Diabetes patient discharge preparation analysis method and system based on data fusion

Abstract

The application discloses a diabetes patient discharge preparation analysis method and system based on data fusion, and relates to the technical field of medical data processing. The method comprises the following steps: obtaining in-hospital blood glucose monitoring data, complication risk assessment data, life habit data, patient willingness data and home blood glucose safety threshold of a patient to be analyzed; constructing a metabolic evolution model according to the in-hospital blood glucose monitoring data and the life habit data; inputting the complication risk assessment data into the metabolic evolution model to calculate an initial risk envelope boundary; identifying a high-risk scene according to a self-management deviation degree and comparing the high-risk scene with the home blood glucose safety threshold to generate a target risk set; extracting a risk subset to be processed based on the target risk set and matching the risk subset to a evidence-based knowledge base, and combining the patient willingness data to generate candidate discharge care suggestion data for medical staff to review. The application improves the consistency and reproducibility of home risk identification, and reduces the false negative rate of metabolic abnormal events that exceed the absolute threshold.

Description

Technical Field

[0001] This invention relates to the field of medical data processing technology, and in particular to a method and system for analyzing the discharge preparation of diabetic patients based on data fusion. Background Technology

[0002] With the global prevalence of diabetes continuing to rise, diabetes has become one of the major chronic diseases seriously threatening human health. Diabetic patients face numerous challenges in returning home and to their communities after hospitalization, including unstable blood glucose management, increased risk of complications, and insufficient self-management skills. Studies show that a significant proportion of diabetic patients lack disease management knowledge and have low levels of self-management behavior, making them prone to problems such as poor blood glucose control and inadequate diet and exercise management after discharge, leading to increased readmission rates and decreased quality of life.

[0003] Currently, there is a certain research foundation for discharge preparation services for diabetic patients. Discharge preparation services aim to improve patients' post-discharge recovery and reintegration into society by assessing their psychological, physiological, and social conditions and developing personalized care plans. Domestic and international studies have shown that structured discharge preparation services can shorten hospital stays, reduce readmission rates, and improve patient satisfaction. However, most existing discharge preparation protocols are based on clinical experience and lack in-depth integration and analysis of individualized patient data, resulting in shortcomings in data-driven risk assessment and intervention plan generation.

[0004] Specifically, existing technologies suffer from the following problems: First, current data processing methods for diabetes discharge risk assessment lack unified alignment and geometric modeling techniques for multi-source data including in-hospital blood glucose time series, complication indicators, and lifestyle habits, resulting in insufficient consistency and reproducibility of risk identification across different samples. Second, existing methods often use single threshold rules to identify metabolic abnormalities, leading to underreporting of events with slow changes that have already exceeded the absolute thresholds for hypoglycemia or hyperglycemia, resulting in insufficient coverage of home-based risk exposure identification. Third, existing methods lack structured scoring rules for matching candidate care measures with risk elements, resulting in insufficient stability and auditability of care matching results. Therefore, there is an urgent need for a discharge preparation analysis method that can improve the accuracy and reproducibility of home-based risk identification based on multi-dimensional data fusion processing and support review by healthcare professionals. Summary of the Invention

[0005] In view of this, the purpose of this invention is to provide a data fusion-based method and system for analyzing discharge preparation for diabetic patients. This system improves the consistency, accuracy, and reproducibility of home-based risk identification by aligning multi-source medical time-series data, normalized vector dimensionality reduction projection, geometric envelope construction, and geometric intersection operations. It also reduces the false negative rate for metabolic abnormalities exceeding absolute thresholds and outputs candidate discharge care recommendations for review by medical staff. The specific solution is as follows: In a first aspect, the present invention discloses a method for analyzing discharge preparation of diabetic patients based on data fusion, comprising: Obtain in-hospital blood glucose monitoring data, complication risk assessment data, lifestyle data, patient willingness data, and home blood glucose safety thresholds for the patients to be analyzed; Based on in-hospital blood glucose monitoring data and lifestyle data, a metabolic evolution model reflecting the blood glucose fluctuation patterns of the patients to be analyzed was constructed. By inputting complication risk assessment data into the metabolic evolution model, the initial risk envelope boundary, which characterizes the individualized risk data boundary of the patient to be analyzed, is calculated. Based on the degree of self-management deviation in lifestyle data, high-risk scenarios where the risk data of the patients to be analyzed exceeds the limit are identified, and the high-risk scenarios are compared with the home blood glucose safety threshold to generate a target risk set that characterizes the degree of home risk exposure of the patients to be analyzed. Based on the target risk set, a subset of risks to be processed is extracted and matched with a pre-built evidence-based knowledge base. Combined with patient willingness data, candidate discharge care recommendations are generated for review by medical staff. The candidate discharge care recommendations are structured and ranked.

[0006] Optionally, before obtaining the in-hospital blood glucose monitoring data, complication risk assessment data, lifestyle data, patient preference data, and home blood glucose safety thresholds for the patient to be analyzed, the following may also be included: Based on the PIPOST model, evidence-based questions for diabetic patients are extracted. Based on these questions, the retrieved medical literature data is summarized according to a pre-built evidence model to construct an evidence-based knowledge base containing various diabetic discharge care measures. Each care measure in the evidence-based knowledge base is associated with a risk label, applicable conditions, evidence level, and recommendation strength.

[0007] Optionally, lifestyle data may include dietary behavior characteristics, exercise behavior characteristics, and the degree of self-management bias. Based on in-hospital blood glucose monitoring data and lifestyle data, a metabolic evolution model reflecting the blood glucose fluctuation patterns of the patients to be analyzed was constructed, including: Extract characteristic data points from in-hospital blood glucose monitoring data under corresponding dietary and exercise behavior characteristics as metabolic abnormality characteristic points. Metabolic abnormality characteristic points are data points whose blood glucose data in the same period meet any of the following abnormal conditions: the absolute value of the slope of blood glucose change exceeds the preset slope threshold, the blood glucose value exceeds the preset hyperglycemia threshold, or the blood glucose value is lower than the preset hypoglycemia threshold. Multiple metabolic abnormality feature points are fitted with curves according to time series to generate a metabolic evolution model.

[0008] Optionally, complication risk assessment data are input into a metabolic evolution model to calculate an initial risk envelope boundary characterizing the individualized risk data boundary of the patient to be analyzed, including: The indicators in the complication risk assessment data were normalized according to their respective clinical reference thresholds to obtain a normalized vector. A correlation matrix is ​​constructed based on the synergistic influence coefficients among the indicators, and the normalized vector is mapped to geometric constraint parameters including the peak blood glucose safety limit, the hypoglycemia risk threshold, and the metabolic fluctuation tolerance window based on the correlation matrix. Geometric constraint parameters are applied to the metabolic evolution model to construct a closed three-dimensional risk envelope in a three-dimensional metabolic state space with time, blood glucose level, and comprehensive complication risk index as coordinate axes, and the closed three-dimensional risk envelope is determined as the initial risk envelope boundary.

[0009] Optionally, high-risk scenarios are compared with home blood glucose safety thresholds to generate a target risk set characterizing the home risk exposure level of the patients to be analyzed, including: Perform a geometric intersection operation between the metabolic data trajectory corresponding to high-risk scenarios and the home safety risk envelope determined by the home blood glucose safety threshold; Identify the points where metabolic data trajectories penetrate the envelope of home safety risks, and perform a local topological search centered on these points to extract the resulting risk-crossing regions and generate a target risk set.

[0010] Optionally, identify high-risk scenarios where the patient's risk data to be analyzed exceeds the limits, including: The degree of self-management deviation is transformed into behavioral deviation simulation parameters and superimposed on the metabolic evolution model. The abnormal deviation vector of the patient data trajectory after superimposing behavioral deviation simulation parameters relative to the initial risk envelope boundary is calculated. Based on the areas where the abnormal deviation vector exceeds the preset safety range, the corresponding time period information and environmental parameters are extracted to identify the high-risk scenarios for the patients to be analyzed.

[0011] Optionally, a subset of risks to be processed is extracted based on the target risk set, and this subset is matched with a pre-built evidence-based knowledge base. Combined with patient preference data, candidate discharge care recommendations are generated for review by healthcare professionals. The candidate discharge care recommendations are structured ranking results, including: The target risk set is grouped and extracted according to the time sequence of risk occurrence to determine the risk subset to be processed; The subset of risks to be processed is matched with candidate care measures in the evidence-based knowledge base. A comprehensive score is calculated based on the risk matching degree of the candidate care measures to the risk subset, the corresponding level of evidence, and the moderating weight determined by the patient's willingness data. The candidate care measures are then merged according to the comprehensive score from high to low, and candidate discharge care recommendations are output for medical staff to review.

[0012] Secondly, this invention discloses a data fusion-based analysis system for the discharge preparation of diabetic patients, comprising: The acquisition module is used to acquire in-hospital blood glucose monitoring data, complication risk assessment data, lifestyle data, patient wishes data, and home blood glucose safety thresholds for the patients to be analyzed. The module is used to construct a metabolic evolution model that reflects the blood glucose fluctuation pattern of the patient to be analyzed, based on in-hospital blood glucose monitoring data and lifestyle data. The calculation module is used to input complication risk assessment data into the metabolic evolution model and calculate the initial risk envelope boundary that characterizes the individualized risk data boundary of the patient to be analyzed. The first generation module is used to identify high-risk scenarios where the risk data of the patient to be analyzed exceeds the limit based on the degree of self-management deviation in the lifestyle data, and compare the high-risk scenarios with the home blood glucose safety threshold to generate a target risk set that represents the degree of home risk exposure of the patient to be analyzed. The second generation module is used to extract a subset of risks to be processed based on the target risk set, match the subset of risks to be processed with a pre-built evidence-based knowledge base, and generate candidate discharge care suggestion data for medical staff to review by combining patient willingness data; the candidate discharge care suggestion data is a structured ranking result.

[0013] Thirdly, the present invention discloses a computer-readable storage medium storing a computer program, which, when executed by a processor, can realize the above-mentioned data fusion-based analysis method for preparing diabetic patients for discharge.

[0014] The beneficial effects of this invention are as follows: First, by aligning in-hospital blood glucose time series with lifestyle data and performing curve fitting on metabolic abnormality feature points, a metabolic evolution model reflecting the blood glucose fluctuation pattern of the patients to be analyzed is constructed, which improves the alignment accuracy of multi-source medical time series data and the stability of metabolic abnormality feature extraction.

[0015] Second, by normalizing the complication risk assessment data and performing dimensionality reduction projection based on the correlation matrix, geometric constraint parameters are obtained and applied to the metabolic evolution model to construct a closed three-dimensional risk envelope, so that the boundary of individualized risk data has a calculable, visualized and reproducible data representation form.

[0016] Third, by parameterizing the degree of self-management bias and superimposing it onto the metabolic evolution model, and by identifying out-of-bounds points and risk-over-bounds areas based on geometric intersection operations and local topological search, combined with anomaly judgment rules that combine slope and absolute threshold, the accuracy of home risk exposure identification is improved and the underreporting rate of metabolic abnormal events that exceed the absolute threshold is reduced.

[0017] Fourth, by integrating candidate care measures through a comprehensive score based on risk matching degree, evidence level, and patient willingness adjustment weights, the consistency and auditability of care matching results are improved, and candidate discharge care recommendations are output for medical staff to review. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0019] Figure 1 This is a flowchart of a data fusion-based analysis method for preparing diabetic patients for discharge, as disclosed in this invention. Figure 2 This is a schematic diagram illustrating the process of constructing a metabolic evolution model disclosed in this invention; Figure 3 This is a schematic diagram of the evolution curve of a metabolic abnormality trend disclosed in this invention; Figure 4 This is a schematic diagram of a three-dimensional envelope surface of the initial risk envelope boundary disclosed in this invention; Figure 5 This is a schematic diagram of the process for generating a target risk set and outputting candidate discharge care recommendations disclosed in this invention; Figure 6 This is a schematic diagram of the structure of a data fusion-based discharge preparation analysis system for diabetic patients disclosed in this invention. Detailed Implementation

[0020] 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.

[0021] Currently, diabetic patients face numerous challenges when returning home and to their communities after hospitalization, including unstable blood glucose management, increased risk of complications, and insufficient self-management abilities. Existing discharge risk assessment data processing solutions lack unified alignment and geometric modeling methods for multi-source data, and often rely on single threshold rules for anomaly detection, resulting in missed reports of metabolic abnormalities exceeding absolute thresholds. This leads to insufficient consistency and reproducibility in home-based risk identification.

[0022] To overcome the aforementioned technical problems, this invention discloses a method and system for analyzing discharge preparation for diabetic patients based on data fusion. This method improves the consistency and reproducibility of home-based risk identification through alignment processing of multi-dimensional medical time-series data, normalized vector dimensionality reduction projection, geometric envelope construction, and geometric intersection operations. Furthermore, by combining an evidence-based knowledge base with a comprehensive score of patient preferences, it outputs candidate discharge care recommendations for review by medical staff.

[0023] Before explaining the specific discharge preparation analysis process, we will first elaborate on the construction process of the evidence-based knowledge base. The evidence-based knowledge base is a crucial foundation for generating the discharge care plan of this invention. It contains a variety of scientifically validated diabetes discharge care measures, providing evidence-based support for subsequent risk-based care matching.

[0024] Before acquiring the various data from the patients to be analyzed, it is necessary to build an evidence-based knowledge base in advance. The specific process is as follows: First, evidence-based questions for diabetic patients are extracted based on the PIPOST model.

[0025] The PIPOST model is a standardized framework for constructing clinical evidence-based questions in the field of evidence-based nursing. In this invention, it is used as a methodological tool to extract evidence-based questions related to diabetes discharge preparation. The meanings of each element in the PIPOST model and their specific content in this invention are as follows: P represents the target population, which in this invention refers to diabetic patients; I represents care measures, i.e., various interventions in discharge preparation services, such as dietary management guidance, exercise management recommendations, and blood glucose monitoring programs; P represents the professionals implementing the interventions, i.e., the medical and nursing team involved in discharge preparation services; O represents outcome indicators, including measurable health outcomes such as improved self-management ability, improved quality of life, improved blood glucose control, and reduced readmission rate; S represents the evidence application location, i.e., the hospital and home environments; T represents the type of evidence, including clinical decision-making, best practices, guidelines, evidence summaries, expert consensus, and systematic reviews. Through the systematic decomposition of the PIPOST model, the direction and scope of evidence retrieval can be clearly defined, ensuring that the extracted evidence-based questions are targeted and complete.

[0026] Then, based on the evidence-based questions, the retrieved medical literature data is summarized according to the pre-constructed evidence model to build an evidence-based knowledge base containing various diabetes discharge care measures. In this embodiment, the evidence model adopts the "6S Pyramid" evidence model. The "6S Pyramid" evidence model is a hierarchical evidence retrieval framework widely used in the field of evidence-based medicine. This model divides medical evidence into six levels from low to high according to the degree of processing and integration. The specific contents of each level are as follows: The first level consists of single studies, which are original clinical research papers and represent primary evidence that has not undergone secondary processing. The second layer is the summary, which is a structured summary of information formed after a critical evaluation of a single study; The third level is comprehensive, which includes literature that conducts comprehensive statistical analysis of multiple studies, such as systematic reviews and meta-analysis. The fourth layer is the evidence summary, which is a summary of clinically usable evidence formed by further summarizing and integrating the systematic reviews. The fifth layer is the clinical decision support system, which is an information-based decision support tool that combines the best evidence with clinical expertise and the specific circumstances of the patient. The sixth level consists of guidelines, which are clinical practice guidelines developed by authoritative institutions based on systematic reviews and expert consensus. These represent the most processed form of evidence.

[0027] During the search process, the evidence is retrieved from the top to the bottom level according to the "6S Pyramid" evidence model. High-level evidence is prioritized and supplemented by searching the next lower level when high-level evidence is insufficient. This ensures that all care measures included in the knowledge base have reliable evidence-based basis.

[0028] In the specific literature search process, searchable databases, guideline websites, and professional society websites include UpToDate, BMJ Best Practice, the Joanna Briggs Institute Evidence-Based Health Care Centre Database in Australia, the International Guidelines Collaboration Network, the American Clinical Guidelines Network, the New Zealand Guidelines Collaboration Group website, the Scottish Inter-Hospital Guidelines Network, Cochrane Library, PubMed, CINAHL, CNKI, Wanfang Database, and the China Biomedical Literature Database, etc.

[0029] Following the search, the included guidelines were evaluated for quality using the Clinical Guideline Research and Evaluation System, and the included expert consensus and systematic reviews were evaluated using the JBI Evidence-Based Healthcare Center Literature Quality Assessment Tool. Finally, by extracting, summarizing, and generalizing all evidence meeting the quality criteria, an evidence-based knowledge base containing various diabetes discharge care measures was constructed. This knowledge base covers multiple dimensions of care measures, including dietary management interventions, exercise management interventions, blood glucose monitoring guidance, medication record management tips, psychological support interventions, complication prevention interventions, risk factor management interventions, and community referral services. Each care measure is accompanied by corresponding evidence level and recommendation strength information. The knowledge base is stored in a structured format, with each care measure associated with fields such as intervention number, risk label, applicable conditions, contraindications, evidence level, recommendation strength, frequency of implementation, and whether healthcare review is required, to facilitate procedural matching and retrieval.

[0030] Through the systematic retrieval and summarization process based on the PIPOST model and the "6S pyramid" evidence model, the constructed evidence-based knowledge base has the advantages of transparent evidence sources, distinct quality levels, and traceability and updability. This provides a foundation for accurately matching individualized patient risks with scientifically validated care measures. After the preliminary construction of the evidence-based knowledge base is completed, the discharge preparation analysis process for specific patients can begin.

[0031] See Figure 1 As shown in the figure, this invention discloses a method for analyzing discharge preparation of diabetic patients based on data fusion, including the following steps: Step S1: Obtain the in-hospital blood glucose monitoring data, complication risk assessment data, lifestyle data, patient willingness data, and home blood glucose safety threshold for the patient to be analyzed.

[0032] In this embodiment, the first step is to acquire multi-dimensional data of the patient to be analyzed in order to establish a comprehensive patient data profile. This multi-dimensional data forms the basis for subsequent construction of metabolic evolution models, risk assessment, and generation of discharge care plans. The specific meanings of each data point are as follows: In-hospital blood glucose monitoring data refers to time-series blood glucose data collected by blood glucose monitoring devices during the hospitalization of the patient to be analyzed. This includes fasting blood glucose levels, postprandial blood glucose levels, random blood glucose levels, and blood glucose measurements at various time points, such as multiple daily finger-prick blood glucose monitoring data or dynamic blood glucose data recorded by a continuous glucose monitor. This in-hospital blood glucose monitoring data can reflect the blood glucose fluctuation patterns and metabolic characteristics of the patient under medical supervision.

[0033] Complication risk assessment data refers to various indicators related to diabetic complications obtained through clinical testing and assessment tools, including but not limited to glycated hemoglobin levels, lipid indicators such as total cholesterol and triglyceride levels, renal function indicators such as urinary microalbumin and creatinine levels, fundus examination results, foot neuropathy examination results, and cardiovascular risk assessment results. This complication risk assessment data can reflect the current complication risk status of the patient being analyzed from multiple organ system dimensions.

[0034] Lifestyle data includes three subcategories: dietary behavior characteristics, exercise behavior characteristics, and the degree of self-management deviation. Dietary behavior characteristics refer to information such as the patient's eating patterns, dietary structure, meal times, and portion sizes. For example, does the patient adhere to diabetic dietary requirements and exhibit behaviors such as overeating or irregular eating habits? Exercise behavior characteristics refer to information such as the patient's exercise frequency, duration, intensity, and type. For example, does the patient maintain at least 30 minutes of moderate-intensity exercise daily? The degree of self-management deviation refers to the extent to which the patient deviates from medical advice in areas such as blood glucose self-monitoring frequency, medication adherence, and lifestyle management. This degree of self-management deviation can be assessed using quantitative scoring tools, such as the Diabetes Self-Management Behavior Scale; a lower score indicates a greater degree of self-management deviation.

[0035] Patient willingness data refers to information on the willingness and preferences of patients to accept various post-discharge care measures. This includes the degree of acceptance of different types of care measures, the patients' expectations regarding follow-up methods and frequency, the patients' family support, and the patients' initiative and enthusiasm for their own health management. Patient willingness data can be obtained through structured questionnaires or interviews.

[0036] Home blood glucose safety thresholds refer to the safe range of blood glucose fluctuations set for patients under analysis in a home environment, based on their age, disease duration, complications, and clinical guidelines. This includes the upper and lower limits of fasting blood glucose, the upper limit of postprandial blood glucose, and hypoglycemia warning values. For example, for general type 2 diabetes patients, the home fasting blood glucose safety threshold can be set to 4.4 to 7.0 mmol / L, and the 2-hour postprandial blood glucose safety threshold can be set to no more than 10.0 mmol / L. Specific values ​​can be personalized by clinicians based on the individual circumstances of each patient. In this embodiment, the above clinical thresholds are taken from the Adult Diabetes Management Guidelines published by the Chinese Diabetes Society and can be parameterized by doctors through the system interface for individual patients.

[0037] Step S2: Based on in-hospital blood glucose monitoring data and lifestyle data, construct a metabolic evolution model that reflects the blood glucose fluctuation pattern of the patient to be analyzed.

[0038] In this embodiment, after acquiring multi-dimensional data of the patient to be analyzed, a metabolic evolution model is constructed based on in-hospital blood glucose monitoring data and lifestyle data. A metabolic evolution model is a mathematical model reflecting the individualized blood glucose fluctuation patterns of the patient to be analyzed. This model can characterize the dynamic characteristics of blood glucose changes over time under different dietary and exercise behaviors. Unlike existing technologies that use static statistical indicators to describe blood glucose status, the metabolic evolution model captures the direction, rate, and periodicity of blood glucose fluctuations in the form of time series data, thus providing a dynamic data foundation for subsequent risk simulation and prediction. The purpose of constructing the metabolic evolution model is to digitize and model the blood glucose fluctuation patterns observed in the hospital, enabling the metabolic evolution model to be used in subsequent steps to simulate and predict the metabolic responses of the patient to be analyzed in different scenarios.

[0039] See Figure 2 As shown, the specific process of constructing a metabolic evolution model is as follows: Step S21: Extract the feature data points of in-hospital blood glucose monitoring data under the corresponding dietary behavior characteristics and exercise behavior characteristics as metabolic abnormality feature points.

[0040] Among them, metabolic abnormality feature points are data points whose blood glucose data in the same period meet any of the following abnormal conditions: the absolute value of the slope of blood glucose change exceeds the preset slope threshold, the blood glucose value exceeds the preset high blood glucose threshold, or the blood glucose value is lower than the preset low blood glucose threshold.

[0041] Specifically, in-hospital blood glucose monitoring data is time-aligned with lifestyle data from the same period to analyze blood glucose changes in patients under specific dietary and exercise characteristics. Time alignment refers to matching the timestamps of in-hospital blood glucose monitoring data with the timestamps of dietary and exercise behavior records, ensuring that each blood glucose data point is associated with dietary and exercise behavior information within the corresponding time period.

[0042] This embodiment employs three types of anomaly detection rules executed in parallel: the first type is the slope rule, which calculates the slope of change between adjacent sampling points for blood glucose data within each time window. When the absolute value of the slope exceeds a preset slope threshold, it is recorded as a metabolic abnormality feature point. In this embodiment, the preset slope threshold is 0.1 mmol / L / min. The second type is the hyperglycemia absolute threshold rule, which records a metabolic abnormality feature point when the blood glucose value at a sampling point exceeds a preset hyperglycemia threshold. In this embodiment, the preset hyperglycemia threshold is 10.0 mmol / L, corresponding to the safe upper limit of blood glucose 2 hours after a meal for general type 2 diabetic patients. The third type is the hypoglycemia absolute threshold rule, which records a metabolic abnormality feature point when the blood glucose value at a sampling point is lower than a preset hypoglycemia threshold. In this embodiment, the preset hypoglycemia threshold is 3.9 mmol / L, corresponding to the clinical hypoglycemia warning value. When any of the above rules is triggered, the corresponding data point is included as a metabolic abnormality feature point in subsequent processing, thereby capturing both rapid fluctuation events and avoiding underreporting of metabolic abnormality events that change slowly but have exceeded the absolute threshold.

[0043] Extracting metabolic abnormality feature points can focus on the critical moments when the metabolic regulation ability of the patient being analyzed is weak, and filter out stable blood glucose fluctuations within the normal range, thereby making the metabolic evolution model constructed subsequently more targeted.

[0044] Step S22: Perform curve fitting on multiple metabolic abnormality feature points according to the time series to generate a metabolic evolution model.

[0045] Specifically, the extracted metabolic abnormality feature points are arranged chronologically to form a time series reflecting the pattern of metabolic abnormalities. Then, curve fitting is performed on this time series to generate an evolution curve that continuously describes the trend of metabolic abnormalities in the patient being analyzed. Curve fitting refers to constructing a smooth and continuous mathematical curve based on discrete data points, ensuring that the curve passes through or approximates all metabolic abnormality feature points as closely as possible. Curve fitting can employ methods such as polynomial fitting, spline interpolation, or Gaussian process regression; in this embodiment, cubic spline interpolation is preferred. Cubic spline interpolation uses a cubic polynomial function to connect every two adjacent metabolic abnormality feature points, ensuring the continuity of the first and second derivatives of the curve at each feature point. This results in a curve that faithfully passes through each metabolic abnormality feature point and provides smooth and reasonable interpolation estimates between adjacent feature points.

[0046] The mathematical expression for cubic spline interpolation is as follows: Suppose there are n metabolic abnormality feature points. ,..., ,in For time, This represents the glucose metabolism response value. It is located within each adjacent interval of metabolic abnormality characteristic points. Above, fit a cubic polynomial: ; in , , , Let be the polynomial coefficients over this interval. The constraints include: First, That is, the curve passes through each metabolic abnormality feature point; second, That is, the function values ​​are continuous at the connection point between adjacent intervals; third, That is, the first derivative is continuous; fourth, This means that the second derivative is continuous. By solving the linear equations formed by the above constraints, the polynomial coefficients on all intervals are determined, thereby generating a complete metabolic evolution model.

[0047] The generated metabolic evolution model exists in the form of a mathematical function, with independent variables being time and related lifestyle data parameters (i.e., dietary behavior characteristics and exercise behavior characteristics), and dependent variable being the glucose metabolism response value.

[0048] To more clearly illustrate the implementation process of steps S21 and S22, the following will be combined with... Figure 3Here's a concrete example. Suppose we have the following continuous blood glucose monitoring data for a patient, Zhang, during his 7-day hospitalization: On day 1, the blood glucose level was 8.2 mmol / L one hour after breakfast and rose to 12.5 mmol / L 1.5 hours later, with a change rate of 0.14 mmol / L per minute, exceeding the preset threshold of 0.1 mmol / L. This time point was marked as metabolic abnormality feature point A1. On the second day, after lunch, the patient to be analyzed walked for 30 minutes. Within 30 minutes after the end of the exercise, the blood glucose dropped from 11.8 mmol / L to 6.3 mmol / L, with an absolute slope of 0.18 mmol / L per minute, which exceeded the preset threshold and was marked as metabolic abnormality feature point A2. On the third day, after dinner, the patient did not exercise and ate a large amount of food. Blood glucose rose from 9.0 mmol / L to 14.2 mmol / L within 40 minutes, with a slope of 0.13 mmol / L per minute. This was marked as metabolic abnormality characteristic point A3. Between 2:00 AM and 4:00 AM on the 5th day, the patient's blood glucose level dropped from 7.5 mmol / L to 3.8 mmol / L, with an absolute slope of 0.03 mmol / L per minute. This did not exceed the preset slope threshold, but the endpoint blood glucose level of 3.8 mmol / L was lower than the preset hypoglycemia threshold of 3.9 mmol / L, triggering the hypoglycemia absolute threshold rule and marking it as metabolic abnormality feature point A4. On day 6, before breakfast, in a fasting state, the patient's blood glucose level rose from 5.2 mmol / L to 9.8 mmol / L within 30 minutes, with a slope of 0.15 mmol / L per minute, which was marked as metabolic abnormality characteristic point A5.

[0049] like Figure 3 As shown, the metabolic abnormality feature points A1, A2, A3, A4, and A5 are arranged in chronological order. With time on the horizontal axis and glucose metabolism response value on the vertical axis, cubic spline interpolation is used to fit these feature points to generate a continuous and smooth metabolic abnormality trend evolution curve. The peaks of this curve correspond to high-risk periods of weak metabolic regulation in the analyzed patients, while the troughs correspond to low-risk periods of relatively stable metabolic regulation. From... Figure 3 It can be observed that in patient Zhang, especially during the postprandial period when diet is uncontrolled and exercise is lacking, metabolic abnormality feature points are densely distributed and the evolution curve shows a significant upward trend. Simultaneously, the hypoglycemic abnormality feature point A4 in the early morning is also significantly captured, indicating that this type of scenario represents a high-risk region for metabolic regulation failure in the patient Zhang. This evolution curve constitutes the metabolic evolution model of patient Zhang, and subsequent steps will be based on this model to calculate the initial risk envelope boundary and identify high-risk scenarios.

[0050] The metabolic evolution model constructed through the above steps can dynamically depict the metabolic fluctuation patterns of the patients to be analyzed in the form of time series, and capture transient abnormal patterns that are difficult to detect by traditional static indicators, thereby providing a more refined data foundation for subsequent risk prediction.

[0051] After determining the metabolic evolution model, it is necessary to further introduce complication risk assessment data to determine the individualized risk data boundary of the patient to be analyzed, that is, the initial risk envelope boundary in the next step.

[0052] Step S3: Input the complication risk assessment data into the metabolic evolution model to calculate the initial risk envelope boundary that characterizes the individualized risk data boundary of the patient to be analyzed.

[0053] In this embodiment, the metabolic evolution model describes the blood glucose fluctuation patterns of the patient under analysis, but it has not yet determined at what level of fluctuation the patient's physiological regulation will fail. Complication risk assessment data reflects the current state and reserve capacity of the patient's various organ systems. The initial risk envelope boundary refers to the maximum range within which the patient can maintain metabolic homeostasis under current physical conditions through its own physiological regulatory mechanisms, i.e., the risk boundary model; compensation refers to the process by which the body activates internal regulatory mechanisms to compensate for a certain functional impairment. Exceeding this initial risk envelope boundary means that the patient's metabolic regulation may be decompensated, facing the risk of acute complication or disease exacerbation. By incorporating complication risk assessment data into the metabolic evolution model, this initial risk envelope boundary can be calculated.

[0054] The specific process for calculating the initial risk envelope boundary is as follows: In this embodiment, each indicator in the complication risk assessment data is normalized according to its respective clinical reference threshold to obtain a normalized vector. A correlation matrix is ​​constructed based on the synergistic influence coefficients between the indicators. The normalized vector is then mapped to geometric constraint parameters, including the peak blood glucose safety limit, the hypoglycemia risk threshold, and the metabolic fluctuation tolerance window, according to the correlation matrix. These geometric constraint parameters are applied to a metabolic evolution model to construct a closed three-dimensional risk envelope in a three-dimensional metabolic state space with time, blood glucose level, and the comprehensive complication risk index as coordinate axes. This closed three-dimensional risk envelope is then defined as the initial risk envelope boundary. In this embodiment, each indicator in the complication risk assessment data is used as a component of a high-dimensional vector. For example, glycated hemoglobin level, blood lipid level, and renal function indicators are each used as a dimension of the high-dimensional vector. By normalizing each indicator to clinical thresholds and performing weighted linear projection based on the correlation matrix, the high-dimensional vector is mapped to a three-dimensional metabolic state space with time, blood glucose level, and the comprehensive complication risk index as coordinate axes. The elements in the correlation matrix represent the synergistic influence coefficients between indicator pairs, determined jointly by clinical medical knowledge, epidemiological statistics, and sample data regression analysis. This allows physicians to make parameterized adjustments for individual patients through the system interface. The mapped geometric constraint parameters include three terms: the peak blood glucose safety limit, the hypoglycemia risk threshold, and the metabolic fluctuation tolerance window. These constrain the upper and lower boundaries of the three-dimensional metabolic state space at blood glucose levels and the fluctuation window over time, respectively. Then, the extracted geometric constraint parameters are applied as constraints to the metabolic evolution model, performing surface reconstruction on the metabolic state space described by the model. Surface reconstruction refers to constructing a continuous, closed surface in three-dimensional space based on the geometric constraint parameters. The three dimensions of this closed three-dimensional risk envelope correspond to blood glucose level, time, and the comprehensive complication risk index, respectively. The comprehensive complication risk index is calculated by weighted fusion of various complication risk assessment data. The internal region of the closed three-dimensional risk envelope represents the range of metabolic states of the patient under analysis, supported by historical in-hospital data and safety threshold rules. The surface of the closed three-dimensional risk envelope represents the limiting boundary of the patient's physiological regulation. The generated closed three-dimensional risk envelope surface covers all data activity boundaries of the patient to be analyzed under the current physiological conditions, and this closed three-dimensional risk envelope surface is determined as the initial risk envelope boundary.

[0055] To more intuitively illustrate the construction process of the initial risk envelope boundary, let's continue with the example of the patient Zhang to be analyzed. Assume Zhang's complication risk assessment data are as follows: glycated hemoglobin 8.5%, total cholesterol 6.2 mmol / L, triglycerides 2.8 mmol / L, urinary microalbumin 45 mg / L, creatinine 95 μmol / L, and fundus examination showing mild non-proliferative retinopathy. The following details the specific process by which the spatial manifold unfolding algorithm transforms the above six complication risk assessment indicators into geometric constraint parameters in a three-dimensional metabolic state space.

[0056] The first step is normalization. The six complication risk assessment indicators are linearly normalized according to their respective upper limits of clinical normality and risk thresholds, mapping them to a range of 0 to 1. The normalization formula is: ; When the actual value is lower than the normal reference upper limit, the normalized value is 0, indicating that the indicator is within the normal range.

[0057] The severity of fundus lesions was assessed using a grading system, with specific grading rules shown in Table 1 below. Table 1: Rules for Assigning the Severity of Fundus Lesions

[0058] The normalized calculation results of each indicator are shown in Table 2 below: Table 2: Normalized Calculation Results of Various Indicators

[0059] After the above normalization process, the six complication risk assessment indicators of the patient Zhang to be analyzed form a six-dimensional normalized vector: [0.57, 0.38, 0.28, 0.14, 0, 0.3].

[0060] The second step involves calculating the correlation matrix and identifying synergistic constraints. A quantifiable analysis of the clinical correlations between the indicators in the six-dimensional normalized vector is performed, constructing a 6×6 correlation matrix. Each element in the correlation matrix represents the synergistic influence coefficient between two corresponding indicators. This synergistic influence coefficient is pre-determined based on clinical medical knowledge, epidemiological statistics, and regression analysis of sample data. The specific synergistic influence coefficients and their clinical basis are shown in Table 3 below. Table 3: Synergistic Influence Coefficient and Its Clinical Basis

[0061] The third step involves performing dimensionality reduction projection based on the correlation matrix to generate geometric constraint parameters. In this embodiment, the projection process projects a six-dimensional normalized vector onto a three-dimensional metabolic state space with blood glucose level, time, and comprehensive complication risk index as coordinate axes, based on the synergistic influence coefficients in the correlation matrix, generating three geometric constraint parameters. The calculation process for each geometric constraint parameter is as follows: Determination of the safe limit for peak blood glucose levels. The safe limit for peak blood glucose levels is determined jointly by the normalized values ​​and synergistic influence coefficients of four indicators: glycated hemoglobin, total cholesterol, triglycerides, and the degree of fundus lesions. The specific calculation method is as follows: First, the normalized value of total cholesterol (0.38) and the normalized value of triglycerides (0.28) are weighted and averaged to obtain the comprehensive lipid normalized value. Since total cholesterol has a stronger indicative effect on cardiovascular risk, its weights are set to 0.6 and 0.4, respectively, resulting in: ; Then, the normalized values ​​of glycated hemoglobin (0.57), comprehensive lipids (0.34), and fundus lesion severity (0.3) were weighted and fused according to the synergistic influence coefficients in the correlation matrix to calculate the joint risk factor: ; Where 0.6 is the synergistic effect coefficient between glycated hemoglobin and overall blood lipids, and 0.7 is the synergistic effect coefficient between glycated hemoglobin and the degree of fundus lesions. Finally, the combined risk factors are substituted into the blood glucose peak mapping function: Peak blood glucose safety limit = theoretical maximum tolerable blood glucose - combined risk factors × (theoretical maximum tolerable blood glucose - normal postprandial peak blood glucose upper limit) / 2.0; The theoretical maximum tolerable blood glucose level is set at 20.0 mmol / L; exceeding this value carries a risk of ketoacidosis. The upper limit for normal postprandial peak blood glucose is set at 10.0 mmol / L. Substituting these values ​​into the calculation, we get: The safe limit for peak blood glucose level = 20.0 - 0.984 × (20.0 - 10.0) / 2.0 = 20.0 - 4.92 = 15.08 mmol / L; Rounded to 15.0 mmol / L, this result indicates that due to the combined effects of moderately elevated glycated hemoglobin, mild dyslipidemia, and mild fundus lesions in the patient Zhang, the safe limit for peak blood glucose has been tightened from the theoretical maximum of 20.0 mmol / L to 15.0 mmol / L.

[0062] Determination of the hypoglycemic risk threshold. The hypoglycemic risk threshold is determined by the normalized values ​​of three indicators: urinary microalbumin, creatinine, and the degree of fundus lesions, along with their synergistic influence coefficients. Impaired renal function and microvascular disease reduce the body's ability to perceive and respond to hypoglycemia; therefore, the worse the relevant indicators, the higher the hypoglycemic risk threshold needs to be to allow for a greater safety margin. The specific calculation method is as follows: First, the normalized value of urinary microalbumin (0.14) and the normalized value of creatinine (0) were fused using a synergistic effect coefficient of 0.8. ; Then, the combined renal function indicators and the normalized value of fundus lesion severity (0.3) were fused with a synergistic influence coefficient of 0.5. ; Finally, substituting into the hypoglycemia threshold mapping function: ; The baseline hypoglycemia threshold is 3.0 mmol / L. The clinical definition of severe hypoglycemia is adjusted upwards by 3.0 mmol / L, meaning it can be increased from 3.0 to a maximum of 6.0 mmol / L. Substituting these values ​​into the calculation, we get: ; Rounded to 3.9 mmol / L, this result indicates that due to the combined effects of mildly elevated urinary microalbumin and mild fundus lesions in the patient Zhang, the hypoglycemic risk threshold was raised from the baseline value of 3.0 mmol / L to 3.9 mmol / L. That is, when blood glucose drops below 3.9 mmol / L, it is considered a dangerous state.

[0063] Determination of the metabolic fluctuation tolerance window. The metabolic fluctuation tolerance window refers to the maximum allowable blood glucose variation within a given time window. In this embodiment, it is set to 2 hours. This parameter is determined by the combined normalized values ​​of two renal function indicators: urinary microalbumin and creatinine. The kidneys are vital organs for maintaining homeostasis of glucose metabolism in the body. Impaired renal function weakens the body's ability to buffer against rapid fluctuations in blood glucose. Therefore, the worse the renal function-related indicators, the tighter the allowable blood glucose variation needs to be. The specific calculation method is as follows: Maximum permissible variation range = baseline permissible variation range - combined renal function indicators × tightening range / 0.5; The baseline allowable range is set at 10.0 mmol / L, which is the typical limit for blood glucose changes in healthy individuals over 2 hours. The tightening range is set at the difference between the baseline and the minimum allowable range, i.e., 10.0 - 4.0 = 6.0 mmol / L. The minimum allowable range of 4.0 mmol / L represents the limit for patients with severe renal insufficiency. 0.5 is a normalization correction factor used to calibrate the mapping relationship between the range of values ​​for the combined renal function indicators and the tightening range. Substituting into the calculation, we get: Maximum permissible variation = 10.0 - 0.14 × 6.0 / 0.5 = 10.0 - 1.68 = 8.32 mmol / L; Rounded to 8.0 mmol / L, this result indicates that the blood glucose level of the patient Zhang under analysis should not exceed 8.0 mmol / L within 2 hours.

[0064] like Figure 4 As shown, the three geometric constraint parameters mentioned above were applied to the metabolic evolution model of the patient Zhang to be analyzed, and surface reconstruction was performed. In a three-dimensional space with "time" as the x-axis, "blood glucose level" as the y-axis, and "comprehensive complication risk index" as the z-axis, a closed three-dimensional risk envelope was generated. The upper boundary of the closed three-dimensional risk envelope is located at a blood glucose level of 15.0 mmol / L, determined by the blood glucose peak safety limit. This upper boundary is not a strictly horizontal plane, but is slightly fluctuating due to the adjustment of the comprehensive complication risk index dimension; that is, when the comprehensive complication risk index is high, the upper boundary will be further compressed. The lower boundary of the closed three-dimensional risk envelope is located at a blood glucose level of 3.9 mmol / L, determined by the hypoglycemia risk threshold, and is also adjusted by the comprehensive complication risk index. The vertical distance between the upper and lower boundaries is the safe activity space of the patient Zhang in the blood glucose level dimension, which is approximately 11.1 mmol / L (15.0-3.9). The front and back sides of the closed three-dimensional risk envelope are determined by the metabolic fluctuation tolerance window in the time dimension, which limits the maximum allowable change in blood glucose level within any 2-hour window to 8.0 mmol / L. This closed three-dimensional risk envelope is the initial risk envelope boundary for the patient Zhang to be analyzed. All metabolic states within the initial risk envelope boundary are safe and maintainable by the patient Zhang's body according to the risk threshold rules, while metabolic states exceeding the initial risk envelope boundary indicate a risk breach.

[0065] The initial risk envelope boundary constructed using the above method integrates multiple complication risk assessment indicators into the metabolic state space in a geometric manner, thereby achieving a quantitative and visual representation of the individualized risk data boundary of the patients to be analyzed.

[0066] After determining the initial risk envelope boundary, it is necessary to consider potential deviations in the self-management behavior of patients being analyzed upon discharge and transitioning to their home environment. These deviations can affect metabolic status and may cause metabolic trajectories to exceed the initial risk envelope boundary. Therefore, the next step will simulate the impact of the degree of self-management deviation on the metabolic evolution model to identify high-risk scenarios that could lead to risk data exceeding the boundary.

[0067] Step S4: Based on the degree of self-management deviation in the lifestyle data, identify high-risk scenarios where the risk data of the patient to be analyzed exceeds the limit, and compare the high-risk scenarios with the home blood glucose safety threshold to generate a target risk set that represents the degree of home risk exposure of the patient to be analyzed.

[0068] In this embodiment, after determining the initial risk envelope boundary, it is necessary to assess the actual risks that the patients under analysis may face in the home environment after discharge. Because the patients under analysis lack the strict supervision of medical institutions in the home environment, their self-management behavior often exhibits a certain degree of deviation. This deviation can cause the actual metabolic state of the patients under analysis to deviate from the ideal state within the hospital. Therefore, it is necessary to incorporate the degree of self-management deviation into the metabolic evolution model for simulation analysis to identify high-risk scenarios that may lead to the risk data of the patients under analysis exceeding the boundary.

[0069] See Figure 5 As shown, Figure 5 The process of generating candidate discharge care recommendation data is shown in steps S4-S5.

[0070] The specific process of identifying high-risk scenarios and generating a target risk set is as follows: Step S41: Convert the degree of self-management deviation into behavioral deviation simulation parameters and superimpose them into the metabolic evolution model. Calculate the abnormal deviation vector of the patient data trajectory to be analyzed relative to the initial risk envelope boundary after superimposing the behavioral deviation simulation parameters.

[0071] Specifically, the degree of self-management deviation is a quantified deviation score that reflects the extent to which the patient being analyzed deviates from standard behaviors in areas such as diet management, exercise management, blood glucose monitoring, and medication use. The process of converting this degree of self-management deviation into behavioral deviation simulation parameters involves generating corresponding perturbation parameters based on the magnitude and direction of the deviation score. These behavioral deviation simulation parameters are a set of numerical parameters obtained by quantifying the degree of self-management deviation in the patient being analyzed. They are used in a metabolic evolution model to simulate the impact of deviant behaviors on blood glucose fluctuations. For example, if the patient being analyzed has a high diet management deviation score, indicating a large deviation, the generated behavioral deviation simulation parameters will add a positive perturbation related to the dietary behavior characteristics to the metabolic evolution model, simulating the additional blood glucose increases that the patient might experience due to irregular eating habits or an unbalanced diet in a home environment. After superimposing the behavioral deviation simulation parameters onto the metabolic evolution model, the model will output a new metabolic data trajectory affected by the deviation. This trajectory reflects the blood glucose fluctuations that the patient might experience in a home environment with a degree of self-management deviation.

[0072] Furthermore, the behavioral deviation simulation parameters are stored in a parameter table, which includes at least the following fields: dietary deviation coefficient, exercise deviation coefficient, blood glucose monitoring missing coefficient, medication record management deviation coefficient, perturbation direction, perturbation amplitude, scale source, calibration time, and medical staff review status. Each deviation coefficient is normalized by default by subtracting the ratio of the corresponding scale score to the scale's full score from 1, and is limited to the range of 0 to 1.

[0073] Then, the abnormal deviation vector of the metabolic data trajectory affected by the bias relative to the initial risk envelope boundary determined in step S3 is calculated. The abnormal deviation vector is a spatial vector pointing from the surface of the initial risk envelope boundary to the portion of the metabolic data trajectory that exceeds the initial risk envelope boundary. The direction of the abnormal deviation vector indicates the direction in which the metabolic data trajectory deviates from the initial risk envelope boundary, and the magnitude of the abnormal deviation vector indicates the degree of deviation. When the magnitude of the abnormal deviation vector exceeds zero, it indicates that the metabolic state of the patient being analyzed has exceeded the limits of physiological regulation at that moment.

[0074] Step S42: Based on the area where the abnormal deviation vector exceeds the preset safety range, extract the corresponding time period information and environmental parameters to determine the high-risk scenario for the patient to be analyzed.

[0075] Specifically, the preset safety range is an allowable threshold set for the magnitude of the abnormal deviation vector, used to distinguish between negligible minor deviations and significant deviations requiring attention. When the magnitude of the abnormal deviation vector exceeds this preset safety range, it indicates that the corresponding metabolic state has significantly exceeded the preset risk data boundary of the patient being analyzed. The time period information and corresponding environmental parameters corresponding to the areas exceeding the preset safety range are extracted, and the combination of the above time period information and environmental parameters is identified as a high-risk scenario. A high-risk scenario refers to a typical situation that may lead to metabolic decompensation in the patient being analyzed, composed of specific time periods, behavioral patterns, and environmental conditions.

[0076] Furthermore, the preset safety range is set according to the normalized value of the abnormal deviation vector magnitude, and the default threshold can be 0.05. When the normalized abnormal deviation vector magnitude is greater than 0.05 and the duration is not less than two consecutive sampling intervals, the corresponding area is marked as a candidate high-risk scenario, and the time period, behavior pattern and environmental parameters are output for medical staff to review.

[0077] By incorporating the degree of self-management bias into a metabolic evolution model for simulation, this method can predict individualized high-risk scenarios that patients may face in their home environment even before they are discharged from the hospital, thus achieving a forward-looking and individualized risk assessment.

[0078] After identifying high-risk scenarios, it is necessary to further assess the actual exposure levels of these scenarios in a home environment. Therefore, high-risk scenarios are compared with home blood glucose safety thresholds to generate a target risk set. The specific process for generating the target risk set is as follows: Step S43: Perform a geometric intersection operation between the metabolic data trajectory corresponding to the high-risk scenario and the home safety risk envelope determined by the home blood glucose safety threshold.

[0079] Specifically, the home blood glucose safety threshold determines the safe range of blood glucose fluctuations for the patient being analyzed in a home environment. This safe range can be represented as a home safety risk envelope in the metabolic state space. The home safety risk envelope is a closed surface formed by the parameters of the home blood glucose safety threshold in a three-dimensional metabolic state space. The internal region of the home safety risk envelope represents the acceptable range of blood glucose fluctuations in a home environment. Geometric intersection operation is a process in computational geometry used to solve the intersection relationship between trajectory line segments and surfaces. This embodiment uses the line segment-surface intersection method to determine the intersection relationship between the metabolic data trajectory and the home safety risk envelope segment by segment. Through the intersection operation, it is possible to accurately determine which parts of the metabolic data trajectory fall within the home safety risk envelope and which parts penetrate the home safety risk envelope and are exposed to the danger zone.

[0080] Step S44: Identify the points where metabolic data trajectories penetrate the home safety risk envelope and perform a local topological search centered on the points where the data crosses the boundary to extract the risk boundary regions and generate a target risk set.

[0081] Specifically, boundary crossing points refer to the boundary intersections where the metabolic data trajectory crosses from the inside to the outside of the home safety risk envelope, i.e., the intersection of the metabolic data trajectory and the home safety risk envelope. After identifying boundary crossing points, a local topological search is performed centered on each boundary crossing point. Topological search refers to a structured search within the spatial neighborhood of the metabolic data trajectory based on connectivity and geometric deformation characteristics.

[0082] The specific process of local topology search is as follows: First, extract the edge tangent vector of the boundary crossing point. The edge tangent vector refers to the tangent direction vector of the metabolic data trajectory along the surface of the home safety risk envelope at the boundary crossing point. Searching along this direction can discover other potential risk nodes associated with the boundary crossing point.

[0083] Then, curvature detection is performed on the surrounding nodes of the metabolic data trajectory along the edge tangent vector. Curvature is a measure of the degree of bending of a curve or surface at a point; a larger curvature indicates more severe bending. When the curvature distortion parameter at a surrounding node reaches a preset deformation condition, it indicates that the metabolic state at that node has also deviated significantly from the preset condition. The curvature distortion parameter is the deviation between the actual curvature and the normal curvature at that node. Connecting the abnormal nodes whose curvature distortion parameters reach the preset deformation condition forms a closed contour line. The area enclosed by the closed contour line is the risk boundary region. The risk boundary region indicates that the risk envelope coverage of the patient to be analyzed has a gap in this area, and the metabolic state cannot be effectively covered by physiological regulatory mechanisms in this area. By collecting all the extracted risk boundary regions, a target risk set representing the degree of home risk exposure of the patient to be analyzed is generated.

[0084] Furthermore, the local topology search parameters include at least the neighborhood search radius, maximum search steps, curvature distortion threshold, minimum number of nodes for the closed contour, lower limit of the region area, and medical review status field; in this embodiment, the neighborhood search radius is taken by default as three adjacent sampling points, the curvature distortion threshold is taken by default as 1.5 times the average normal curvature, and the minimum number of nodes for the closed contour is taken by default as 4.

[0085] Step S5: Extract the risk subset to be processed based on the target risk set, and match the risk subset to be processed with the pre-built evidence-based knowledge base. Combine the patient's willingness data to generate candidate discharge care suggestion data for medical staff to review. The candidate discharge care suggestion data is a structured ranking result and does not directly output disease diagnosis conclusions or clinical treatment decisions.

[0086] In this embodiment, after obtaining the target risk set, it is necessary to extract the risk subset requiring care from the target risk set, match corresponding care measures to it, and finally generate candidate discharge care recommendation data for medical staff to review. The specific process is as follows: Step S51: Group and extract the target risk set according to the time sequence of risk occurrence to determine the risk subset to be processed.

[0087] Specifically, the target risk set contains risk elements corresponding to multiple risk out-of-bounds regions, and these risk elements exhibit different occurrence patterns according to their distribution over time. Grouping risks according to their chronological order of occurrence allows for the grouping of risk elements that are temporally adjacent or causally related. Based on this grouping, considering the severity and interveneability of each group, a subset of risks requiring proactive care is extracted. Severity is defined as the size of the corresponding risk out-of-bounds region, and interveneability refers to the existence of effective care measures in the evidence-based knowledge base that can reduce the risk.

[0088] Step S52: Match the risk subset to be processed with candidate care measures in the evidence-based knowledge base, and calculate a comprehensive score based on the risk matching degree of the candidate care measures to the risk subset, the corresponding level of evidence, and the adjustment weight determined by the patient's willingness data. Then, merge the candidate care measures according to the comprehensive score from high to low, and output candidate discharge care recommendation data for medical staff to review.

[0089] Specifically, each risk in the subset of risks to be processed is semantically matched and correlated with care measures in the evidence-based knowledge base. A set of candidate care measures corresponding to the risk is retrieved from the evidence-based knowledge base. Semantic matching refers to establishing a correspondence between the semantic relevance between the description of the risk element and the applicable scenario tag of the care measure. In this embodiment, a method is used to vectorize the description of the risk element and the applicable condition field of the care measure and calculate the cosine similarity. Care measures with a similarity higher than a preset matching threshold are included in the candidate set.

[0090] Then, a comprehensive score is calculated for each candidate care measure using the following formula: Score = α × Risk Matching Degree + β × Evidence Level Normalized Value + γ × Patient Willingness Weight - δ × Contraindication Penalty Item, where α, β, γ, and δ are the weight coefficients of the four components, satisfying α + β + γ = 1 and δ ≥ 0. In this embodiment, α = 0.4, β = 0.3, γ = 0.3, and δ = 1.0 are defaulted, and medical staff can adjust each weight coefficient through the system interface. The risk matching degree is the similarity score obtained in the step matching. The evidence level normalized value is mapped from the "6S pyramid" level to a value in the range of 0 to 1. The patient willingness weight is obtained by normalizing the patient's willingness data according to a preset scale. The contraindication penalty item is 1 when the care measure conflicts with the patient's previous contraindications, and 0 otherwise. The candidate care measures are sorted from high to low according to the comprehensive score, and care measures are selected sequentially from them under the premise of meeting the minimum care suitability requirements. Multiple care measures are integrated into a set of coordinated and consistent candidate discharge care recommendation data. The purpose of the above-mentioned fusion processing is to ensure the consistency of matching care measures while making the output results have reproducible scoring criteria, which facilitates review by medical staff.

[0091] The final output of candidate discharge care recommendations is a structured dataset awaiting review. It includes candidate care items arranged chronologically, field references for the implementation method, frequency, and precautions for each intervention item, as well as matching scores and evidence level markers, for healthcare professionals to reference and review in clinical decision-making. This candidate discharge care recommendation data comprehensively reflects the metabolic evolution characteristics, complication risk assessment results, self-management bias levels, and patient wishes of the analyzed patients. However, it does not directly output disease diagnoses or clinical treatment decisions, nor does it replace physicians' treatment decisions. Through structured sorting based on comprehensive scores, the output candidate discharge care recommendation data exhibits higher matching consistency and reviewability compared to the general template outputs in existing technologies. In summary, this embodiment acquires multi-dimensional patient data, constructs a metabolic evolution model, calculates the initial risk envelope boundary, simulates self-management bias to identify high-risk scenarios, generates a target risk set through geometric intersection operations and topological search, and finally combines an evidence-based knowledge base and patient wishes to generate candidate discharge care recommendation data. This method constitutes a complete data processing flow in steps such as alignment of multi-source medical time series data, normalized vector dimensionality reduction projection, geometric envelope construction, geometric intersection and comprehensive score ranking, which improves the accuracy, consistency and reproducibility of diabetes discharge risk identification and care matching.

[0092] See Figure 6 As shown, this embodiment of the invention discloses a data fusion-based discharge preparation analysis system for diabetic patients, comprising: The acquisition module 61 is used to acquire in-hospital blood glucose monitoring data, complication risk assessment data, lifestyle data, patient willingness data, and home blood glucose safety thresholds of the patient to be analyzed. Module 62 is used to construct a metabolic evolution model that reflects the blood glucose fluctuation pattern of the patient to be analyzed based on in-hospital blood glucose monitoring data and lifestyle data. Calculation module 63 is used to input complication risk assessment data into the metabolic evolution model and calculate the initial risk envelope boundary that characterizes the individualized risk data boundary of the patient to be analyzed. The first generation module 64 is used to identify high-risk scenarios where the risk data of the patient to be analyzed exceeds the limit based on the degree of self-management deviation in the lifestyle data, and compare the high-risk scenarios with the home blood glucose safety threshold to generate a target risk set that represents the degree of home risk exposure of the patient to be analyzed. The second generation module 65 is used to extract a subset of risks to be processed based on the target risk set, match the subset of risks to be processed with a pre-built evidence-based knowledge base, and generate candidate discharge care suggestion data for medical staff to review by combining patient willingness data; the candidate discharge care suggestion data is a structured sorting result and does not directly output disease diagnosis conclusions or clinical treatment decisions.

[0093] The modules mentioned above are connected via data interfaces. Data collected by module 61 is sequentially transmitted to module 62, module 63, first generation module 64, and second generation module 65 for processing, forming a complete data processing pipeline. Each module can be deployed on the same server or distributed across different computing nodes according to a microservice architecture, interacting with data via network communication. The data interface uses encrypted transmission and access control for the transmitted sensitive patient medical data, and records audit logs for the input and output of each module. The candidate discharge care recommendations output by the second generation module 65 are submitted to medical staff for review through an audit interface. Medical staff can confirm, modify, or return the recommendations. Only the results confirmed by medical staff are included in the patient's discharge file.

[0094] The data fusion-based discharge preparation analysis system for diabetic patients in this embodiment of the invention is used to implement the aforementioned data fusion-based discharge preparation analysis method for diabetic patients. Therefore, the specific implementation of the data fusion-based discharge preparation analysis system for diabetic patients can be found in the embodiment section of the data fusion-based discharge preparation analysis method for diabetic patients mentioned above. The specific implementation can be referred to the description of the corresponding embodiments, and will not be repeated here.

[0095] The present invention also provides an electronic device, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the data fusion-based analysis method for preparing diabetic patients for discharge as described above.

[0096] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the above-described data fusion-based analysis methods for preparing diabetic patients for discharge.

[0097] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as USB flash drives, read-only memory, random access memory, portable hard drives, magnetic disks, or optical disks.

[0098] Embodiments of the present invention also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above embodiments of the data fusion-based analysis method for preparing diabetic patients for discharge.

[0099] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0100] The foregoing has provided a detailed description of the data fusion-based analysis method and system for hospital discharge preparation of diabetic patients provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of this invention. It should be noted that those skilled in the art can make various improvements and modifications to this invention without departing from its principles, and these improvements and modifications also fall within the protection scope of this invention.

Claims

1. A data fusion-based analysis method for discharge preparation of diabetic patients, characterized in that, include: Obtain in-hospital blood glucose monitoring data, complication risk assessment data, lifestyle data, patient willingness data, and home blood glucose safety thresholds for the patients to be analyzed; Based on the in-hospital blood glucose monitoring data and lifestyle data, a metabolic evolution model reflecting the blood glucose fluctuation pattern of the patient to be analyzed was constructed. The complication risk assessment data are input into the metabolic evolution model to calculate the initial risk envelope boundary that characterizes the individualized risk data boundary of the patient to be analyzed. Based on the degree of self-management deviation in the lifestyle data, high-risk scenarios where the risk data of the patient to be analyzed exceeds the limit are identified, and the high-risk scenarios are compared with the home blood glucose safety threshold to generate a target risk set that characterizes the degree of home risk exposure of the patient to be analyzed. Based on the target risk set, a risk subset to be processed is extracted, and the risk subset to be processed is matched with a pre-built evidence-based knowledge base. Combined with the patient's intention data, candidate discharge care suggestion data is generated for medical staff to review; the candidate discharge care suggestion data is a structured sorting result.

2. The method according to claim 1, characterized in that, Before obtaining the in-hospital blood glucose monitoring data, complication risk assessment data, lifestyle data, patient preference data, and home blood glucose safety thresholds for the patients to be analyzed, the following is also included: Based on the PIPOST model, evidence-based questions for diabetic patients are extracted. Based on these questions, the retrieved medical literature data is summarized according to a pre-constructed evidence model to build an evidence-based knowledge base containing various diabetic discharge care measures. Each care measure in the evidence-based knowledge base is associated with a risk label, applicable conditions, evidence level, and recommendation strength.

3. The method according to claim 1, characterized in that, The lifestyle data includes dietary behavior characteristics, exercise behavior characteristics, and the degree of self-management deviation. The step involves constructing a metabolic evolution model reflecting the blood glucose fluctuation patterns of the patient under analysis, based on the in-hospital blood glucose monitoring data and lifestyle data, including: The characteristic data points of the in-hospital blood glucose monitoring data under the corresponding dietary behavior characteristics and exercise behavior characteristics are extracted as metabolic abnormality characteristic points. The metabolic abnormality characteristic points are data points of blood glucose data in the same period that meet any of the following abnormal conditions: the absolute value of the slope of blood glucose change exceeds a preset slope threshold, the blood glucose value exceeds a preset hyperglycemia threshold, and the blood glucose value is lower than a preset hypoglycemia threshold. Multiple metabolic abnormality feature points are fitted with curves according to time series to generate a metabolic evolution model.

4. The method according to claim 1, characterized in that, The step of inputting the complication risk assessment data into the metabolic evolution model to calculate the initial risk envelope boundary characterizing the individualized risk data boundary of the patient to be analyzed includes: The indicators in the complication risk assessment data are normalized according to their respective clinical reference thresholds to obtain normalized vectors; A correlation matrix is ​​constructed based on the synergistic influence coefficients among the indicators, and the normalized vector is mapped to geometric constraint parameters including the peak blood glucose safety limit, the hypoglycemia risk threshold, and the metabolic fluctuation tolerance window according to the correlation matrix. The geometric constraint parameters are applied to the metabolic evolution model to construct a closed three-dimensional risk envelope in a three-dimensional metabolic state space with time, blood glucose level, and comprehensive complication risk index as coordinate axes, and the closed three-dimensional risk envelope is determined as the initial risk envelope boundary.

5. The method according to claim 1, characterized in that, The step of comparing the high-risk scenarios with the home blood glucose safety threshold to generate a target risk set characterizing the home risk exposure level of the patient to be analyzed includes: Perform a geometric intersection operation between the metabolic data trajectory corresponding to the high-risk scenario and the home safety risk envelope determined by the home blood glucose safety threshold; Identify the points where the metabolic data trajectory penetrates the home safety risk envelope, and perform a local topological search centered on the points where the data crosses the boundary to extract the risk boundary regions and generate a target risk set.

6. The method according to claim 1, characterized in that, The identification of high-risk scenarios where the risk data of the patient to be analyzed exceeds the limits includes: The degree of self-management deviation is converted into behavioral deviation simulation parameters and superimposed on the metabolic evolution model. The abnormal deviation vector of the patient data trajectory to be analyzed relative to the initial risk envelope boundary is calculated after superimposing the behavioral deviation simulation parameters. Based on the area where the abnormal deviation vector exceeds the preset safety range, the corresponding time period information and environmental parameters are extracted to determine the high-risk scenario for the patient to be analyzed.

7. The method according to claim 1, characterized in that, The process involves extracting a subset of risks to be processed based on the target risk set, matching the subset of risks to be processed with a pre-built evidence-based knowledge base, and combining the patient's willingness data to generate candidate discharge care recommendations for medical staff to review. The candidate discharge care recommendations data are structured sorting results, including: The target risk set is grouped and extracted according to the time sequence of risk occurrence to determine the risk subset to be processed; The subset of risks to be processed is matched with candidate care measures in the evidence-based knowledge base. A comprehensive score is calculated based on the risk matching degree of the candidate care measures to the risk subset, the corresponding level of evidence, and the adjustment weight determined by the patient's willingness data. The candidate care measures are then fused according to the comprehensive score from high to low, and candidate discharge care recommendations are output for medical staff to review.

8. The method according to claim 5, characterized in that, The step of performing a local topology search centered on the boundary-crossing point to extract the resulting risk boundary-crossing region includes: Extract the edge tangent vector of the boundary crossing point, perform curvature detection on the surrounding nodes of the metabolic data trajectory along the edge tangent vector, connect the nodes whose curvature distortion parameters reach the preset deformation conditions to form a closed contour line, and extract the area enclosed inside the closed contour line as the risk boundary crossing area.

9. A data fusion-based analysis system for discharge preparation of diabetic patients, characterized in that, include: The acquisition module is used to acquire in-hospital blood glucose monitoring data, complication risk assessment data, lifestyle data, patient wishes data, and home blood glucose safety thresholds for the patients to be analyzed. The module is used to construct a metabolic evolution model that reflects the blood glucose fluctuation pattern of the patient to be analyzed, based on the in-hospital blood glucose monitoring data and lifestyle data. The calculation module is used to input the complication risk assessment data into the metabolic evolution model and calculate the initial risk envelope boundary that characterizes the individualized risk data boundary of the patient to be analyzed. The first generation module is used to identify high-risk scenarios where the risk data of the patient to be analyzed exceeds the limit based on the degree of self-management deviation in the lifestyle data, and compare the high-risk scenarios with the home blood glucose safety threshold to generate a target risk set that characterizes the degree of home risk exposure of the patient to be analyzed. The second generation module is used to extract a subset of risks to be processed based on the target risk set, match the subset of risks to be processed with a pre-built evidence-based knowledge base, and generate candidate discharge care suggestion data for medical staff to review by combining the patient's intention data; the candidate discharge care suggestion data is a structured sorting result.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, enables the implementation of the data fusion-based analysis method for discharge preparation of diabetic patients as described in any one of claims 1 to 8.

Images