Health risk intelligent monitoring and early warning method and platform for chronic disease population

Abstract

This invention belongs to the field of smart healthcare and digital health technology, providing a method and platform for intelligent monitoring and early warning of health risks for people with chronic diseases. The method includes: collecting physiological indicators, lifestyle habits, and traditional Chinese medicine constitution identification information, and fusing them to generate multi-dimensional time-series data on chronic disease characteristics; constructing and dynamically updating a personal dynamic health model; establishing a three-tiered progressive early warning mechanism of L1-level trend deviation, L2-level organ dysfunction, and L3-level complication risk, clarifying the triggering conditions for each level of early warning; based on the early warning level, calling upon a knowledge base of traditional Chinese and Western medicine intervention rules to generate and push personalized intervention plans; collecting feedback data to optimize the model and evaluate the intervention effect. The platform includes a smart terminal layer, a platform service layer, a model knowledge base, and an application interaction layer, adopting an edge-cloud collaborative architecture to improve monitoring real-time performance. This invention, through multi-dimensional data fusion and hierarchical early warning, can achieve early identification and accurate early warning of chronic disease risks.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare and digital health technology, and in particular to intelligent monitoring and early warning methods and platforms for health risks of people with chronic diseases. Background Technology

[0002] With the aging population and changing lifestyles, the incidence of chronic diseases such as diabetes, hypertension, and chronic kidney disease is rising year by year, becoming a major global public health issue. Chronic diseases are characterized by long duration, slow progression, and numerous complications, placing a heavy burden on patients' quality of life and healthcare resources. Therefore, achieving early monitoring, accurate early warning, and personalized intervention for chronic disease risks is a core requirement for improving chronic disease management. Currently, the development of smart wearable devices and mobile health technologies has provided new technological pathways for chronic disease monitoring. Existing technologies include solutions based on the collection of physiological indicators (such as heart rate, blood sugar, and blood pressure) by wearable devices and the issuance of risk warnings. For example, some existing technologies construct personal health benchmark models, compare real-time collected physiological indicators with benchmark values, and trigger warnings when indicators deviate from thresholds; other technologies attempt to combine traditional Chinese medicine constitution identification to achieve integrated traditional Chinese and Western medicine chronic disease management. However, existing chronic disease monitoring and early warning technologies still have many shortcomings and are difficult to meet the needs of precise clinical management. For example, some existing technologies have relatively simple early warning levels and mostly use fixed threshold comparisons, making it difficult to distinguish between short-term fluctuations and long-term abnormal trends in indicators. In addition, some schemes do not fully integrate data on lifestyle habits and environmental factors. Furthermore, personal health models are updated at a fixed frequency and do not fully take into account individual differences.

[0003] Furthermore, existing integrated traditional Chinese and Western medicine chronic disease management programs are mostly limited to a simple combination of "traditional Chinese medicine classification + Western medicine testing," without establishing a deep integration mechanism between the evolution of traditional Chinese medicine syndromes and Western medicine pathological models, nor incorporating the trajectory of syndrome evolution into the core logic of risk warning, making it difficult to fully leverage the advantages of the synergy between traditional Chinese and Western medicine. Summary of the Invention

[0004] This invention provides a method and platform for intelligent monitoring and early warning of health risks for people with chronic diseases. It aims to solve the problems of ambiguous early warning levels, insufficient data fusion dimensions, poor personalization adaptability, and unbalanced data processing and transmission efficiency in existing chronic disease monitoring and early warning technologies. By constructing a personal dynamic health model that integrates multimodal data and a three-layer progressive risk early warning mechanism, combined with the synergistic intervention strategy of traditional Chinese and Western medicine, it can achieve accurate identification, graded early warning, and personalized management of health risks for people with chronic diseases, thereby improving the intelligence and clinical applicability of chronic disease management.

[0005] This invention provides a method for intelligent monitoring and early warning of health risks for people with chronic diseases, including the following steps: S1: Collect physiological indicator data of patients with chronic diseases through smart wearable devices, and extract key features closely related to the target chronic disease from the physiological indicator data to form a physiological indicator feature dataset; at the same time, collect lifestyle habit data and TCM constitution identification information actively reported by patients with chronic diseases through mobile terminals or health management platforms; perform time-series alignment and structuring processing on the physiological indicator feature dataset, lifestyle habit data and TCM constitution identification information, and based on multivariate time-series analysis, mine the dynamic correlation between different modal data to generate multidimensional chronic disease feature time-series data that integrates spatiotemporal correlation features; S2: Based on multi-dimensional chronic disease feature time series data with spatiotemporal correlation characteristics, call and coordinate chronic disease-specific knowledge graphs, TCM syndrome evolution models, organ health status models and pathological models to construct a personal dynamic health model, and perform parameter learning and structure optimization based on continuous monitoring data and feedback information to achieve dynamic updates of the personal dynamic health model; S3: Based on a personal dynamic health model, a three-tiered progressive risk warning triggering mechanism is established, specifically including: L1-level trend deviation warning: Based on the key features extracted in S1, a dynamic baseline range based on personal historical data is maintained. When the key features continuously deviate from the personal dynamic baseline range and meet the preset statistical significance conditions, a trend deviation warning is triggered. L2-level organ dysfunction early warning: The feature data that triggers the L1-level trend deviation early warning is input into the organ health status sub-model optimized by chronic disease data to obtain the organ function status index and TCM syndrome attributes. When the organ function status index falls into the preset risk range, the organ function abnormality early warning is triggered. The organ health status sub-model is a sub-model of the personal dynamic health model. L3 chronic disease complication risk warning: Combining the dynamic correlation in S1, the output of L2 organ dysfunction warning is dynamically matched and jointly analyzed with the pathological model library of chronic disease complications. The probability of occurrence of specific chronic disease complications is calculated through the risk prediction model. When the probability of occurrence exceeds the preset threshold determined by clinical validation, a high risk warning for chronic disease complications is triggered. S4: Based on the three-tiered progressive risk warning, the system calls upon the knowledge base of Chinese and Western medicine intervention rules to generate a structured intervention plan and push it to the corresponding user terminal. At the same time, it collects feedback data after the intervention to optimize the personal dynamic health model and evaluate the intervention effect.

[0006] Furthermore, lifestyle data includes quantified dietary data and medication adherence data; TCM constitution identification information includes constitution type, syndrome evolution trajectory, and pulse characteristics data.

[0007] Furthermore, in step S1, in addition to collecting physiological indicator feature datasets, lifestyle data, and TCM constitution identification information, the exercise status, sleep quality, and ambient light data of patients with chronic diseases are passively collected through the sensors of smart wearable devices and integrated into multivariate time series analysis to form lifestyle behavior data.

[0008] Furthermore, the individual dynamic health model is dynamically updated using a hierarchical adaptive mechanism, setting differentiated model update frequencies and feature weight adjustment rules for patients with different chronic disease stages, age groups, or complication histories.

[0009] Furthermore, the L1-level trend deviation warning uses a change point detection algorithm to identify changes in indicator trends; the L3-level chronic disease complication risk warning uses a machine learning model for risk prediction, and the chronic disease complication pathology model library includes prediction models for various acute complications of chronic diseases and target organ damage.

[0010] Furthermore, based on the multi-dimensional chronic disease characteristic time-series data generated in step S1, the method also includes a deep processing step for this multi-dimensional chronic disease characteristic time-series data to optimize the accuracy of subsequent personal dynamic health model construction and risk warning. Specifically, this includes: A cross-modal association reasoning mechanism based on spatiotemporal graph neural network (ST-GNN) is introduced to construct a personalized health status evolution map. The personalized health status evolution map uses key physiological indicators as nodes and the evolution trajectory of TCM syndromes and dynamic changes in lifestyle habits as edge weights to capture the implicit risk transmission path in the three-element coupling relationship of physiology-behavior-constitution. In L2-level organ dysfunction early warning, multi-hop reasoning is used to identify subclinical health states that have not yet shown clinical symptoms, thus triggering subclinical risk warnings in advance.

[0011] Furthermore, in step S3, the triggering of the L3 level chronic disease complication risk warning requires the simultaneous fulfillment of the following three conditions: the target organ function index shows a clinically defined abnormal value, the TCM syndrome of the chronic disease patient matches the preset high-risk syndrome, and a pattern in the lifestyle data that is significantly associated with the risk of chronic disease complications appears.

[0012] Furthermore, the knowledge base of Chinese and Western medicine intervention rules stores the mapping relationship between risk patterns and intervention measures. The intervention measures are matched with the hidden risk transmission paths identified in the personalized health status evolution map to generate intervention plans targeting the hidden risk transmission paths.

[0013] Furthermore, the intervention plan is pushed out using a dynamic priority ranking mechanism. The ranking criteria include warning level, risk urgency, the patient's situation, the first key criterion, and the second key criterion. A comprehensive priority score is calculated based on the weights assigned to each criterion to determine the push order. The first key criterion is the weight value of the abnormal physiological indicator that triggers the warning in the latent risk transmission path, and the second key criterion is the strength of the dynamic correlation between the abnormal physiological indicator that triggers the warning and other data in the multi-dimensional chronic disease characteristic time series data.

[0014] A smart health risk monitoring and early warning platform for people with chronic diseases, including: The intelligent terminal layer includes intelligent wearable devices for collecting physiological indicator data, as well as client applications for data entry, information viewing, and interaction. The platform service layer includes a data generation module, a risk assessment engine, and an intervention module. The data generation module is used for data collection, multivariate time series analysis, and generation of multi-dimensional chronic disease characteristic time series data that integrates spatiotemporal correlation features. The risk assessment engine is used to build a personal dynamic health model and implement a three-layer progressive risk warning logic. The intervention module is used to generate, push, and evaluate intervention plans. The platform service layer integrates edge computing and cloud collaborative processing architecture, supporting the preliminary calculation of L1 level trend deviation warning on the smart wearable device, and only uploading abnormal data fragments and context time series information to the cloud for L2 and L3 level in-depth analysis; The model knowledge base, connected to the platform service layer, is used to store and maintain personal dynamic health models, organ health status models, pathological models, chronic disease-specific knowledge graphs, TCM syndrome evolution models, chronic disease complication pathological model libraries, and TCM and Western medicine intervention rule knowledge bases. The application interaction layer is used to provide users, their families, and doctors with differentiated health information displays, early warning notifications, and interactive functions.

[0015] Compared with the prior art, the present invention has the following advantages and beneficial effects: First, a personal dynamic health model integrating multimodal data was constructed, breaking through the limitations of traditional single-dimensional data monitoring. By integrating physiological indicators collected by smart wearable devices, proactively reported lifestyle habits and TCM constitution identification information, as well as passively collected lifestyle behavior data such as exercise, sleep, and environment, multivariate time series analysis was used to mine the dynamic correlation between data and generate multidimensional time series data integrating spatiotemporal characteristics. On this basis, in conjunction with chronic disease knowledge graphs, TCM syndrome evolution models, and other multi-domain models, combined with a hierarchical adaptive update mechanism, the model is dynamically optimized as individual data accumulates and the condition changes, making health assessments more in line with individual realities and improving the model's ability to represent complex chronic disease states.

[0016] Second, a three-tiered progressive risk warning triggering mechanism was designed, enabling refined graded warnings from abnormal trends to organ dysfunction and then to complication risk. The L1 level trend deviation warning is based on an individual's dynamic baseline and captures early abnormal trends of indicators through strict statistical conditions. The L2 level organ dysfunction warning calls upon the organ health status sub-model and combines organ function index with TCM syndrome attributes to judge organ-level abnormalities. The L3 level complication risk warning achieves accurate identification of high-risk complications through dynamic pattern matching with the pathological model library and prediction by machine learning models. The three levels of warnings are progressive, and the L3 level warning must simultaneously meet the core conditions of target organ dysfunction, TCM syndrome matching, and lifestyle risk patterns, effectively improving the specificity and accuracy of the warnings and avoiding over-warning caused by a single abnormal indicator.

[0017] Third, the introduction of integrated traditional Chinese and Western medicine intervention strategies enhances the personalization and clinical applicability of intervention plans. Based on the three-tiered early warning results, the system invokes the knowledge base of traditional Chinese and Western medicine intervention rules, and combines the implicit risk transmission paths identified by the personalized health status evolution map to generate targeted intervention plans. The intervention plan push adopts dynamic priority sorting, comprehensively considering multiple factors such as early warning level, risk urgency, patient scenario, and the weight and correlation strength of indicators in the risk transmission path to ensure that key intervention information is delivered first. At the same time, by collecting intervention feedback data, the model is continuously optimized and the effect is evaluated, forming a closed-loop management of "monitoring-early warning-intervention-feedback-optimization", realizing intelligent support for the entire chain from risk identification to intervention implementation. In particular, the integration of traditional Chinese medicine constitution and syndrome elements provides a more holistic and personalized solution for chronic disease management.

[0018] Fourth, it adopts an edge computing and cloud collaborative processing architecture, which optimizes data processing and transmission efficiency and system response performance. The initial calculation of L1 level early warning is completed on the smart wearable device, and only abnormal data fragments and context information are uploaded to the cloud for in-depth analysis. This significantly reduces the amount of data transmission and cloud computing pressure, ensuring the real-time performance of L1 level early warning, while also ensuring the computing power support for L2 and L3 level in-depth analysis. This layered processing mode balances the needs of real-time monitoring and in-depth analysis, enabling the system to run efficiently on terminal devices with limited resources, and improving the platform's practicality and user experience.

[0019] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description and the accompanying drawings.

[0020] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0021] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 A schematic diagram illustrating the steps of intelligent health risk monitoring and early warning methods for people with chronic diseases. Figure 2 A schematic diagram illustrating the components of a three-tiered, progressive risk warning triggering mechanism; Figure 3 This is a schematic diagram of the structure of a smart health risk monitoring and early warning platform for people with chronic diseases. Detailed Implementation

[0022] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0023] This invention provides a method for intelligent monitoring and early warning of health risks for people with chronic diseases, such as... Figure 1 As shown, it includes the following steps: S1: Collect physiological indicator data from patients with chronic diseases through smart wearable devices, and extract key features closely related to the target chronic disease from the physiological indicator data to form a physiological indicator feature dataset; simultaneously, collect lifestyle habit data and TCM constitution identification information actively reported by patients with chronic diseases through mobile terminals or health management platforms; perform time-series alignment and structured processing on the physiological indicator feature dataset, lifestyle habit data, and TCM constitution identification information, and based on multivariate time-series analysis, mine the dynamic correlation between different modalities of data to generate multidimensional chronic disease feature time-series data that integrates spatiotemporal correlation features; key features include, but are not limited to, systolic and diastolic blood pressure, fasting and postprandial blood glucose, and heart rate variability; S2: Based on multi-dimensional chronic disease feature time series data with spatiotemporal correlation characteristics, call and coordinate chronic disease-specific knowledge graphs, TCM syndrome evolution models, organ health status models and pathological models to construct a personal dynamic health model, and perform parameter learning and structure optimization based on continuous monitoring data and feedback information to achieve dynamic updates of the personal dynamic health model; The specific implementation of the personal dynamic health model is as follows: A Transformer-based temporal prediction network is used, with an input feature dimension of 128, including time windows (7 days in length) for physiological indicators, lifestyle habits, and TCM constitution characteristics. The network consists of a 4-layer encoder-decoder structure, with 8 attention heads in each layer. The model employs a joint optimization of mean squared error loss and contrastive learning loss, with an initial learning rate of 1e-4, using the AdamW optimizer, and a batch size of 32. S3: Based on a personal dynamic health model, establish a three-tiered progressive risk warning triggering mechanism, such as... Figure 2 As shown, it specifically includes: L1-level trend deviation warning: Based on the key features extracted in S1, a dynamic baseline range based on personal historical data is maintained. When the key features continuously deviate from the personal dynamic baseline range and meet the preset statistical significance conditions, a trend deviation warning is triggered. The statistical significance conditions are: the key features deviate from the personal dynamic baseline range for three consecutive collection periods, and the deviation exceeds ±10% of the baseline mean, verified by a t-test with P<0.05. All the above conditions were verified and determined through multi-center chronic disease cohort data. The personal dynamic baseline range is generated by the personal dynamic health model based on historical monitoring data. L2-level organ dysfunction early warning: The feature data that triggers the L1-level trend deviation early warning is input into the organ health status sub-model optimized by chronic disease data to obtain the organ function status index and TCM syndrome attributes. When the organ function status index falls into the preset risk range, the organ function abnormality early warning is triggered. The organ health status sub-model is a sub-model of the personal dynamic health model. L3 chronic disease complication risk warning: Combining the dynamic correlation in S1, the output of L2 organ dysfunction warning is dynamically matched and jointly analyzed with the pathological model library of chronic disease complications. The probability of occurrence of specific chronic disease complications is calculated through the risk prediction model. When the probability of occurrence exceeds the preset threshold determined by clinical validation, a high risk warning for chronic disease complications is triggered. S4: Based on the three-tiered progressive risk warning, call the knowledge base of Chinese and Western medicine intervention rules to generate a structured intervention plan and push it to the corresponding user terminal. At the same time, collect feedback data after the intervention to optimize the personal dynamic health model and evaluate the intervention effect. In practical application, taking an elderly user with type 2 diabetes and hypertension as an example, the specific implementation process is as follows: In step S1, the user wears a smart wearable device that supports continuous monitoring, which continuously collects physiological data such as heart rate, blood oxygen saturation, blood pressure (estimated by the pulse wave transit time PTT algorithm), blood glucose trend (calculated by photoplethysmography PPG signal combined with calibration model), body temperature and activity level every day. The device uploads raw data to the health management cloud platform every 15 minutes. In the mobile health management app, the user fills out a lifestyle questionnaire once a week, including dietary structure (frequency of high salt / high fat intake), exercise frequency and duration, sleep quality score, and self-assessment of emotional state; at the same time, the user completes the TCM constitution identification scale (based on the "Classification and Judgment of TCM Constitution" standard), and is determined to be a mixed constitution type of "phlegm-dampness constitution + yin deficiency and yang hyperactivity constitution". The system performs time-series alignment processing on the aforementioned heterogeneous data: using UTC timestamps as a benchmark, the high-frequency sampling data (time granularity 15 minutes) from wearable devices is downsampled into a daily average sequence, and correlated with daily reported lifestyle habit tags (such as "walking less than 3000 steps on the day", "dinner is too salty") and weekly updated TCM constitution attributes; multivariate time-series analysis is used to explore the dynamic influence relationship between different modalities; for example, it was found that when sodium intake is high for two consecutive days and nocturnal heart rate variability (HRV) decreases by more than 15%, the probability of systolic blood pressure rising the next morning is 78.3% (p<0.01); and individuals with "phlegm-dampness constitution" are more likely to experience worsening insulin resistance in the context of a sustained increase in resting heart rate; The final dataset generated is a multi-dimensional time-series dataset of chronic disease characteristics, including: physiological indicators: fasting blood glucose trend, mean ambulatory blood pressure (MAP), HRV-LF / HF ratio, and body mass index change rate; lifestyle data: daily steps, sleep efficiency, alcohol consumption frequency, and dietary score; traditional Chinese medicine constitution identification information: constitution imbalance score and pulse wiry and slipperyness score; and spatiotemporal correlation features: the lagged response strength between the previous day's high-salt diet and the increase in blood pressure upon waking on the current day, and the intertemporal correlation coefficient between exercise absence and increased blood glucose fluctuations. This dataset is indexed by user ID and stored in a personal health database on a daily basis for subsequent modeling. In step S2, based on the multi-dimensional chronic disease feature time-series data generated in S1, the system calls the following four core knowledge modules to collaboratively construct the user's personal dynamic health model: Chronic disease knowledge graph: Loaded with a diabetes-hypertension comorbidity sub-graph, which covers disease nodes (such as "type 2 diabetes" and "essential hypertension"), complication pathways (such as "diabetic nephropathy to chronic renal failure"), risk factor chains (such as "insulin resistance to dyslipidemia and then to atherosclerosis") and their logical relationships (causation, comorbidity, inhibition, etc.). Traditional Chinese Medicine Syndrome Evolution Model: Based on the user's potential syndrome transformation path from "phlegm-dampness accumulation to liver yang hyperactivity and then to qi and yin deficiency", a state transition matrix and causal weights are set, and the current main syndrome tendency is dynamically assessed in combination with the user's constitution type. Organ health status model: Sub-models are established for the heart, kidneys, blood vessels and pancreas. Key physiological parameters (such as eGFR, carotid IMT growth rate, HOMA-IR index) are input and the functional status index of each organ (0-100 points) is output. Pathological Model Library: Integrates mechanistic models such as diabetic microvascular complications and hypertension target organ damage, simulating the cumulative damage process of long-term metabolic disorders to tissues and organs; The initial model is configured based on the user's historical physical examination reports (e.g., HbA1c range of 6.8%-8.2% over the past 3 years, and office blood pressure of 140-160 / 85-95 mmHg) and baseline characteristics. Subsequently, the system receives newly collected data daily and adjusts the model parameters using an online learning mechanism. For example, if monitoring for a week shows that the user's fasting blood glucose level decreases by an average of 0.6 mmol / L the day after an evening walk, the causal weight of "regular aerobic exercise leading to improved insulin sensitivity" is strengthened. If the TCM identification results show an increase in the score of "yin deficiency and fire excess" in a certain week, the activation threshold of liver yang hyperactivity syndrome is increased accordingly. The model performs structural optimization every 7 days: the most predictive feature combination is identified through an attention mechanism, low-contribution variables are pruned, a personalized path-dependent network is formed, and the updated model version is timestamped and retains rollback capability. In step S3, the three-layer progressive risk warning triggering mechanism operates as follows: For L1 level trend deviation warning, if the system detects that the user's dynamic blood pressure values ​​for three consecutive collection cycles (45-minute intervals) are 158 / 92 mmHg, 161 / 94 mmHg, and 159 / 93 mmHg, which exceed the user's personal dynamic baseline range (average of 142±8 / 86±6 mmHg over the past 3 months), the systolic blood pressure deviation is calculated to be +13.4% (>±10%), and the diastolic blood pressure deviation is +8.1%. After a one-sample t-test, P=0.037<0.05, which meets the statistical significance condition, the system immediately triggers the L1 level "abnormally high blood pressure trend" warning, and the event time is recorded as T0. For L2-level organ dysfunction warnings, multidimensional features within 72 hours before and after time T0 are input into the organ health status sub-model to comprehensively assess cardiac load, renal filtration function, and vascular elasticity. The cardiovascular system functional status index output by the model drops to 52 points (normal >75, borderline 60-75, risk <60), and the "liver yang hyperactivity" syndrome attribute is identified (confidence 86%). Since the index falls into the "risk range", the system triggers an L2-level "cardiovascular dysfunction with predominance of TCM syndrome" warning and pushes it to the family doctor workstation and family member app. For the L3 level chronic disease complication risk warning, the system retrieves the pathological model library of chronic disease complications and matches the current state pattern: long-term hypertension + recent sudden rise in blood pressure + continuous decline in HRV + TCM "liver yang hyperactivity" manifestation. Through the Bayesian network inference engine, it jointly analyzes the activation probability of key nodes in the stroke occurrence path and finds that the probability of ischemic stroke in the next 7 days is 12.7% (the baseline is about 0.3%). The preset threshold is 5%, so the L3 level "high risk of acute cerebrovascular event" warning is triggered and the emergency intervention process is initiated. For step S4, the system automatically calls the knowledge base of traditional Chinese and Western medicine intervention rules and generates a graded intervention strategy based on the three-level warning results: For L1 level trend deviation warning: it is recommended to increase the frequency of daily blood pressure measurement to twice a day, morning and evening, avoid drastic emotional fluctuations, and reduce daily sodium intake; For L2 level organ dysfunction warning: activate the remote family doctor consultation channel and recommend adding a small dose of CCB antihypertensive drug (such as amlodipine 5mg). The system simultaneously prescribed a traditional Chinese medicine herbal tea formula (modified Tianma Gouteng Decoction: 6g Tianma, 10g Gouteng, 10g Chrysanthemum, 15g Cassia Seed), one dose to be brewed and drunk daily. For L3-level chronic disease complication risk warnings: a red alert notification was sent to the user's community health service center, priority outpatient appointments were arranged, and the user was reminded to seek immediate medical evaluation via SMS and voice call. Emergency guidance videos were simultaneously pushed to family members, including emergency response steps for sudden headaches, slurred speech, etc. All intervention measures were packaged into structured JSON format instruction packages. After the intervention was implemented, the system continuously collected feedback data: including whether the user took medication on time, whether they went to the hospital, the blood pressure change curve after medication, and the improvement of TCM symptom scores. This feedback was used as reward signals input into the reinforcement learning module to optimize the accuracy of the intervention response prediction branch in the personal dynamic health model. For example, if the blood pressure returned to a safe range within 3 days after using the Tianma Gouteng Decoction, the recommendation priority of this formula in the "liver yang hyperactivity type hypertension" scenario would be enhanced.

[0024] The working principle of the above technical solution is as follows: In the data acquisition and feature fusion stage (S1), physiological indicator data of patients with chronic diseases, such as heart rate, blood pressure, blood oxygen saturation, and blood glucose fluctuations, are continuously collected through smart wearable devices. Key temporal features highly correlated with target chronic diseases (such as diabetes, hypertension, and coronary heart disease) are extracted from these data to form a high-dimensional physiological indicator feature dataset. The smart wearable device is a wrist-worn physiological monitoring device that integrates a PPG photoplethysmography (PPG) sensor, a three-axis accelerometer, and a gyroscope sensor. The PPG signal sampling rate is no less than 50Hz, and the accelerometer and gyroscope sampling rates are no less than 25Hz, enabling continuous and non-invasive acquisition of physiological and motion state data. At the same time, information on patients' proactively reported lifestyle habits (such as dietary structure, exercise frequency, sleep quality, and emotional state) and TCM constitution identification results (such as nine constitution classifications including Qi deficiency, Yang deficiency, and phlegm-dampness) are collected through mobile terminals or health management platforms. To achieve effective collaborative analysis of cross-modal data, the system performs strict timestamp alignment and structured coding on the above heterogeneous data to ensure consistency and comparability of data from different sources on the timeline. Based on this, multivariate time series analysis methods (such as vector autoregression (VAR) model, dynamic time warping (DTW), Granger causality test, etc.) are used to explore the dynamic correlation between changes in physiological indicators and lifestyle habits and TCM constitution. For example, the combined effect pattern of "nocturnal blood pressure rise" and "high salt intake at dinner + phlegm-dampness constitution" is identified, or the coupling trend of "continuous rise in fasting blood glucose" and "lack of regular exercise for three consecutive days + Qi and Yin deficiency syndrome" is identified. These deep correlations are abstracted into multidimensional chronic disease feature time series data that integrates spatiotemporal dependence, which serve as the basic input for subsequent modeling. Entering the personal health modeling stage (S2), driven by fused multi-dimensional time-series data, four major medical knowledge models are integrated: chronic disease-specific knowledge graph (covering disease staging, complication pathways, and risk factor networks), TCM syndrome evolution model (describing the evolution of syndromes over time, such as the transmission path of spleen deficiency → qi deficiency → yang deficiency), organ health status model (quantifying the degree of functional decline of the heart, liver, kidneys, etc.), and pathological model (reflecting microscopic mechanisms such as tissue damage and metabolic disorders). Through the model synergy mechanism, the above-mentioned prior medical knowledge is embedded into the individualized data-driven modeling process to construct a personal dynamic health model with interpretability. This personal dynamic health model employs an online learning architecture (such as recurrent neural networks (RNNs), a Transformer-based time series model combined with a Bayesian update strategy), which can continuously adjust internal parameters, correct the weights of each sub-module, and optimize structural connections based on newly incoming monitoring data. This enables accurate characterization and dynamic approximation of the evolution trajectory of an individual's health status. For example, when a diabetic patient shows a declining trend in glomerular filtration rate, the model will automatically increase the influence weight of the kidney function sub-model and link it with the TCM "kidney deficiency" syndrome node to enhance the sensitivity of risk perception for future complications (such as diabetic nephropathy). In terms of risk warning triggering mechanism design (S3), a three-level progressive warning system is proposed, which focuses and filters layer by layer, taking into account both early sensitivity and clinical specificity: L1-level trend deviation early warning: Focuses on detecting abnormal fluctuations at the individual level. It establishes a dynamic baseline range for key physiological characteristics based on personal historical data (such as rolling 60-day mean ± standard deviation) and sets strict statistical criteria. That is, when a key characteristic exceeds the baseline range for three consecutive sampling periods and the deviation exceeds ±10% of the baseline mean, and is verified by paired t-test with P<0.05, it is determined that there is a significant trend deviation and triggers a primary early warning. This mechanism avoids the false alarm problem caused by the group threshold and improves individual adaptability. Level L2 organ dysfunction warning: Based on the L1 warning, the optimized organ health status sub-model (belonging to the personal dynamic health model) is invoked to perform in-depth analysis of the abnormal feature sequence, and output organ function status index (such as the NYHA classification mapping value of cardiac function, FIB-4 score of liver fibrosis, etc.) and the corresponding TCM syndrome attributes (such as "blood stasis in the heart" and "qi stagnation in the liver"). If the organ function status index falls into the preset risk range (such as >70% functional impairment threshold), the level 2 warning is triggered, indicating that there may be organic changes. L3 level chronic disease complication risk warning: Further combining the multimodal dynamic correlations mined in S1, the L2 output results are dynamically matched with the built-in chronic disease complication pathology model library. For example, the combination pattern of "deterioration of renal function index, aggravation of TCM kidney yang deficiency syndrome, increased nocturia, and poor blood pressure control" is compared with the diabetic nephropathy development path model for similarity calculation. The ensemble learning model is used to predict the probability of progression to overt proteinuria within the next 3 months. When the probability of this risk exceeds the clinical intervention threshold (e.g., ≥30%), the highest level of complication high-risk warning is triggered, and the emergency intervention process is initiated. Finally, in the intervention response and closed-loop optimization stage (S4), based on the current triggered warning level, the knowledge base of integrated traditional Chinese and Western medicine intervention rules is invoked to generate a structured and personalized intervention plan. This knowledge base includes Western medicine lifestyle recommendations, medication adjustment prompts, and traditional Chinese medicine conditioning plans (such as recommendations for Chinese herbal prescriptions, acupuncture point guidance, and constitution conditioning suggestions), and is output in order of priority. The intervention plan is pushed to the user's APP, doctor's workstation, or family caregiver's terminal. At the same time, the user's compliance behavior and physiological feedback data (such as blood pressure improvement after implementing the exercise plan) are recorded to back-train the personal dynamic health model and improve its predictive accuracy and intervention effectiveness assessment capabilities. In summary, this technical solution achieves seamless transformation from raw data to clinical decision support through a full-link design of "multi-source perception - feature fusion - dynamic modeling - hierarchical early warning - closed-loop intervention". It integrates modern medical quantitative indicators while retaining the holistic view and syndrome differentiation and treatment characteristics of traditional Chinese medicine, forming an intelligent prevention and control system for chronic disease risks with self-evolution capabilities.

[0025] The beneficial effects of the above technical solution are as follows: The solution provided in this embodiment can significantly improve the accuracy and timeliness of health risk monitoring for people with chronic diseases. Through deep fusion and dynamic modeling of multi-source heterogeneous data, it breaks through the limitations of traditional monitoring where physiological indicators, lifestyle habits, and traditional Chinese medicine constitution information are separated, achieving a comprehensive and three-dimensional depiction of an individual's health status. Its three-level progressive early warning system, from trend deviation to organ dysfunction to complication risk, delves deeper layer by layer, ensuring both the keen detection of early risks and, through the deep involvement of rigorous statistical criteria and medical knowledge models, effectively reducing the false alarm rate and improving the reliability of clinical decision-making. The online learning capability of the dynamic health model enables it to continuously adapt to changes in individual health status and constantly optimize the prediction model, thus achieving personalized health management. The intervention plan generation mechanism that integrates traditional Chinese and Western medicine not only provides patients with scientific and specific health guidance but also provides doctors with strong decision support, promoting the transformation of health management from passive response to proactive prevention. Ultimately, through closed-loop optimization of data collection, model prediction, early warning intervention, and effect feedback, a sustainable and evolving intelligent prevention and control ecosystem for chronic disease risks is constructed, which helps to delay disease progression, reduce the occurrence of complications, improve patients' quality of life, and provide strong support for the efficient use of medical and health resources.

[0026] In one embodiment, lifestyle data includes quantified dietary data and medication adherence data; TCM constitution identification information includes constitution type, syndrome evolution trajectory, and pulse characteristics data.

[0027] The working principle of the above technical solution is as follows: First, the system collects users' dietary behavior data (such as the types, weights, cooking methods, eating times and frequencies of food consumed daily) through smart terminal devices (such as smart bracelets, smart pillboxes, diet tracking apps, etc.), and converts it into standardized nutritional component indicators (such as calories, protein, fat, carbohydrates, key vitamins and minerals content, etc.), thus completing the quantitative processing of dietary data; simultaneously, based on the smart pillbox's on / off records, medication reminder response status, and medication records reported by the patient, combined with the preset medication plan... The system calculates medication adherence indicators (such as on-time medication rate, dosage accuracy, and treatment completion rate) to quantify medication adherence data. For TCM constitution identification information, it integrates the results of TCM constitution identification scales, doctor's consultation records, and pulse waveform data (such as pulse rate, pulse strength, pulse width, and smoothness) collected by intelligent pulse imaging devices. It uses TCM constitution identification algorithms to determine the user's basic constitution type (such as balanced constitution, yang deficiency constitution, yin deficiency constitution, etc.) and combines the results of previous constitution identifications with syndrome description information to construct a syndrome evolution trajectory model and dynamically record the changes in constitution and syndrome.

[0028] The beneficial effects of the above technical solution are as follows: by adopting the solution provided in this embodiment, it is possible to achieve multi-dimensional and refined collection and quantitative analysis of the lifestyle habits and TCM constitution information of people with chronic diseases, and provide more comprehensive underlying data support for health risk monitoring.

[0029] In one embodiment, in step S1, in addition to collecting physiological indicator feature datasets, lifestyle data, and TCM constitution identification information, the exercise status, sleep quality, and ambient light data of patients with chronic diseases are passively collected through the sensors of smart wearable devices and integrated into multivariate time series analysis to form lifestyle behavior data.

[0030] The working principle of the above technical solution is as follows: Smart wearable devices (such as smartwatches and wristbands) have built-in three-axis accelerometers, gyroscopes, and heart rate sensors. These continuously collect raw data such as the patient's motion acceleration, angular velocity, and heart rate variability. Combined with timestamp information, dynamic time warping algorithms and threshold judgment methods are used to identify the patient's movement type (such as sitting, walking, running, climbing stairs, etc.). The duration, intensity (by calculating the MET value, i.e., metabolic equivalent), and frequency of different movement types are statistically analyzed to form a motion state dataset. Simultaneously, the device's light and motion sensors monitor the number of body movements, turning frequency, heart rate fluctuations, and changes in ambient light intensity during sleep. Based on the principles of polysomnography monitoring, a sleep structure analysis model is constructed, dividing the sleep process into wakefulness... The study analyzed sleep quality indicators such as the duration of each sleep stage (sleep, light sleep, deep sleep, and REM sleep), total sleep duration, sleep latency, and number of awakenings during the night. Additionally, it collected real-time data on light intensity, duration, and spectral composition of the patient's environment using an ambient light sensor. This data, combined with geographic location and time information, analyzed the correlation between light exposure patterns and circadian rhythms, creating an ambient light dataset. This data on movement, sleep quality, and ambient light was stored in time-series format and preprocessed using an edge computing module (e.g., noise reduction, outlier removal, data completion). It was then aligned with physiological indicator data, lifestyle data, and traditional Chinese medicine constitution identification information, integrating them into multi-dimensional lifestyle behavior data to provide richer behavioral pattern characteristics for subsequent health risk monitoring.

[0031] The beneficial effects of the above technical solution are as follows: By adopting the solution provided in this embodiment, objective behavioral data such as exercise, sleep and ambient light in the daily life of patients with chronic diseases can be passively and imperceptibly collected through intelligent wearable devices. This avoids the subjective bias and data loss problems that may occur when relying on patients to actively record data in the traditional way, and significantly improves the continuity, objectivity and comprehensiveness of life behavior data collection.

[0032] In one embodiment, the personal dynamic health model is dynamically updated using a hierarchical adaptive mechanism, and differentiated model update frequency and feature weight adjustment rules are set for patients with different chronic disease stages, age groups or complication histories. The specific implementation of the hierarchical adaptive mechanism is as follows: Patients are divided into 8 categories based on their chronic disease stage (early / mid / late stage), age group (≤45 years / 46-65 years / ≥66 years), and history of complications (present / absent). The model update frequency is as follows: the late stage + ≥66 years + history of complications group is updated daily; the early stage + ≤45 years + history of complications group is updated every 14 days; and the remaining groups are updated every 7 days. Feature weight adjustment adopts a gradient-based feature importance assessment. After each round of updates, the top-20% features are weighted (multiplied by 1.2), the bottom-10% features are weighted (multiplied by 0.8), and then renormalized.

[0033] The working principle of the above technical solution is as follows: First, patients are stratified based on their chronic disease type (e.g., diabetes, hypertension, coronary heart disease), current disease stage (e.g., preclinical diabetes, diabetic stage, complication stage), age group (e.g., youth, middle-aged, elderly), and history of complications (e.g., diabetic nephropathy, hypertensive heart disease). For patients in a rapidly progressing stage (e.g., diabetic complication stage) or elderly patients, the model update frequency is set to a higher level (e.g., daily updates) to quickly capture dynamic changes in their health status. For early-stage patients with stable disease control or younger patients, the model update frequency can be appropriately reduced (e.g., weekly updates) to reduce computational resource consumption while ensuring model accuracy. Regarding feature weight adjustment, different health influencing factors are assigned differentiated weights for patients in different strata; for example... For patients with hypertension, the weight of blood pressure-related physiological indicators (such as the fluctuation range of systolic and diastolic blood pressure) is higher than that of other features. For patients with diabetes, the weight of blood glucose monitoring data and dietary behavior characteristics (such as carbohydrate intake and meal timing regularity) is significantly increased. For patients with a history of complications, specific indicators related to complications (such as the urinary microalbumin / creatinine ratio in diabetic patients and blood lipid indicators in coronary heart disease patients) will be given higher weights to ensure that the model can focus on the key factors that have the greatest impact on their health risks. During the model update process, the system will periodically review the patient's historical health data and prediction results. If the prediction error is found to exceed the preset threshold (such as three consecutive prediction errors greater than 15%), an emergency update mechanism will be automatically triggered to readjust the feature weights and model parameters to ensure the model's adaptability to changes in individual health status and prediction accuracy.

[0034] The beneficial effects of the above technical solution are as follows: by adopting the solution provided in this embodiment, the personal dynamic health model can be significantly improved in terms of its personalized adaptation ability and dynamic response speed to the health status of people with chronic diseases.

[0035] In one embodiment, the L1-level trend deviation warning uses a change point detection algorithm to identify changes in indicator trends; the risk prediction model used for the L3-level chronic disease complication risk warning is a machine learning model. The chronic disease complication pathological model library includes prediction models for various acute complications of chronic diseases and target organ damage. The change point detection algorithm uses the CUSUM algorithm and the PELT algorithm. The machine learning model includes, but is not limited to, random forest, long short-term memory network LSTM, and Bayesian network, which can be adaptively selected according to the type of chronic disease and data characteristics.

[0036] The working principle of the above technical solution is as follows: The L1 level trend deviation warning uses the CUSUM algorithm or PELT algorithm to scan the time series data of the patient's continuously collected physiological indicators (such as blood pressure, blood sugar, blood lipids, etc.) in real time. When the algorithm detects a significant deviation from the normal baseline trend in the indicator sequence (such as a continuous increase in blood pressure or an abnormal increase in blood sugar fluctuations), the L1 level warning is triggered, prompting medical staff and patients to pay attention to the early abnormal trends of the indicators. The L3 level chronic disease complication risk warning first calls the corresponding prediction model from the chronic disease complication pathology model library based on the patient's underlying disease type (such as diabetes, hypertension, etc.). For example, for diabetic patients, the prediction model for diabetic nephropathy or diabetic retinopathy built based on random forest or LSTM models is called. This machine learning model will input the patient's recent clinical indicators, lifestyle data, and historical health records, etc. Through the feature calculation and weight allocation within the model, it outputs the risk probability value of the patient developing specific complications in a certain period of time in the future (such as 6 months or 1 year). When the risk probability value exceeds the preset warning threshold, the system triggers the L3 level warning, providing decision support for clinical intervention.

[0037] The beneficial effects of the above technical solution are as follows: by adopting the solution provided in this embodiment, the health risks of people with chronic diseases can be accurately and progressively managed through a hierarchical early warning mechanism.

[0038] In one embodiment, based on the generation of multi-dimensional chronic disease feature time-series data in step S1, the method further includes a deep processing step for this multi-dimensional chronic disease feature time-series data to optimize the accuracy of subsequent personal dynamic health model construction and risk warning, specifically including: A cross-modal association reasoning mechanism based on spatiotemporal graph neural network (ST-GNN) is introduced to construct a personalized health status evolution map. The personalized health status evolution map uses key physiological indicators as nodes and TCM syndrome evolution trajectory and dynamic changes in lifestyle as edge weights. The spatiotemporal graph neural network uses key physiological indicators as nodes and TCM syndrome evolution and lifestyle changes as edge weights. It adopts a two-stage learning strategy of pre-training and fine-tuning, and uses time-series prediction loss and association matching loss to jointly optimize the model, so as to realize the mining of implicit risk transmission paths in the physiological-behavioral-constitutional three-element coupling relationship. In L2-level organ dysfunction early warning, multi-hop reasoning is used to identify subclinical health states that have not yet shown clinical symptoms, thus triggering subclinical risk warnings in advance.

[0039] The working principle of the above technical solution is as follows: A cross-modal association reasoning mechanism based on spatiotemporal graph neural networks (ST-GNN) is introduced. Its working principle involves modeling multi-dimensional chronic disease characteristic time-series data into a dynamic graph structure. Key physiological indicators (such as blood pressure, blood sugar, and heart rate variability) serve as nodes in the graph, while the edge weights are jointly encoded by the evolution trajectory of TCM syndromes (such as the dynamic transformation of Qi deficiency and Yin deficiency syndromes) and changes in lifestyle habits (such as sleep rhythm, dietary patterns, and exercise frequency). This achieves a multi-level expression of an individual's health status from "physiological manifestations" to "overall functional state." By introducing a time-aware message passing mechanism, ST-GNN can capture the nonlinear temporal dependencies and spatial interaction patterns between different physiological systems, such as the lagging effect of blood sugar fluctuations on autonomic nervous function, or the metabolic disorder transmission pathway caused by long-term emotional stress through the HPA axis. This allows for the construction of a personalized health status evolution map with physiological interpretability. The multi-hop inference mechanism in L2 organ dysfunction early warning is based on the principle of using a pre-constructed personalized health status evolution map for high-order semantic deduction. This means that it not only focuses on the abnormal propagation between directly adjacent nodes (such as hyperglycemia leading to elevated glycated hemoglobin), but also tracks the risk diffusion chain across systems and time through multi-layer message aggregation (such as chronic insomnia causing cortisol rhythm disorder, which in turn leads to decreased insulin sensitivity, which in turn leads to liver fat deposition, which in turn leads to a slight increase in ALT). This multi-hop inference capability enables the model to identify the synergistic aggregation of multiple weak abnormal signals in the map when traditional medical test indicators are still within the normal reference range, thereby determining a subclinical health state. For example, when the three nodes "increased nighttime awakenings", "decreased morning HRV", and "fasting blood glucose approaching the upper limit" in the map simultaneously show marginal shifts and form a closed-loop feedback pathway, the system can trigger an early metabolic disorder warning. Personalized health status evolution maps and the hidden risk transmission paths they reveal further enrich the optimization process of personal dynamic health models. On the one hand, the high-dimensional embedding vectors output by the personalized health status evolution maps are used as inputs for health status representation, replacing traditional manually designed feature engineering and significantly improving the model's accuracy in modeling complex health degradation processes. On the other hand, key risk transmission subgraphs (such as the path of "prolonged sitting leading to slowed venous return in the lower limbs, which in turn leads to the accumulation of microinflammatory factors, which in turn leads to endothelial dysfunction") are extracted through the interpretability analysis module, providing mechanistic evidence for the generation of intervention strategies. In addition, the map supports incremental updates, automatically correcting the node state distribution and edge weights after each new data injection, ensuring that the personal dynamic health model always reflects the latest physiological evolution trends, thereby achieving a fundamental shift from static risk scoring to dynamic trajectory prediction.

[0040] The beneficial effects of the above technical solution are as follows: the solution provided in this embodiment can significantly improve the accuracy and foresight of health risk monitoring and early warning for people with chronic diseases; by constructing a personalized health status evolution map, it breaks through the limitations of traditional monitoring methods that rely on single indicators or static data; the application of the multi-hop inference mechanism in the early warning of L2-level organ dysfunction enables the early identification of subclinical health status; the continuous optimization of the personalized health status evolution map and the implicit risk transmission path for the personal dynamic health model not only improves the prediction accuracy of the model, but also provides physiologically interpretable evidence for the risk warning results, enabling the early warning to trace the deep-seated causes of the risk and providing strong support for the formulation of personalized intervention strategies.

[0041] In one embodiment, in step S3, the triggering of the L3 level chronic disease complication risk warning requires the simultaneous fulfillment of the following three conditions: the target organ function index shows a clinically defined abnormal value, the TCM syndrome of the chronic disease patient matches the preset high-risk syndrome, and the appearance of a pattern in the lifestyle data that is significantly associated with the risk of chronic disease complications.

[0042] The working principle of the above technical solution is as follows: First, the system dynamically monitors the target organ function status of patients with chronic diseases by accessing electronic health records, medical test reports, and physiological parameters collected in real time by wearable devices. When a certain key indicator (such as glomerular filtration rate eGFR < 60 mL / min / 1.73 m) is detected, the system will monitor the target organ function. 2 When the left ventricular ejection fraction (LVEF) < 50% or the retinopathy grade reaches stage III or above falls into the functional abnormality range defined by clinical guidelines, the system determines that the condition of "abnormal target organ function index" is met. This judgment is based on authoritative medical standards (such as KDIGO chronic kidney disease staging, ACC / AHA cardiac function classification, etc.) to ensure that the warning has clear clinical basis and interpretability. Secondly, at the level of TCM syndrome identification, the system relies on the structured consultation module and natural language processing technology to extract the patient's chief complaint, pulse and constitution identification information, and map them to the standardized TCM syndrome system (such as "Qi and Yin deficiency with blood stasis syndrome" and "Liver and Kidney Yin deficiency with wind syndrome" in "Internal Medicine of Traditional Chinese Medicine"). Through the pre-constructed "complication-high-risk syndrome" association knowledge graph (for example, the progression of diabetic nephropathy often corresponds to "spleen and kidney yang deficiency with water overflow syndrome", and patients with high risk of stroke recurrence often see "phlegm and blood stasis syndrome with liver yang hyperactivity syndrome"), the system compares the current diagnosis results with the preset high-risk syndrome database. Only when the patient's TCM syndrome completely matches at least one verified syndrome closely related to the occurrence and development of a specific complication can the condition be activated, thereby realizing the cross-validation and fusion decision of TCM and Western medicine information. Finally, the system conducts long-term tracking and pattern mining of lifestyle data from channels such as smart bracelets, mobile apps, and home monitoring terminals, including but not limited to daily step fluctuation trends, nocturnal heart rate variability, sleep fragmentation, changes in dietary calories and nutritional structure, and medication adherence scores. Machine learning models (such as LSTM time series analysis and random forest classifiers) are used to identify abnormal behavioral patterns significantly associated with complication risk, such as prolonged sitting (average daily activity time < 30 minutes), non-dipping nocturnal blood pressure patterns, and regular missed doses of antihypertensive medication. These patterns have been validated through multi-center cohort studies, demonstrating statistical significance (p < 0.01) and clinical predictive value (AUC > 0.75). Only when such high-risk behavioral patterns are continuously detected and reach a preset threshold can the system confirm that the third condition is met. All three conditions must be met simultaneously to trigger an L3 level warning. This multi-verification mechanism effectively avoids false positive alarms caused by misjudgment from a single dimension, improving the specificity and clinical applicability of the warning. Once triggered, the system automatically generates a comprehensive report that includes Western medicine diagnostic evidence, traditional Chinese medicine syndrome differentiation analysis, behavioral risk factors, and personalized intervention suggestions, and pushes it to the doctor's workstation and the patient's end, initiating a multi-level linkage early intervention algorithm process.

[0043] The beneficial effects of the above technical solution are as follows: The solution provided in this embodiment improves the accuracy and clinical operability of risk warning for chronic disease complications. By integrating the three dimensions of Western medicine objective physiological indicators, traditional Chinese medicine syndrome differentiation system and dynamic behavioral data, a joint discrimination model with biological rationality and medical logic consistency is constructed. This not only enhances the sensitivity and specificity of high-risk state identification, but also realizes the synergistic interpretation and complementary verification of Chinese and Western medicine diagnosis and treatment information.

[0044] In one embodiment, the knowledge base of Chinese and Western medicine intervention rules stores the mapping relationship between risk patterns and intervention measures. The intervention measures are matched with the hidden risk transmission paths identified in the personalized health status evolution map to generate intervention plans targeting the hidden risk transmission paths.

[0045] The working principle of the above technical solution is as follows: The Traditional Chinese Medicine (TCM) and Western Medicine Intervention Rule Knowledge Base is constructed based on a large amount of clinical data, medical guidelines and expert experience. It systematically integrates the TCM syndrome differentiation and treatment system with the Western medicine pathophysiological model. The TCM and Western Medicine Intervention Rule Knowledge Base pre-stores a variety of typical risk patterns (such as "spleen deficiency and dampness combined with pre-metabolic syndrome" and "qi stagnation and blood stasis associated with activation of chronic inflammatory pathways"). Each risk pattern corresponds to a set of quantifiable abnormal biomarkers, symptom combinations, physical characteristics and behavioral factors. These risk patterns are automatically identified from individuals' long-term health monitoring records through multimodal data fusion technology, ensuring accurate capture of subclinical states and early functional disorders. The TCM and Western Medicine Intervention Rule Knowledge Base is organized using ontology modeling, supports semantic reasoning and dynamic updates, and enables the intervention logic under different medical systems to be uniformly expressed and cross-validated. The system constructs a personalized health status evolution map based on time-series health data (including physical examination indicators, vital signs collected by wearable devices, information from the four diagnostic methods of traditional Chinese medicine, lifestyle logs, etc.). The map uses nodes to represent key health states (such as "insulin resistance", "liver qi stagnation", "gut flora imbalance") and edges to represent potential causal or related evolutionary paths between states. A temporal graph neural network is introduced to learn path weights. By comparing the deviation between the group baseline model and the individual trajectory, the system can identify abnormal evolutionary trends that deviate from the normal homeostasis, especially those intermediate state chains that do not constitute diseases under traditional diagnostic criteria but have a risk of continuous deterioration. These abnormal paths constitute "hidden risk transmission paths", such as the cascade reaction of endocrine disorders and emotional disorders caused by long-term sleep disorders leading to autonomic nervous system imbalance. When the system detects an individual's atypical but progressive health evolution path, it abstracts the path into a structured representation (such as state transition sequences, key driving factors, and rates of change of mediating variables) and performs semantic similarity matching with known risk patterns in the knowledge base of Chinese and Western medicine intervention rules. The matching process not only relies on symbolic reasoning (rule-based logical judgment) but also combines deep representation learning methods to embed the path into a low-dimensional vector space for nearest neighbor retrieval, thereby achieving high generalization recognition in complex multi-factor co-occurrence situations. For some novel paths that are not fully matched, the system can use a transfer learning mechanism to refer to similar patterns for analogy inference, thereby improving its ability to deal with rare or emerging health problems. After identifying the target risk transmission path, the system invokes the corresponding set of intervention measures from the knowledge base of traditional Chinese and Western medicine intervention rules. These measures cover a variety of methods, including recommendations for traditional Chinese medicine prescriptions (such as prescriptions for soothing the liver and regulating qi), pharmacological interventions in Western medicine (such as using metformin to improve insulin sensitivity), nutritional prescriptions (such as low glycemic load diets), exercise therapy (such as aerobic combined with resistance training), and psychological adjustment (such as mindfulness-based stress reduction). Based on the individual's current state, tolerance, compliance preferences, and factors such as the target point, onset period, and synergistic effect of each intervention measure, the system uses a multi-objective optimization algorithm to generate the optimal combination strategy. In particular, for key hub nodes in the transmission path (such as "chronic low-grade inflammation" as a common exit point for multiple systems), strong intervention measures are prioritized. For marginal perturbations, a mild adjustment approach is adopted to avoid overtreatment. The final output intervention plan has a clear time rhythm (with phased key adjustments), effect prediction curves, and risk warning thresholds, forming an executable, traceable, and iterative personalized health management plan.

[0046] The beneficial effects of the above technical solution are as follows: By using the solution provided in this embodiment, through the construction of an intervention rule knowledge base integrating traditional Chinese and Western medicine and a personalized health status evolution map, the accurate identification and dynamic intervention of the hidden risk transmission path of people with chronic diseases are realized.

[0047] In one embodiment, the intervention plan is pushed out using a dynamic priority ranking mechanism. The ranking criteria include warning level, risk urgency, the scenario in which the chronic disease patient is located, the first key criterion, and the second key criterion. A comprehensive priority score is calculated based on the weights assigned to each criterion to determine the push order. The first key criterion is the weight value of the abnormal physiological indicator that triggers the warning in the latent risk transmission path, and the second key criterion is the strength of the dynamic correlation between the abnormal physiological indicator that triggers the warning and other data in the multi-dimensional chronic disease characteristic time series data.

[0048] The working principle of the above technical solution is as follows: First, the system receives health monitoring data from people with chronic diseases in real time. When an abnormal physiological indicator triggers an early warning, a dynamic priority ranking mechanism is activated. At this time, the system extracts ranking criteria from multiple dimensions, including the warning level (e.g., high-risk, medium-risk, low-risk), the urgency of the risk (e.g., immediate intervention required, intervention within 24 hours, routine attention, etc.), and the current scenario of the patient with chronic diseases (e.g., daily life at home, hospital follow-up, during exercise, etc.). Simultaneously, it focuses on obtaining the first and second key criteria. The first key criteria determines the weight of the abnormal physiological indicator triggering the early warning in the path by querying the implicit risk transmission path model; a higher weight indicates a greater impact of the indicator on the overall risk transmission. The second key criteria are obtained by analyzing the abnormal physiological indicator... The dynamic correlation between the patient's multi-dimensional chronic disease characteristics and other data (such as past medical history, medication records, lifestyle habits, and trends in other physiological indicators) is calculated, and the correlation strength is determined. A higher correlation strength indicates that the abnormality of the indicator may trigger a more complex chain reaction. Next, according to a preset weighting system (the weights can be dynamically adjusted under different application scenarios; for example, the weight of risk urgency can be increased in emergency medical scenarios), each basis item is assigned a corresponding weight, and the specific value of each item is multiplied by its corresponding weight and then summed to obtain a comprehensive priority score. Finally, the intervention plans are sorted from high to low according to the comprehensive priority score, and the intervention plan with the highest score will be pushed to the patient or medical staff first to ensure that high-risk, high-urgency, and highly correlated intervention needs are responded to in a timely manner, thereby improving the accuracy and timeliness of the intervention.

[0049] The beneficial effects of the above technical solution are as follows: by adopting the solution provided in this embodiment, the intervention plan can be accurately sorted and pushed through a dynamic priority sorting mechanism, which can effectively avoid the problem of delay or omission of key intervention information that may occur in the traditional fixed order push mode.

[0050] Intelligent health risk monitoring and early warning platforms for people with chronic diseases, such as Figure 3 As shown, it includes: The intelligent terminal layer includes intelligent wearable devices for collecting physiological indicator data, as well as client applications for data entry, information viewing, and interaction. The platform service layer includes a data generation module, a risk assessment engine, and an intervention module. The data generation module is used for data collection, multivariate time series analysis, and generation of multi-dimensional chronic disease characteristic time series data that integrates spatiotemporal correlation features. The risk assessment engine is used to build a personal dynamic health model and implement a three-layer progressive risk warning logic. The intervention module is used to generate, push, and evaluate intervention plans. The platform service layer integrates edge computing and cloud collaborative processing architecture: the smart wearable device has a built-in L1 level early warning computing unit. After completing the initial calculation of L1 level trend deviation early warning, it only uploads abnormal data fragments and context time series information to the cloud for L2 and L3 level in-depth analysis; after receiving the uploaded data, the cloud analysis module calls the model knowledge base to execute the subsequent early warning logic. The model knowledge base, connected to the platform service layer, is used to store and maintain personal dynamic health models, organ health status models, pathological models, chronic disease-specific knowledge graphs, TCM syndrome evolution models, chronic disease complication pathological model libraries, and TCM and Western medicine intervention rule knowledge bases. The application interaction layer is used to provide users, their families, and doctors with differentiated health information displays, early warning notifications, and interactive functions.

[0051] The working principle of the above technical solution is as follows: The intelligent terminal layer collects physiological indicator data of people with chronic diseases in real time through intelligent wearable devices, such as heart rate, blood pressure, blood oxygen saturation, body temperature, respiratory rate, blood glucose (continuous monitoring), sleep quality, body movement and other multi-dimensional physiological parameters; these devices have high-precision sensors and low-power communication capabilities, support long-term continuous monitoring, and transmit the raw data to the client application via Bluetooth or Wi-Fi. The client application not only serves as the entry point for users to manually supplement data (such as diet, medication, symptom self-assessment), but also undertakes information visualization, early warning display and human-computer interaction functions, so as to realize the dynamic perception of users' own health status; The platform service layer adopts an edge-cloud collaborative computing architecture to ensure a balance between response efficiency and computational depth. In the data generation module, step S1 is executed first: the raw physiological data from the terminal is cleaned, denoised, standardized, and time-aligned to eliminate data anomalies caused by device errors or signal interference. Subsequently, a spatiotemporal correlation modeling mechanism is introduced, combining the time-series characteristics of individual historical data with the spatial distribution characteristics of common group patterns to construct a multi-dimensional chronic disease feature time-series dataset that integrates context-dependent data. This dataset not only contains the change trajectory of a single indicator but also integrates the coupling relationship between cross indicators (such as the dynamic correlation between blood pressure fluctuations and heart rate variability), providing a high-dimensional, structured input foundation for subsequent risk assessment. The risk assessment engine receives the processed feature time-series data and executes steps S2 and S3. In step S2, based on the individual dynamic health model in the model knowledge base, the system establishes an individualized baseline health status profile. This model is based on long-term monitoring data and uses an adaptive learning algorithm to continuously update each person's physiological homeostasis range and identify trends deviating from the normal fluctuation range. Based on this, step S3 implements a three-layer progressive risk warning logic: Level 1 warning is completed on the smart terminal side, using a lightweight model (such as moving sliding window threshold detection or simple LSTM) to determine in real time whether there is a significant trend deviation (such as cardiac...). If the rate continues to rise beyond the individual baseline upper limit by 2 standard deviations, a local alarm will be triggered and the abnormal segment and its preceding and following temporal context will be uploaded; after receiving the uploaded data in the cloud, the L2 warning will combine multivariate joint analysis and organ health status model to assess whether the current abnormality constitutes a specific organ dysfunction (such as cardiac overload or autonomic nervous system disorder) and determine its degree of matching with known pathological patterns; the L3 warning will further integrate chronic disease knowledge graph and complication pathological model library to perform cross-system correlation reasoning, and finally output a hierarchical, interpretable risk level (low / medium / high risk) and explanation of potential clinical significance. The intervention module is responsible for executing step S4, automatically generating a personalized intervention plan based on the risk assessment results. This process calls upon the knowledge base of intervention rules in both traditional Chinese and Western medicine, integrating modern medical guidelines (such as ACC / AHA recommendations for hypertension management) with the principles of syndrome differentiation and treatment in traditional Chinese medicine (such as the method of calming the liver and suppressing yang for "liver yang hyperactivity") to form a multimodal intervention strategy that includes lifestyle adjustments (exercise prescriptions, dietary recommendations), medication reminders and optimization, emotional regulation guidance, and traditional Chinese medicine conditioning suggestions (such as acupoint massage and herbal tea recommendations). The plan is pushed to users through the client and simultaneously synchronized to family members' terminals or doctors' workstations according to their roles and permissions. The system continuously tracks the implementation of the intervention and subsequent changes in physiological indicators, uses causal inference methods (counterfactual modeling) to evaluate the intervention effect, and dynamically iterates and optimizes subsequent suggestions to achieve closed-loop management. The model knowledge base serves as the cognitive foundation of the entire platform, centrally storing and managing all core models and rule systems. Among them, the personal dynamic health model evolves continuously over time, reflecting the physiological remodeling brought about by individual aging, disease progression, or intervention responses; the organ health status model depicts the functional degradation pathways of key systems (cardiovascular, metabolic, respiratory, etc.); the chronic disease-specific knowledge graph integrates structured medical knowledge such as disease staging, risk factors, and complication chains; and the TCM syndrome evolution model is based on the "syndrome-disease-time" three-dimensional framework, describing the transformation rules of syndrome types. All models support version control, incremental updates, and cross-institutional collaborative evolution under federated learning, ensuring the scientific nature and timeliness of the knowledge system. At the application interaction level, differentiated service interfaces are provided for different user roles. Ordinary users can view health overviews, receive alerts, perform intervention tasks, and provide feedback on their subjective feelings via mobile devices. The family member interface focuses on remote monitoring views and emergency event reminders to enhance family care support capabilities. The doctor's interface provides professional-grade data analysis tools, including trend comparison, risk tracing, and intervention effect review functions, to assist clinical decision-making. All interactive behaviors are recorded and fed back to the platform service layer for optimizing model training and user experience design.

[0052] The beneficial effects of the above technical solution are as follows: The solution provided in this embodiment enables precise full-cycle management of physiological indicator data for people with chronic diseases through a collaborative architecture of the intelligent terminal layer, platform service layer, model knowledge base, and application interaction layer; the terminal layer, composed of intelligent wearable devices and client applications, ensures real-time collection of physiological parameters while also meeting the user's needs for manual data entry and information interaction; the edge-cloud collaborative computing architecture of the platform service layer is highly innovative, completing preliminary L1-level early warning calculations on the device side and only uploading abnormal data to the cloud for in-depth analysis. This design significantly reduces data transmission pressure. This system improves the speed of early warning response while ensuring the depth and accuracy of L2 and L3 level early warning analysis. The risk assessment engine constructs a personal dynamic health model and a three-layer progressive risk early warning logic, enabling precise judgment from simple trend deviations to organ dysfunction and then to the risk of complications, making early warnings more hierarchical and clinically instructive. The intervention module generates multimodal intervention plans combining traditional Chinese and Western medicine, improving the personalization and effectiveness of interventions. The application interaction layer provides differentiated services for users, families, and doctors, meeting the information needs and interactive functions of different roles and enhancing family care support and clinical decision-making assistance capabilities.

[0053] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for intelligent monitoring and early warning of health risks for people with chronic diseases, characterized in that: Includes the following steps: S1: Collect physiological indicator data of patients with chronic diseases through smart wearable devices, and extract key features closely related to the target chronic disease from the physiological indicator data to form a physiological indicator feature dataset; at the same time, collect lifestyle habit data and TCM constitution identification information actively reported by patients with chronic diseases through mobile terminals or health management platforms; perform time-series alignment and structuring processing on the physiological indicator feature dataset, lifestyle habit data and TCM constitution identification information, and based on multivariate time-series analysis, mine the dynamic correlation between different modal data to generate multidimensional chronic disease feature time-series data that integrates spatiotemporal correlation features; S2: Based on multi-dimensional chronic disease feature time series data with spatiotemporal correlation characteristics, call and coordinate chronic disease-specific knowledge graphs, TCM syndrome evolution models, organ health status models and pathological models to construct a personal dynamic health model, and perform parameter learning and structure optimization based on continuous monitoring data and feedback information to achieve dynamic updates of the personal dynamic health model; S3: Based on a personal dynamic health model, a three-tiered progressive risk warning triggering mechanism is established, specifically including: L1-level trend deviation warning: Based on the key features extracted in S1, a dynamic baseline range based on personal historical data is maintained. When the key features continuously deviate from the personal dynamic baseline range and meet the preset statistical significance conditions, a trend deviation warning is triggered. L2-level organ dysfunction early warning: The feature data that triggers the L1-level trend deviation early warning is input into the organ health status sub-model optimized by chronic disease data to obtain the organ function status index and TCM syndrome attributes. When the organ function status index falls into the preset risk range, the organ function abnormality early warning is triggered. The organ health status sub-model is a sub-model of the personal dynamic health model. L3 chronic disease complication risk warning: Combining the dynamic correlation in S1, the output of L2 organ dysfunction warning is dynamically matched and jointly analyzed with the pathological model library of chronic disease complications. The probability of occurrence of specific chronic disease complications is calculated through the risk prediction model. When the probability of occurrence exceeds the preset threshold determined by clinical validation, a high risk warning for chronic disease complications is triggered. S4: Based on the three-tiered progressive risk warning, the system calls upon the knowledge base of Chinese and Western medicine intervention rules to generate a structured intervention plan and push it to the corresponding user terminal. At the same time, it collects feedback data after the intervention to optimize the personal dynamic health model and evaluate the intervention effect.

2. The intelligent health risk monitoring and early warning method for people with chronic diseases according to claim 1, characterized in that, Lifestyle data includes quantified dietary data and medication adherence data; TCM constitution identification information includes constitution type, syndrome evolution trajectory, and pulse characteristics data.

3. The intelligent health risk monitoring and early warning method for people with chronic diseases according to claim 1, characterized in that, In step S1, in addition to collecting physiological indicator feature datasets, lifestyle data, and TCM constitution identification information, the exercise status, sleep quality, and ambient light data of patients with chronic diseases are passively collected through the sensors of smart wearable devices and integrated into multivariate time series analysis to form lifestyle behavior data.

4. The intelligent health risk monitoring and early warning method for people with chronic diseases according to claim 1, characterized in that, The system dynamically updates individual health models using a hierarchical adaptive mechanism, setting differentiated model update frequencies and feature weight adjustment rules for patients with different chronic disease stages, age groups, or complication histories.

5. The intelligent health risk monitoring and early warning method for people with chronic diseases according to claim 1, characterized in that, The L1 level trend deviation warning uses a change point detection algorithm to identify changes in indicator trends; the L3 level chronic disease complication risk warning uses a machine learning model for risk prediction, and the chronic disease complication pathology model library includes prediction models for various acute complications of chronic diseases and target organ damage.

6. The intelligent health risk monitoring and early warning method for people with chronic diseases according to claim 1, characterized in that, Based on the generation of multi-dimensional chronic disease feature time-series data in step S1, the process also includes in-depth processing steps for this multi-dimensional chronic disease feature time-series data to optimize the accuracy of subsequent personal dynamic health model construction and risk warning. Specifically, this includes: A cross-modal association reasoning mechanism based on spatiotemporal graph neural network (ST-GNN) is introduced to construct a personalized health status evolution map. The personalized health status evolution map uses key physiological indicators as nodes and the evolution trajectory of TCM syndromes and dynamic changes in lifestyle habits as edge weights to capture the implicit risk transmission path in the three-element coupling relationship of physiology-behavior-constitution. In L2-level organ dysfunction early warning, multi-hop reasoning is used to identify subclinical health states that have not yet shown clinical symptoms, thus triggering subclinical risk warnings in advance.

7. The intelligent health risk monitoring and early warning method for people with chronic diseases according to claim 3, characterized in that, In step S3, the triggering of the L3 level chronic disease complication risk warning requires the simultaneous fulfillment of the following three conditions: the target organ function index shows a clinically defined abnormal value, the TCM syndrome of the chronic disease patient matches the preset high-risk syndrome, and the appearance of a pattern in the lifestyle data that is significantly associated with the risk of chronic disease complications.

8. The intelligent health risk monitoring and early warning method for people with chronic diseases according to claim 6, characterized in that, The knowledge base of Chinese and Western medicine intervention rules stores the mapping relationship between risk patterns and intervention measures. Intervention measures are matched with hidden risk transmission paths identified in the personalized health status evolution map to generate intervention plans targeting the hidden risk transmission paths.

9. The intelligent health risk monitoring and early warning method for people with chronic diseases according to claim 6, characterized in that, The intervention plan is delivered using a dynamic priority ranking mechanism. The ranking criteria include warning level, risk urgency, the scenario in which the patient with chronic disease is located, the first key criterion, and the second key criterion. The order of push notifications is determined by calculating a comprehensive priority score based on the weights assigned to each item. The first key criterion is the weight value of the abnormal physiological indicator that triggers the warning in the hidden risk transmission path, and the second key criterion is the strength of the dynamic correlation between the abnormal physiological indicator that triggers the warning and other data in the multi-dimensional chronic disease characteristic time series data.

10. A smart health risk monitoring and early warning platform for people with chronic diseases, characterized in that: include: The intelligent terminal layer includes intelligent wearable devices for collecting physiological indicator data, as well as client applications for data entry, information viewing, and interaction. The platform service layer includes a data generation module, a risk assessment engine, and an intervention module; The data generation module is used for data collection, multivariate time series analysis, and generation of multi-dimensional chronic disease characteristic time series data by integrating spatiotemporal correlation features; the risk assessment engine is used to realize the construction of personal dynamic health models and three-level progressive risk warning logic; the intervention module is used to realize the steps of intervention plan generation, push and effect evaluation. The platform service layer integrates edge computing and cloud collaborative processing architecture, supporting the preliminary calculation of L1 level trend deviation warning on the smart wearable device, and only uploading abnormal data fragments and context time series information to the cloud for L2 and L3 level in-depth analysis; The model knowledge base, connected to the platform service layer, is used to store and maintain personal dynamic health models, organ health status models, pathological models, chronic disease-specific knowledge graphs, TCM syndrome evolution models, chronic disease complication pathological model libraries, and TCM and Western medicine intervention rule knowledge bases. The application interaction layer is used to provide users, their families, and doctors with differentiated health information displays, early warning notifications, and interactive functions.

Images