Multi-level abnormal index early warning method and system

Abstract

The invention discloses a multi-level abnormal index early warning method and system, and the method comprises the steps: collecting the multi-source health data of a user, carrying out the standardization processing of the multi-source health data, and constructing a standardized physical sign feature matrix which comprises a plurality of health indexes; performing graph algorithm analysis on the sign feature matrix through a pre-constructed medical knowledge graph, and identifying an implicit association mode between different health indexes; based on the implicit association mode, combining a health data prediction sequence generated according to the historical multi-source health data and a user health portrait, and generating a study and judgment result of the user health risk; the early warning of the corresponding level is triggered according to the research and judgment result, and whether the early warning is upgraded or not is determined according to the response state of the user to the early warning, so that the accuracy and reliability of the early warning are remarkably improved, potential health abnormality can be timely identified in the early stage, and the health safety guarantee level of the user is improved.

Description

Technical Field

[0001] This application relates to the field of medical and health monitoring, specifically to a multi-level abnormal indicator early warning method and system. Background Technology

[0002] Currently, existing medical and health monitoring generally adopts a judgment mechanism based on fixed thresholds to identify abnormalities in various physiological indicators. However, this method has several significant drawbacks and is difficult to meet the requirements of accuracy, personalization, and responsiveness in practical applications.

[0003] First, the false alarm rate is high. Using uniform, fixed thresholds, existing health monitoring systems cannot distinguish data fluctuations caused by individual differences or normal physiological states (such as exercise or emotional fluctuations), easily misjudging normal physiological changes as abnormal events. At the same time, existing systems often rely on single-dimensional indicators for analysis, failing to comprehensively consider various aspects of user health information, resulting in one-sided and incomplete health risk assessment results.

[0004] Secondly, the early warning grading mechanism is simplistic. Most systems only set a single early warning level, using the same prompts or notifications regardless of whether the deviation from the indicator is slight or severe, lacking differentiated response strategies for different risk levels. Furthermore, their early warning rules and health recommendations are usually preset templates, lacking the ability to adapt dynamically to the user's individual characteristics and health status, making it difficult to achieve truly personalized health management.

[0005] Secondly, there is a lack of a post-warning processing mechanism. Warnings are simply communicated without confirmation, a one-way, feedback-free approach that may lead to unnecessary waste of medical resources and delays in intervention due to users' failure to notice the warning in a timely manner.

[0006] Finally, health monitoring involves data from multiple sources with varying formats, units, and semantic standards. Existing systems generally lack effective integration and standardization mechanisms, resulting in poor correlation between data and affecting the accuracy of health status assessments.

[0007] Therefore, existing health monitoring technologies have significant shortcomings in areas such as false alarm control, early warning classification, closed-loop response, and multi-source data collaboration. There is an urgent need for an intelligent health early warning solution that can achieve accurate, dynamic, multi-level, and closed-loop processing capabilities. Summary of the Invention

[0008] This application provides a multi-level abnormal indicator early warning method and system, which can solve the technical problems of high false alarm rate, inflexible classification and imperfect processing mechanism of existing health early warning methods.

[0009] In a first aspect, embodiments of this application provide a multi-level anomaly indicator early warning method, the multi-level anomaly indicator early warning method comprising: Collect multi-source health data from users, and standardize the multi-source health data to construct a standardized vital sign feature matrix, which contains multiple health indicators. By using a pre-constructed medical knowledge graph to analyze the vital sign feature matrix, implicit correlation patterns between different health indicators can be identified. Based on the implicit correlation pattern, combined with the health data prediction sequence generated from historical multi-source health data and the user health profile, the assessment result of the user's health risk is generated. The system will trigger a warning of the corresponding level based on the assessment results, and determine whether to escalate the warning based on the user's response to the warning.

[0010] Secondly, embodiments of this application provide a multi-level anomaly indicator early warning system, the multi-level anomaly indicator early warning system comprising: The data acquisition module is used to collect multi-source health data from users and to standardize the multi-source health data to construct a standardized vital sign feature matrix, which contains multiple health indicators. The analysis module is used to perform graph algorithm analysis on the vital sign feature matrix through a pre-constructed medical knowledge graph to identify implicit correlation patterns between different health indicators. The generation module is used to generate a judgment result on user health risk based on the implicit correlation pattern, combined with the health data prediction sequence generated from historical multi-source health data and the user health profile. The early warning module is used to trigger an early warning of the corresponding level based on the assessment results, and to determine whether to escalate the early warning based on the user's response to the early warning.

[0011] The beneficial effects of the technical solutions provided in this application include: By collecting multi-source health data from users and standardizing this data, a standardized vital sign feature matrix is ​​constructed, containing multiple health indicators. A graph algorithm analysis is performed on the vital sign feature matrix using a pre-constructed medical knowledge graph to identify implicit correlation patterns between different health indicators. Based on these implicit correlation patterns, combined with health data prediction sequences generated from historical multi-source health data and user health profiles, a judgment result on user health risk is generated. A corresponding level of warning is triggered based on the judgment result, and the decision to escalate the warning is determined based on the user's response to the warning. This approach solves problems in related technologies such as rigid warning rules, poor individual adaptability, difficulty in multi-source data fusion, difficulty in mining correlations between indicators, and the lack of a dynamic response mechanism based on user feedback. It achieves personalized, dynamic health risk warnings with closed-loop feedback capabilities, significantly improving the accuracy and reliability of warnings, timely identification of potential health abnormalities in the early stages, securing a critical time window for intervention, and effectively improving the level of user health and safety protection. Attached Figure Description

[0012] Figure 1 This is a flowchart illustrating an embodiment of the multi-level abnormal indicator early warning method of this application; Figure 2 This is a schematic diagram of the architecture of an embodiment of the multi-level abnormal indicator early warning system of this application. Detailed Implementation

[0013] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0014] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0015] Firstly, embodiments of this application provide a multi-level abnormal indicator early warning method.

[0016] In one embodiment, reference is made to Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the multi-level abnormal indicator early warning method of this application. Figure 1 As shown, the multi-level abnormal indicator early warning method includes: Step S101: Collect multi-source health data of users, and perform standardized processing on the multi-source health data to construct a standardized vital sign feature matrix, which contains multiple health indicators.

[0017] It is worth noting that the multi-level anomaly indicator early warning method in this embodiment can be implemented through a multi-level anomaly indicator early warning system. The specific implementation of the method of the present invention will be further described below in conjunction with this system.

[0018] In one embodiment, before executing step S101, the system pre-constructs a medical knowledge graph through a knowledge graph construction and entity relationship modeling module. This module is responsible for collecting and processing large amounts of health data, extracting general medical entities such as disease entities and health indicator entities, as well as relationship entities describing the relationships between entities, thereby constructing a three-layer medical knowledge graph that includes a foundation layer, a bridge layer, and an application layer.

[0019] The foundational layer is a core component of the medical knowledge graph, containing general medical entities and relationships. Specifically, the foundational layer includes, but is not limited to, the following types of entities: health indicator entities (e.g., heart rate, blood pressure, blood sugar); disease entities (e.g., hypertension, diabetes, coronary heart disease); symptom entities (e.g., headache, chest tightness, shortness of breath); and drug entities (e.g., aspirin, antihypertensive drugs, insulin). Each entity has a unique identifier and can be uniformly represented using standardized terminology to ensure semantic consistency across source data.

[0020] The bridge layer provides relationships between entities, describing semantic connections between different general medical entities. For example, disease-symptom associations: such as "hypertension → leads to → headache"; and also drug-target effects: such as "aspirin → acts on → platelet aggregation inhibition".

[0021] The application layer provides services for specific application scenarios, including but not limited to: personalized health management, generating personalized early warning rules based on user health profiles; and disease diagnosis and treatment recommendations, providing preliminary diagnostic opinions and treatment plans based on user health data and relationships in the knowledge graph. The application layer services rely on the knowledge base provided by the foundation and bridge layers, generating targeted health management and early warning strategies through the analysis of multi-source user health data.

[0022] It's worth noting that each medical entity and its relationship in the medical knowledge graph is assigned an initial weight value, ranging from 0.1 to 0.9. Specifically: 0.5 represents medium importance; 0.1 represents low importance; and 0.9 represents extremely high importance. This can be combined with an automatic learning model to achieve dynamic learning and optimization of the entity and relationship weights in the knowledge graph.

[0023] Specifically, the optimization methods include: Graph Neural Network (GNN) embedding learning, which uses GNNs to embed entities and relationships in the knowledge graph, capturing their deep semantic information; Temporal data reinforcement learning, which utilizes long-term accumulated user time-series health data and dynamically adjusts the weights of entities and relationships through reinforcement learning algorithms, enabling the knowledge graph to better adapt to individual differences and environmental changes; and Incremental learning, where the system automatically updates the entity and relationship weights in the knowledge graph when new data appears, maintaining the timeliness and accuracy of the graph.

[0024] In one embodiment, the multi-source health data in this application includes three categories: physiological indicator data, status description information, and environmental data.

[0025] Physiological data can be automatically collected through wearable devices (such as smart bracelets and biosensors) or professional medical instruments (such as electronic blood pressure monitors and Holter monitors). Physiological data includes, but is not limited to, electrocardiogram (ECG) data, heart rate data, and blood pressure data.

[0026] The status description information is provided by the user and is used to supplement subjective feelings and behavioral backgrounds that cannot be reflected by physiological data. The system obtains this information by setting up a daily data retrospective questionnaire. To improve compliance and data accuracy, the questionnaire triggering mechanism adopts an intelligent strategy, which includes: automatically pushing the questionnaire at a fixed time every day (such as 9:00 pm before bedtime), or triggering the questionnaire push in real time when the system detects potential abnormalities in physiological indicators.

[0027] The questionnaire design focuses on key factors that affect the interpretation of physiological indicators, including but not limited to: whether there was strenuous exercise, emotional fluctuations, alcohol consumption, or medication use today; current physical state (such as resting, walking, sleeping, working, etc.); and self-reported physical state ratings, such as quantitative scoring of fatigue, chest tightness, dizziness, discomfort, etc., from 0 to 10.

[0028] To ensure accurate alignment between status descriptions and physiological indicator data across time, the system associates questionnaires with specific time periods (e.g., "8:00 AM – 12:00 PM", "2:00 PM – 6:00 PM", "10:00 PM – 6:00 AM the next day"). User-submitted questionnaire answers are timestamped with physiological indicator data (e.g., ECG, heart rate, blood pressure) collected within the same time period, forming a fused and annotated dataset. This fused data not only helps build more accurate user health profiles but also serves as a personalized calibration basis for the "symptom-indicator" relationship in medical knowledge graphs, significantly improving the specificity and interpretability of anomaly identification.

[0029] In one embodiment, a timeliness weight is assigned to each collection of multi-source health data. The timeliness weight decreases as the interval between the collection time and the current time increases, with the most recently collected multi-source health data being assigned the highest timeliness weight. The timeliness weight ranges from 0.6 to 0.9, where a higher weight indicates fresher data and a lower weight indicates older data.

[0030] Furthermore, the method also includes: calculating weighted historical benchmark data based on the user's historical multi-source health data and corresponding timeliness weights; if the deviation between the latest collected multi-source health data and the weighted historical benchmark data is greater than a preset deviation threshold, an incremental learning mechanism is triggered to update the association relationships and corresponding weights between general medical entities in the medical knowledge graph.

[0031] It's worth noting that the system incorporates a dynamic adaptive mechanism to ensure the knowledge graph continuously reflects the latest changes in a user's individual health status. Specifically, the system calculates weighted historical baseline data based on the user's historical multi-source health data and its corresponding timeliness weights, serving as a reference baseline for the current health pattern. When the deviation between the latest collected multi-source health data and the weighted historical baseline data exceeds a preset deviation threshold, the system determines that the current data significantly deviates from the existing health pattern, potentially indicating a substantial change in physiological state. At this point, the incremental learning framework automatically initiates the incremental learning process without manual intervention or full model retraining. During this process, the system identifies health indicators that have changed significantly and traces their associated general medical entities in the medical knowledge graph. Subsequently, it dynamically updates the strength of the associations between these entities and their corresponding weights. Through this mechanism, the medical knowledge graph is no longer a static knowledge base but a dynamic reasoning engine with individual perception capabilities, capable of adaptively optimizing as the user's health status evolves, significantly improving the accuracy of subsequent graph algorithm analysis and the personalization of early warnings.

[0032] The deviation threshold can be set according to requirements; in this embodiment, it is set to 15%.

[0033] Furthermore, the multi-source health data is standardized to construct a standardized vital sign feature matrix, including: data cleaning and data integration of the multi-source health data; semantic unification of the integrated multi-source health data using a data standardization pipeline based on medical ontology and a pre-constructed medical terminology mapping matrix to obtain corresponding unified data; and data reduction and feature extraction of the unified data to construct the vital sign feature matrix.

[0034] As an example, the system first cleanses the raw, multi-source health data, removing outliers and invalid data. Next, the data integration module performs time alignment and field normalization on the data from different sources, ensuring consistency in time and format. This process aims to eliminate data inconsistencies caused by differences in equipment or acquisition environments, providing a reliable foundation for subsequent analysis.

[0035] After data cleaning and integration, the system utilizes a data standardization pipeline based on medical ontology and a pre-built medical terminology mapping matrix to perform semantic unification processing on the integrated multi-source health data. Specifically, the system first converts data of different formats into a unified format through data transformation and uses semantic similarity calculation (with a threshold set to 0.8) to ensure semantic consistency. Next, through the medical terminology mapping matrix, local codes (such as device proprietary codes) are mapped to standard medical terms (such as SNOMED CT and LOINC), thereby achieving standardized representation of cross-source data. This step ensures that data from different sources can be compared and analyzed within a unified knowledge framework.

[0036] After obtaining unified data, the system further performs data reduction and feature extraction to reduce data dimensionality and extract key features, ultimately constructing a standardized vital sign feature matrix. Specific steps include: Data reduction: The system uses a Kalman filter algorithm to preprocess the collected health and environmental data, removing outliers and noise interference. Subsequently, principal component analysis (PCA) or singular value decomposition (SVD) methods are used to reduce the dimensionality of the high-dimensional data, retaining the main variation features and reducing computational complexity. Feature extraction: The edge computing module is responsible for performing feature extraction operations. By normalizing the dimensionality-reduced data, the feature values ​​of each dimension fall between 0 and 1, ensuring the comparability and stability between different features. The normalized data is used to construct a standardized vital sign feature matrix, serving as the basis for subsequent graph algorithm analysis and risk assessment.

[0037] Through the standardized processing steps described above, the system can generate a high-quality, standardized vital sign matrix, providing a clean and reliable data foundation for subsequent knowledge-based processing. This standardized matrix not only includes the user's core physiological indicators but also integrates status description information and environmental data, forming a comprehensive dataset that fully reflects the user's health status.

[0038] Step S102: Analyze the physical characteristic matrix using a graph algorithm based on a pre-constructed medical knowledge graph to identify implicit correlation patterns between different health indicators.

[0039] Specifically, each dimension of the vital signs feature matrix corresponds to a standardized health indicator (such as heart rate, systolic blood pressure, blood oxygen saturation, etc.). The analysis module in the system maps these indicators to the corresponding health indicator entities in the basic layer of the medical knowledge graph. Subsequently, based on the predefined inter-entity relationships in the bridge layer, a subgraph with the current user's indicators as nodes is constructed on the graph, and graph algorithms are used to perform depth-first traversal and relational reasoning on this subgraph.

[0040] The graph algorithms employed include, but are not limited to, path search algorithms, subgraph matching algorithms, and graph neural network (GNN) message passing mechanisms. During this process, the system monitors the collaborative anomalies of multiple indicators in the semantic space of the knowledge graph. If two indicators, viewed individually, do not reach the warning threshold, but they have a strong correlation path in the graph, the system can identify this implicit correlation pattern and input it as a potential risk signal to the subsequent analysis module.

[0041] Step S103: Based on the implicit correlation pattern, and combined with the health data prediction sequence generated from historical multi-source health data and the user health profile, generate the assessment result of the user's health risk.

[0042] Specifically, the system's generation module first uses a pre-set time series prediction model (such as the ARIMA model) to generate a health data prediction sequence for the future time window based on the user's historical multi-source health data and the most recently collected multi-source health data. For example, the system can use heart rate, blood pressure, and blood oxygen saturation data from the past 30 days, combined with currently collected data, to predict the changing trends of these indicators over the next 7 days. This prediction sequence not only considers the trend changes of individual indicators but also incorporates environmental data (such as temperature and humidity) for comprehensive analysis to improve the accuracy of the prediction.

[0043] Based on users' historical multi-source health data, the system generates a raw health profile for each user. This profile includes multi-dimensional information such as trends in physiological indicators, lifestyle habits (e.g., exercise frequency, sleep quality), and environmental influencing factors. To further refine user classification, the system introduces a user profile clustering algorithm, dividing users into multiple typical health characteristic categories (six typical health characteristic categories in this embodiment). Each category represents a group of users with similar health characteristics, such as "high-risk individuals for hypertension" or "patients in the stable phase of cardiovascular disease." Through this clustering method, the system can more accurately identify the health characteristic group to which a user belongs, providing a basis for subsequent personalized warnings.

[0044] After obtaining the health data prediction sequence and user health profile, the system combines implicit correlation patterns to calibrate the user's health baseline. Specifically, the system dynamically adjusts the user's health baseline based on the health data prediction sequence, the user's health characteristic group, and implicit correlation patterns in the graph. For example, if a user belongs to the "high-risk group for hypertension" and their heart rate and blood pressure have recently shown a coordinated upward trend, the system will correspondingly raise the user's warning baseline and adaptively adjust the multi-level warning thresholds. The adjustment range is set at ±20% to ensure that the warning rules reflect individual differences while maintaining sufficient sensitivity.

[0045] Finally, the system compares the current physiological indicator data with the adjusted multi-level warning thresholds, and determines the assessment result of the user's health risk based on the level of warning thresholds exceeded by the physiological indicator data.

[0046] Step S104: Trigger the corresponding level of warning based on the judgment result, and determine whether to upgrade the warning based on the user's response to the warning.

[0047] For example, this can be achieved through the system's early warning module. If the user's current heart rate is higher than the first warning threshold after personalized calibration, the system triggers a level one warning and issues a voice prompt to the user. If the heart rate continues to exceed the second warning threshold, it automatically escalates to a level two warning and notifies the family. If the heart rate continues to exceed the third warning threshold, it automatically escalates to a level three warning and notifies relevant medical personnel to take further intervention measures.

[0048] If a valid response is received from the user within the preset response time (such as voice confirmation that there is no abnormality, manual clicking of "false alarm", or completion of a retest in a resting state, etc.), the system will maintain the current warning level or directly lift the warning; if no valid user response is received, the warning will be automatically upgraded to the next level and an upgrade notification will be pushed to the user terminal and relevant parties.

[0049] Through the above mechanisms, the system effectively improves the accuracy of abnormal event identification and the reliability of user interaction while ensuring the timeliness of early warnings.

[0050] In an optional embodiment, the method further includes: in response to a user's information query operation, displaying in a visualization interface: the user's multi-source health data, historical warning records, health risk assessment results, the changing trend of at least one health indicator under different time dimensions, and medical entities and their relationships in the medical knowledge graph.

[0051] Specifically, the visualization interface, provided by the system's interpretability and interaction design module, allows users to query historical trends of specific health indicators and compare changes in health status over different time periods. Simultaneously, the interface integrates interactive exploration functionality of a knowledge graph, allowing users to click on or expand medical entities (such as diseases, symptoms, and medications) to view their relationships and semantic paths within the graph. Furthermore, the system uses a hierarchical attention visualization module to intuitively present the contribution of each health indicator (represented by values ​​ranging from 0 to 1.0) and the confidence level of the adopted decision-making path during the AI ​​decision-making process. Paths with a confidence level ≥ 0.7 are marked as high-confidence paths, helping users understand the basis for the generation of warnings or judgments, thus improving the system's transparency and clinical acceptability.

[0052] In one optional embodiment, the system integration and deployment module adopts a cloud-edge collaborative architecture. Edge devices are responsible for data collection and preliminary processing, while the cloud platform is responsible for knowledge graph construction, AI algorithm training, and complex query processing. The system employs end-to-end encryption technology during data transmission and storage to ensure the security of user health data. Simultaneously, the system implements data anonymization and access control mechanisms, assigning different data access permissions based on different user roles. The system supports integration with third-party systems such as hospital information systems and electronic health record systems via API interfaces, enabling data sharing and functional complementarity.

[0053] In one specific embodiment, the system server is deployed in the health management center of a tertiary hospital and integrated with the hospital's smart bracelet management platform. Simultaneously, a corresponding app is installed on doctors' workstations, nurses' stations, and patients' family members' mobile phones to receive alert information.

[0054] Data Acquisition and Algorithm Training: 1000 patients with chronic cardiovascular disease were selected and equipped with smart bracelets. ECG, heart rate, and blood pressure data were collected continuously for one month. This data was imported into the system's AI algorithm processing module. Based on the TensorFlow framework, a Long Short-Term Memory (LSTM) network model was used for training to establish a personalized health indicator fluctuation model for each patient. The normal fluctuation range for heart rate was set at 60-100 beats / min, and the normal range for blood pressure was set at 90-140 / 60-90 mmHg.

[0055] Warning Rule Setting: In the warning rule setting module, set three levels of warning rules. Level 1 Warning: When the heart rate is 101-110 beats / min or blood pressure is 141-150 / 91-100 mmHg and lasts for 10 minutes, trigger voice confirmation; Level 2 Warning: When the heart rate is 111-120 beats / min or blood pressure is 151-160 / 101-110 mmHg and lasts for 5 minutes, directly notify family members; Level 3 Warning: When the heart rate is greater than 120 beats / min or blood pressure is greater than 160 / 110 mmHg, immediately notify the doctor and contact 120 (emergency services).

[0056] Actual scenario: While taking a walk, Patient A's smart bracelet detected a heart rate of 105 beats per minute for 12 minutes, triggering a Level 1 alert. The system then dialed Patient A's mobile phone via voice interaction, asking, "Are you currently feeling palpitations or discomfort?" Patient A did not answer the call. Three minutes later, the system determined that no confirmation had been received and activated a Level 2 alert, sending a text message to Patient A's family: "Patient's current heart rate is 105 beats per minute, alert triggered. Please contact the patient to confirm the situation." The family promptly contacted the patient, averting potential risks.

[0057] This application provides a multi-level abnormal indicator early warning method, the advantages of which are: by integrating physiological indicators, state descriptions, and environmental data to construct multi-source health data, and combining medical knowledge graphs to identify implicit correlation patterns between indicators, it avoids relying solely on a single threshold for judgment, significantly reducing false alarms and missed alarms. Based on timeliness weighting to calculate a weighted historical benchmark, combined with user health profiles and time series prediction models, the health baseline and multi-level early warning thresholds are adaptively adjusted within ±20%, making the early warning rules more closely aligned with individual health status. A user response feedback mechanism is introduced; if an early warning is not confirmed within a preset time, the system automatically upgrades the warning level and notifies relevant personnel, forming a closed-loop management of "monitoring—early warning—confirmation—intervention," improving emergency response efficiency. A cloud-edge collaborative architecture, end-to-end encryption, data anonymization, and role-based access control are adopted to ensure user privacy and security; simultaneously, through standard API interfaces, it can seamlessly connect with hospital HIS, electronic health record, and other systems, facilitating clinical implementation. A visual interface displays health trends, early warning records, knowledge graph correlations, and AI decision-making contributions, assisting medical staff in understanding the judgment basis and improving system credibility and clinical acceptance.

[0058] Secondly, embodiments of this application also provide a multi-level abnormal indicator early warning device.

[0059] In one embodiment, reference is made to Figure 2 , Figure 2 This is a functional module diagram of an embodiment of the multi-level abnormal indicator early warning device of this application. Figure 2 As shown, the multi-level abnormal indicator early warning device includes: The data acquisition module is used to collect multi-source health data from users and to standardize the multi-source health data to construct a standardized vital sign feature matrix, which contains multiple health indicators. The analysis module is used to perform graph algorithm analysis on the vital sign feature matrix through a pre-constructed medical knowledge graph to identify implicit correlation patterns between different health indicators. The generation module is used to generate a judgment result on user health risk based on the implicit correlation pattern, combined with the health data prediction sequence generated from historical multi-source health data and the user health profile. The early warning module is used to trigger an early warning of the corresponding level based on the assessment results, and to determine whether to escalate the early warning based on the user's response to the early warning.

[0060] Furthermore, in one embodiment, the medical knowledge graph includes: a base layer, a bridge layer, and an application layer; The base layer includes general medical entities, wherein the general medical entities include health indicator entities, disease entities, symptom entities and / or drug entities; The bridge layer includes the association relationships between the general medical entities, and each general medical entity and association relationship is configured with a corresponding initial weight; The application layer includes configuring corresponding early warning rules based on the user's health profile.

[0061] Furthermore, in one embodiment, the acquisition module is also used for: A timeliness weight is assigned to each collection of multi-source health data. The timeliness weight decreases as the time interval between the collection time and the current time increases, with the most recently collected multi-source health data being assigned the highest timeliness weight.

[0062] Furthermore, in one embodiment, the analysis module is also used for: Before performing graph algorithm analysis on the vital sign feature matrix using a pre-constructed medical knowledge graph, weighted historical baseline data is calculated based on the user's historical multi-source health data and corresponding timeliness weights; If the deviation between the newly collected multi-source health data and the weighted historical benchmark data is greater than a preset deviation threshold, an incremental learning mechanism is triggered to update the association relationships and corresponding weights between general medical entities in the medical knowledge graph.

[0063] Furthermore, in one embodiment, the acquisition module is also used for: The multi-source health data is cleaned and integrated. By employing a data standardization pipeline based on medical ontology and a pre-built medical terminology mapping matrix, semantic unification is performed on the integrated multi-source health data to obtain corresponding unified data; The unified data is reduced and features are extracted to construct the vital sign feature matrix.

[0064] Furthermore, in one embodiment, the generation module is further configured to: Using a pre-set time series prediction model, a health data prediction sequence for the future time window is generated based on the user's historical multi-source health data and the latest collected multi-source health data. A user health profile is generated based on the user's historical multi-source health data, and the user health profile is matched with multiple preset typical health characteristic categories to determine the health characteristic group to which the user belongs. Based on the predicted health data sequence, the user's health characteristic group, and the implicit correlation pattern, the user's health baseline is calibrated, and the multi-level early warning thresholds are adaptively adjusted based on the health baseline. The current physiological indicator data is compared with the adjusted multi-level warning thresholds, and the assessment result of the user's health risk is determined based on the warning threshold level exceeded by the physiological indicator data.

[0065] Furthermore, in one embodiment, the early warning module is also used for: If user confirmation or intervention is received within the preset response time after an alert is triggered, the current alert level will be maintained or the alert will be lifted. If no user response is received within the preset response time, the alert level will be automatically upgraded to the next level, and an upgrade notification will be sent.

[0066] Furthermore, in one embodiment, the system is also used for: In response to the user's information query operation, the following are displayed in the visualization interface: the user's multi-source health data, historical warning records, health risk assessment results, the changing trend of at least one health indicator in different time dimensions, and medical entities and their relationships in the medical knowledge graph.

[0067] Furthermore, in one embodiment, the multi-source health data includes: Physiological data, including electrocardiogram data, heart rate data, and / or blood pressure data; Status description information, including exercise status, mood fluctuations, medication status, sleep status, and / or the user's self-reported physical status score; Environmental data, including ambient temperature, humidity, and / or air quality.

[0068] The functions of each module in the above-mentioned multi-level abnormal indicator early warning device correspond to the steps in the above-mentioned multi-level abnormal indicator early warning method embodiment, and their functions and implementation processes will not be described in detail here.

[0069] It should be noted that the sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0070] The terms "comprising" and "having," and any variations thereof, in the specification, claims, and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such process, method, product, or apparatus. The terms "first," "second," and "third," etc., are used to distinguish different objects, etc., and do not indicate a sequence, nor do they limit "first," "second," and "third" to different types.

[0071] In the description of the embodiments of this application, terms such as "exemplary," "for example," or "for instance" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as "exemplary," "for example," or "for instance" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary," "for example," or "for instance" is intended to present the relevant concepts in a concrete manner.

[0072] In the description of the embodiments of this application, unless otherwise stated, " / " means "or". For example, A / B can mean A or B. The "and / or" in the text is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can mean: A exists alone, A and B exist simultaneously, and B exists alone. In addition, in the description of the embodiments of this application, "multiple" means two or more.

[0073] In some processes described in the embodiments of this application, multiple operations or steps are included in a specific order. However, it should be understood that these operations or steps may not be executed in the order they appear in the embodiments of this application, or they may be executed in parallel. The sequence number of the operation is only used to distinguish different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed sequentially or in parallel, and these operations or steps may be combined.

[0074] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device to execute the methods described in the various embodiments of this application.

[0075] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. A multi-level anomaly indicator early warning method, characterized in that, The multi-level anomaly indicator early warning method includes: Collect multi-source health data from users, and standardize the multi-source health data to construct a standardized vital sign feature matrix, which contains multiple health indicators. By using a pre-constructed medical knowledge graph to analyze the vital sign feature matrix, implicit correlation patterns between different health indicators can be identified. Based on the implicit correlation pattern, combined with the health data prediction sequence generated from historical multi-source health data and the user health profile, the assessment result of the user's health risk is generated. The system will trigger a warning of the corresponding level based on the assessment results, and determine whether to escalate the warning based on the user's response to the warning.

2. The multi-level abnormal indicator early warning method as described in claim 1, characterized in that, The medical knowledge graph includes: a basic layer, a bridge layer, and an application layer; The base layer includes general medical entities, wherein the general medical entities include health indicator entities, disease entities, symptom entities and / or drug entities; The bridge layer includes the association relationships between the general medical entities, and each general medical entity and association relationship is configured with a corresponding initial weight; The application layer includes configuring corresponding early warning rules based on the user's health profile.

3. The multi-level abnormal indicator early warning method as described in claim 2, characterized in that, The collection of users' multi-source health data also includes: A timeliness weight is assigned to each collection of multi-source health data. The timeliness weight decreases as the time interval between the collection time and the current time increases, with the most recently collected multi-source health data being assigned the highest timeliness weight.

4. The multi-level abnormal indicator early warning method as described in claim 3, characterized in that, Before performing graph algorithm analysis on the physical characteristic matrix using a pre-constructed medical knowledge graph, the following steps are also included: Based on the user's historical multi-source health data and corresponding timeliness weights, calculate weighted historical baseline data; If the deviation between the newly collected multi-source health data and the weighted historical benchmark data is greater than a preset deviation threshold, an incremental learning mechanism is triggered to update the association relationships and corresponding weights between general medical entities in the medical knowledge graph.

5. The multi-level abnormal indicator early warning method as described in claim 1, characterized in that, The multi-source health data is standardized to construct a standardized vital sign feature matrix, including: The multi-source health data is cleaned and integrated. By employing a data standardization pipeline based on medical ontology and a pre-built medical terminology mapping matrix, semantic unification is performed on the integrated multi-source health data to obtain corresponding unified data; The unified data is reduced and features are extracted to construct the vital sign feature matrix.

6. The multi-level abnormal indicator early warning method as described in claim 1, characterized in that, Based on the implicit correlation pattern, and combined with health data prediction sequences generated from historical multi-source health data and user health profiles, a user health risk assessment result is generated, including: Using a pre-set time series prediction model, a health data prediction sequence for the future time window is generated based on the user's historical multi-source health data and the latest collected multi-source health data. A user health profile is generated based on the user's historical multi-source health data, and the user health profile is matched with multiple preset typical health characteristic categories to determine the health characteristic group to which the user belongs. Based on the predicted health data sequence, the user's health characteristic group, and the implicit correlation pattern, the user's health baseline is calibrated, and the multi-level early warning thresholds are adaptively adjusted based on the health baseline. The current physiological indicator data is compared with the adjusted multi-level warning thresholds, and the assessment result of the user's health risk is determined based on the warning threshold level exceeded by the physiological indicator data.

7. The multi-level abnormal indicator early warning method as described in claim 1, characterized in that, Triggering corresponding alert levels based on the assessment results, and determining whether to escalate the alert based on the user's response to the alert, also includes: If user confirmation or intervention is received within the preset response time after an alert is triggered, the current alert level will be maintained or the alert will be lifted. If no user response is received within the preset response time, the alert level will be automatically upgraded to the next level, and an upgrade notification will be sent.

8. The multi-level abnormal indicator early warning method as described in claim 1, characterized in that, The method also includes: In response to the user's information query operation, the following are displayed in the visualization interface: the user's multi-source health data, historical warning records, health risk assessment results, the changing trend of at least one health indicator in different time dimensions, and medical entities and their relationships in the medical knowledge graph.

9. The multi-level abnormal indicator early warning method as described in claim 1, characterized in that, The multi-source health data includes: Physiological data, including electrocardiogram data, heart rate data, and / or blood pressure data; Status description information, including exercise status, mood fluctuations, medication status, sleep status, and / or the user's self-reported physical status score; Environmental data, including ambient temperature, humidity, and / or air quality.

10. A multi-level anomaly indicator early warning system, characterized in that, The multi-level abnormal indicator early warning system includes: The data acquisition module is used to collect multi-source health data from users and to standardize the multi-source health data to construct a standardized vital sign feature matrix, which contains multiple health indicators. The analysis module is used to perform graph algorithm analysis on the vital sign feature matrix through a pre-constructed medical knowledge graph to identify implicit correlation patterns between different health indicators. The generation module is used to generate a judgment result on user health risk based on the implicit correlation pattern, combined with the health data prediction sequence generated from historical multi-source health data and the user health profile. The early warning module is used to trigger an early warning of the corresponding level based on the assessment results, and to determine whether to escalate the early warning based on the user's response to the early warning.

Images