Health risk dynamic assessment and early warning method and system for old people

Abstract

The invention discloses a health risk dynamic assessment and early warning method and system for old people groups, and relates to the field of health risk assessment of old people groups, and the method comprises the steps: comparing first health detection data with a preset chronic disease diagnosis standard, and building a hierarchical initial health file according to a comparison result; acquiring symptom description information, treatment condition information and detection data fed back by the elderly individuals; extracting stage health features; constructing an old people health fault tree knowledge graph; calculating the probability of occurrence of the health endpoint event of the elderly individual, and identifying a key risk conduction path causing increase of the probability; generating differentiated early warning information and intervention suggestions; and optimizing and adjusting the logic gate weight in the fault tree knowledge graph. The method has the advantages that the senile health fault tree knowledge graph is constructed, expert knowledge in the field of senile medicine is converted into a computable risk conduction logic model, and interpretable early warning and accurate traceability of senile health risks are achieved.

Description

Technical Field

[0001] This invention relates to the field of health risk assessment for the elderly population, specifically to a method and system for dynamic health risk assessment and early warning for the elderly population. Background Technology

[0002] With the accelerating aging of the population, health management for the elderly has become a focus of social attention and an important issue for ensuring people's livelihood and improving the public health service system. Currently, health risk assessment technologies for the elderly mainly fall into two categories: one is continuous monitoring technology based on wearable devices, which uses portable devices such as smart bracelets and smartwatches to collect core physiological data in real time, such as heart rate, blood pressure, blood oxygen saturation, and steps, and combines this with preset physiological indicator thresholds to issue alarms for abnormal data; the other is static assessment technology based on electronic medical records, which systematically analyzes the individual's past medical records, laboratory test reports, medication history, disease diagnosis results, and other historical medical data to assess their current disease risk, potential disease risk, and disease prognosis. In addition, in recent years, some studies have attempted to introduce artificial intelligence algorithms (such as machine learning and deep learning) to construct risk prediction models that integrate multiple factors such as physiological indicators, medical data, and lifestyle habits, aiming to further improve the accuracy and relevance of health risk assessments.

[0003] However, the application of wearable devices among the elderly faces challenges such as insufficient age-friendly design, including common problems such as forgetting to wear them, cumbersome charging, and difficult operation, resulting in poor long-term user compliance. Secondly, assessment technologies based on medical records face the problem of data lag. The elderly often seek medical attention only after symptoms become obvious, which may mean missing the best intervention opportunity. Existing artificial intelligence prediction models are mostly "black box" models, which not only rely on massive amounts of lifestyle, physiological monitoring, and diagnosis and treatment data, but also only output risk probabilities without explaining the source of the risk, making it difficult to gain the trust of clinicians and elderly users. Summary of the Invention

[0004] To address the aforementioned technical issues, this paper provides a method and system for dynamic assessment and early warning of health risks for the elderly population. This technical solution resolves the problems mentioned in the background section regarding existing health risk assessment technologies for the elderly, such as insufficient age-friendly design of wearable devices, data lag, uninterpretable artificial intelligence models, and dependence on training sample data.

[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A dynamic health risk assessment and early warning method for the elderly population includes: Acquire the initial health test data of elderly individuals, compare the initial health test data with the preset chronic disease diagnostic criteria, and establish a hierarchical initial health record based on the comparison results; Based on the hierarchy of initial health records, information on symptoms, treatment, and test results is collected from elderly individuals through APP registration and regular testing. Based on symptom descriptions and treatment information provided by elderly individuals, phased health characteristics are extracted; A fault tree knowledge graph for elderly health is constructed. The fault tree knowledge graph has health endpoint events as top nodes, chronic diseases, acute symptoms and abnormal physiological indicators as intermediate nodes, and stage-specific health characteristics as bottom nodes. The nodes are connected by logic gates to form risk transmission paths. The current state of the base node is input into the fault tree knowledge graph as the phased health characteristics, and the probability of an elderly individual experiencing a health endpoint event is calculated, and the key risk transmission path that leads to the increase of this probability is identified. Based on the probability of health endpoint events and key risk transmission pathways, differentiated early warning information and intervention recommendations are generated; Obtain the actual health event results of elderly individuals and compare them with the probability of health endpoint events. Optimize and adjust the logic gate weights in the fault tree knowledge graph based on the comparison results.

[0006] Preferably, the establishment of hierarchical initial health records specifically includes: If an elderly individual's initial health check data does not identify a chronic disease, a natural disease monitoring file should be established. Natural disease surveillance records use the initial health check data as a reference baseline, record only basic physiological indicators, and are marked as level one. If an elderly individual is diagnosed with a chronic disease based on their initial health check data, obtain the physician's diagnostic data and treatment plan, and establish a chronic disease management file. Chronic disease management records include diagnosis information, medication regimens, and treatment goals; the record level is marked as level two. We continuously monitor and obtain medical diagnosis information from elderly individuals, and dynamically adjust the file level and supplement relevant data based on the feedback.

[0007] Preferably, the collection of symptom descriptions, treatment information, and test data from elderly individuals based on the initial health record hierarchy, through APP registration and regular testing, specifically includes: Differentiated collection cycles and dimensions are set based on the archive hierarchy; By registering an app, the app pushes voice inquiry content that matches the age level to elderly individuals, and simultaneously collects symptom descriptions and treatment information from the elderly individuals in response to the inquiries. Based on regular offline health check-ups, data on various physiological indicators and test results generated by the corresponding testing processes for elderly individuals are collected. The subjective feedback information collected by the APP is linked and integrated with the objective data obtained from regular testing, and then uniformly collected into the health records of the corresponding elderly individuals, which are stored as periodic health characteristic data.

[0008] Preferably, the extraction of phased health characteristics based on symptom descriptions and treatment information from elderly individuals specifically includes: For elderly individuals corresponding to Level 1 records, if new symptom keywords are reported, the symptom name, occurrence time, and duration are extracted as characteristics of the new symptom. For elderly individuals corresponding to Level 1 records, if they report no symptoms and no medical records in several consecutive inquiries, they are set to a continuous health status label. For elderly individuals corresponding to Level 2 records, the proportion of on-time medication responses within a preset time period is counted to extract characteristics of poor medication adherence. For elderly individuals corresponding to Level 2 records, the frequency and severity of various symptoms reported within a preset time period are statistically analyzed, the symptom fluctuation coefficient is calculated, and the characteristics of unstable symptom control are extracted. For elderly individuals corresponding to Level 2 records, when symptoms that have not been previously reported are recorded in several consecutive inquiry responses, and these symptoms match the symptoms in the list of known complications of the chronic disease, they are extracted as complication warning features.

[0009] Preferably, the construction of the elderly health fault tree knowledge graph specifically includes: Collect clinical guidelines, expert experience, and historical case data in geriatric medicine, and extract disease names, complication names, symptom names, and medication names as knowledge nodes; The top node is the elderly's disability, hospitalization, and death; the middle node is each chronic disease and its complications; and the bottom node is the stage-specific health characteristics. Determine the logical transmission relationships between each node, including AND gate relationships and OR gate relationships; Based on the top node, middle node, bottom node, and the logical transmission relationship between them, a knowledge graph of fault tree for elderly health is constructed.

[0010] Preferably, the calculation of the probability of an elderly individual experiencing a health endpoint event and the identification of key risk transmission pathways leading to an increased probability specifically includes: The current state of the base node is input into the fault tree knowledge graph as the phased health feature. If a certain symptom or behavior exists in the phased health feature, the state of the base node corresponding to the symptom or behavior is set to 1, otherwise it is set to 0. The upward method is used to calculate the probability of occurrence of each intermediate node by starting from the bottom node and proceeding upwards layer by layer according to the logic gate rules; Starting from the top node, the downward method is used to search for all activated propagation paths in the current state layer by layer according to the logic gate rules; The minimum cut set is output as the key risk transmission path, which is used to explain the main causes of the risk at the top node.

[0011] Preferably, the step of generating differentiated early warning information and intervention recommendations based on the probability of health endpoint events and key risk transmission pathways specifically includes: The risk level is determined based on the probability of a health endpoint event, and the risk level includes three levels: low risk, medium risk, and high risk. Based on the risk level and key risk transmission pathways, generate differentiated intervention recommendations: If the risk level is low, the existing frequency of periodic inquiries will be maintained without any additional intervention. If the risk level is medium and the key risk transmission path includes a medication adherence node, a medication reminder suggestion will be generated, and a follow-up reminder will be added in the next inquiry. If the risk level is medium and the key risk transmission path includes nodes of newly emerging symptoms, then a medical advice will be generated and follow-up questions will be added in the next inquiry. If the key risk transmission path of a medium-risk individual contains multiple types of nodes, then these will be merged to generate multiple corresponding intervention recommendations. If the risk level is high, an emergency warning message will be generated and pushed to the APP or consult a doctor. The warning message includes the risk level, key risk transmission path and recommended medical department.

[0012] Preferably, the optimization and adjustment of the logic gate weights in the fault tree knowledge graph based on the comparison results specifically includes: Acquire the actual health event outcomes of elderly individuals, including positive and negative events, and record the specific time and type of the events. The predicted probabilities of all elderly individuals within the historical prediction period are paired with the actual results to construct positive event sample sets and negative event sample sets respectively. Calculate the average predicted probability of positive event samples. If the value is lower than the first threshold, it is determined that there is a risk of missed reports in the graph, and the weight coefficients of each logic gate in the corresponding key risk transmission path are increased. Calculate the average predicted probability of negative event samples. If the value is higher than the second threshold, it is determined that the graph has a risk of false alarm, and the weight coefficient of each logic gate in the corresponding key risk transmission path is weakened. The optimized and adjusted fault tree knowledge graph will be used as the benchmark for the next round of risk simulation.

[0013] Furthermore, this solution proposes a dynamic health risk assessment and early warning system for the elderly population, used to implement the aforementioned dynamic health risk assessment and early warning method for the elderly population, including: The health record module is used to acquire the initial health test data of elderly individuals, compare the initial health test data with preset chronic disease diagnostic standards, and establish hierarchical initial health records based on the comparison results. The data acquisition module is used to collect symptom descriptions, treatment information, and test data from elderly individuals based on the initial health record hierarchy, through APP registration and regular testing. The health assessment and optimization module is used to extract phased health features based on symptom descriptions and treatment information provided by elderly individuals; construct an elderly health fault tree knowledge graph, with health endpoint events as top nodes, chronic diseases, acute symptoms, and abnormal physiological indicators as intermediate nodes, and phased health features as bottom nodes, connected by logic gates to form risk transmission paths; input the current state of the phased health features as bottom nodes into the fault tree knowledge graph to calculate the probability of an elderly individual experiencing a health endpoint event and identify key risk transmission paths that lead to an increase in this probability; generate differentiated early warning information and intervention suggestions based on the probability of health endpoint events and key risk transmission paths; obtain the actual health event results of elderly individuals and compare them with the probability of health endpoint events, optimizing and adjusting the weights of logic gates in the fault tree knowledge graph based on the comparison results.

[0014] Preferably, the health assessment and optimization module includes: The feature extraction unit is used to extract phased health features based on symptom descriptions and treatment information provided by elderly individuals. The graph construction unit is used to construct an elderly health fault tree knowledge graph. The fault tree knowledge graph has health endpoint events as top nodes, chronic diseases, acute symptoms and abnormal physiological indicators as intermediate nodes, and stage-specific health characteristics as bottom nodes. The nodes are connected by logic gates to form a risk transmission path. A health assessment unit is used to input the current state of the stage-based health characteristics as the base node into the fault tree knowledge graph, calculate the probability of an elderly individual experiencing a health endpoint event, and identify the key risk transmission path that leads to an increase in this probability. The early warning and recommendation unit is used to generate differentiated early warning information and intervention recommendations based on the probability of health endpoint events and key risk transmission paths. The optimization and adjustment unit is used to obtain the actual health event results of elderly individuals, compare them with the probability of health endpoint events, and optimize and adjust the logic gate weights in the fault tree knowledge graph based on the comparison results.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention proposes a dynamic health risk assessment and early warning scheme for the elderly population. Through a hierarchical filing mechanism combining initial health checks with post-medical feedback, it categorizes elderly individuals into a natural disease monitoring layer and a chronic disease management layer. Dynamic migration of file levels is achieved based on feedback information, solving the data cold-start problem in existing technologies where a large number of elderly individuals who have not sought medical treatment are excluded from the monitoring scope. By using a registered app to push differentiated periodic voice inquiries and regular checks, symptom feedback and medication information are collected in a manner consistent with clinical diagnostic logic, effectively overcoming the shortcomings of wearable devices in terms of age-friendly design and improving the adaptability of the elderly population. Furthermore, by constructing a fault tree knowledge graph with health endpoint events as top nodes, expert knowledge and clinical guidelines in the field of geriatric medicine are transformed into a computable risk transmission logic. This model reveals the mechanistic relationship between multiple coexisting diseases, solving the problem that existing artificial intelligence models cannot explain the source of risk. By inputting stage-specific health characteristics into a fault tree knowledge graph for upward probability calculation and downward path search, it not only calculates the probability of risk occurrence but also identifies the key transmission paths leading to the risk, making the warning results interpretable and providing clear intervention directions for users and clinicians. Through the comparison and feedback between actual health event results and predicted probabilities, the logic gate weights in the fault tree knowledge graph are continuously optimized, enabling dynamic evolution and personalized adaptation of the model. This achieves individualized profiling, communication-friendly data collection, multi-disease risk transmission analysis, interpretable warnings, and closed-loop model evolution, effectively improving the accuracy and clinical applicability of risk assessment. Attached Figure Description

[0016] Figure 1 This is a flowchart of a dynamic health risk assessment and early warning method for the elderly population according to the present invention. Figure 2 The flowchart for extracting stage-specific health characteristics based on symptom description information and treatment information from elderly individuals in this invention is as follows: Figure 3 This is a flowchart illustrating the construction of a fault tree knowledge graph for elderly health in this invention. Figure 4 This invention provides a flowchart for calculating the probability of health endpoint events in elderly individuals and identifying key risk transmission pathways that lead to an increase in this probability. Detailed Implementation

[0017] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0018] Reference Figure 1 As shown, a dynamic health risk assessment and early warning method for the elderly population includes: Acquire the initial health test data of elderly individuals, compare the initial health test data with the preset chronic disease diagnostic criteria, and establish a hierarchical initial health record based on the comparison results; Based on the hierarchy of initial health records, information on symptoms, treatment, and test results is collected from elderly individuals through APP registration and regular testing. Based on symptom descriptions and treatment information provided by elderly individuals, phased health characteristics are extracted; A fault tree knowledge graph for elderly health is constructed. The fault tree knowledge graph has health endpoint events as top nodes, chronic diseases, acute symptoms and abnormal physiological indicators as intermediate nodes, and stage-specific health characteristics as bottom nodes. The nodes are connected by logic gates to form risk transmission paths. The current state of the base node is input into the fault tree knowledge graph as the phased health characteristics, and the probability of an elderly individual experiencing a health endpoint event is calculated, and the key risk transmission path that leads to the increase of this probability is identified. Based on the probability of health endpoint events and key risk transmission pathways, differentiated early warning information and intervention recommendations are generated; Obtain the actual health event results of elderly individuals and compare them with the probability of health endpoint events. Optimize and adjust the logic gate weights in the fault tree knowledge graph based on the comparison results.

[0019] Understandably, this solution establishes hierarchical initial health records by comparing the initial health check with preset chronic disease diagnostic standards, enabling differentiated record-keeping for elderly individuals at the natural disease monitoring and chronic disease management levels. Subsequent medical feedback facilitates dynamic migration of record levels. The solution collects symptom feedback, medication usage, and test data through periodic voice inquiries and regular testing via a registered app, simulating clinical consultation logic and effectively overcoming the limitations of wearable devices' age-friendly design, thus improving the adaptability of the elderly population. By constructing a fault tree knowledge graph with health endpoint events as top nodes and stage-specific health characteristics as bottom nodes, expert knowledge in geriatric medicine is transformed into a computable risk transmission logic model, revealing the mechanism of coupling relationships in the coexistence of multiple diseases. By inputting stage-specific health characteristics into the fault tree knowledge graph for upward probability calculation and downward path search, not only is the probability of risk occurrence calculated, but the key transmission paths leading to increased risk are also identified, making the warning results interpretable. Through comparison and feedback between actual health event outcomes and predicted probabilities, the weights of logic gates in the fault tree knowledge graph are continuously optimized, achieving dynamic evolution and personalized adaptation of the model.

[0020] The establishment of hierarchical initial health records specifically includes: If an elderly individual's initial health check data does not identify a chronic disease, a natural disease monitoring file should be established. Natural disease surveillance records use the initial health check data as a reference baseline, record only basic physiological indicators, and are marked as level one. If an elderly individual is diagnosed with a chronic disease based on their initial health check data, obtain the physician's diagnostic data and treatment plan, and establish a chronic disease management file. Chronic disease management records include diagnosis information, medication regimens, and treatment goals; the record level is marked as level two. We continuously monitor and obtain medical diagnosis information from elderly individuals, and dynamically adjust the file level and supplement relevant data based on the feedback.

[0021] It is understandable that the health risks of the elderly population mainly include natural diseases, chronic diseases, and extensions of chronic diseases, as well as the types of initial treatment and existing treatment records. There are a large number of individuals in the elderly population with health risks who have not yet sought medical treatment. If modeling is based solely on medical data, these individuals will be excluded from the monitoring scope. This solution establishes a baseline profile through initial testing and includes individuals who have not sought medical treatment in a primary profile for continuous monitoring. When an individual in the primary profile reports medical diagnosis information and this information is identified as a confirmed diagnosis of a chronic disease, the individual's profile level is moved from primary to secondary, and the medical diagnosis information is added to the profile as a new dimension. If it is an acute disease or other non-chronic disease diagnosis, the medical information is only recorded in the primary profile without moving the profile level, thus achieving comprehensive coverage of the monitoring scope and dynamic evolution of the profile level. It should be explained that for individuals whose initial health check can detect some chronic diseases and can be directly identified as chronic diseases (such as hypertension and diabetes), they will be directly admitted to the secondary file. For individuals whose initial health check cannot confirm a diagnosis and require further examination (such as early-stage kidney disease and neurodegenerative diseases), or who are aware of their condition but did not provide medical history information during the initial check, they will first be admitted to the primary file for continuous monitoring. Once they provide confirmation information through the app or consultation, they will be transferred to the secondary file.

[0022] The hierarchical structure based on initial health records, through APP registration and regular testing, collects symptom descriptions, treatment information, and test data from elderly individuals, specifically including: Differentiated collection cycles and dimensions are set based on the archive hierarchy; By registering an app, the app pushes voice inquiry content that matches the age level to elderly individuals, and simultaneously collects symptom descriptions and treatment information from the elderly individuals in response to the inquiries. Based on regular offline health check-ups, data on various physiological indicators and test results generated by the corresponding testing processes for elderly individuals are collected. The subjective feedback information collected by the APP is linked and integrated with the objective data obtained from regular testing, and then uniformly collected into the health records of the corresponding elderly individuals, which are stored as periodic health characteristic data.

[0023] It is understandable that collecting symptom descriptions and treatment information from elderly individuals via an app, along with collecting physiological indicators and test results from regular offline health checkups, are crucial data sources for understanding the health status of elderly individuals. This approach avoids age-appropriateness issues in data collection from the outset, ensuring the continuity and completeness of health data. Therefore, this solution sets differentiated cycles and collection dimensions based on the hierarchical structure of health records. A specific implementation example is as follows: For elderly individuals corresponding to Level 1 records, a voice inquiry is periodically pushed through the registered APP. The inquiry content is "Have you experienced any new discomfort symptoms in the past month? For example, dizziness, chest tightness, fatigue, cough? If so, how many days has it lasted?" The voice responses of the elderly individuals are collected, and natural language processing technology is used to analyze the response content and extract symptom keywords and time descriptive words as symptom description information. For elderly individuals corresponding to Level 1 records, a home visit or community-based health check will be arranged regularly to collect basic physiological parameters such as blood pressure, blood sugar, and heart rate as test data. The test data will be compared with the initial health baseline. When the test data deviates from the baseline by more than the normal value, the individual will be marked as a state requiring attention and targeted follow-up questions will be added in subsequent inquiries. For elderly individuals with Level 2 medical records, a voice inquiry is pushed weekly through the registered app. The inquiry content is dynamically generated based on the specific chronic disease type recorded in the record. For example, the inquiry for a hypertensive patient is "Have you taken your antihypertensive medication on time in the past week? Have you experienced dizziness or headaches? What was your blood pressure measured this morning?" The inquiry for a diabetic patient is "Have you taken your medication or injected insulin on time in the past week? Have you experienced palpitations, cold sweats, or increased thirst?" The inquiry for a coronary heart disease patient is "Have you experienced chest tightness or chest pain in the past week? Does it worsen after activity? Have you taken your medication on time?" The voice responses of elderly individuals are collected, and symptom keywords, medication behavior keywords, and self-test data are parsed as symptom description information and treatment information. For elderly individuals with Level II medical records, according to the doctor's orders for regular testing, testing reminders are pushed through the APP to guide them to go to community medical institutions or use home testing devices to complete the testing. The test data is collected and automatically uploaded through the APP or manually entered. The test data is compared with the treatment target range to generate a target achievement analysis. When an elderly individual proactively reports their medical experience through the APP, a guided inquiry process is triggered, which sequentially collects information such as the department visited, diagnosis results, medication adjustments, and follow-up recommendations. The collected medical information is then added to the health record and the record level is updated.

[0024] Reference Figure 2 As shown, the extraction of phased health characteristics based on symptom descriptions and treatment information from elderly individuals specifically includes: For elderly individuals corresponding to Level 1 records, if new symptom keywords are reported, the symptom name, occurrence time, and duration are extracted as characteristics of the new symptom. For elderly individuals corresponding to Level 1 records, if they report no symptoms and no medical records in several consecutive inquiries, they are set to a continuous health status label. For elderly individuals corresponding to Level 2 records, the proportion of on-time medication responses within a preset time period is counted to extract characteristics of poor medication adherence. For elderly individuals corresponding to Level 2 records, the frequency and severity of various symptoms reported within a preset time period are statistically analyzed, the symptom fluctuation coefficient is calculated, and the characteristics of unstable symptom control are extracted. For elderly individuals corresponding to Level 2 records, when symptoms that have not been previously reported are recorded in several consecutive inquiry responses, and these symptoms match the symptoms in the list of known complications of the chronic disease, they are extracted as complication warning features.

[0025] Understandably, natural illnesses are often assumed to be cured without medical attention. Therefore, for individuals in Level 1 records, as long as they do not report symptoms, they are considered healthy and require no intervention. Chronic diseases, on the other hand, are characterized by experiential symptoms, and patients have a direct understanding of their own symptoms. Collecting these subjective feelings through regular inquiries is entirely consistent with the logic of physicians obtaining information through consultations. By combining the health management needs of elderly individuals at different levels of their records, differentiated stage-specific health characteristics can be extracted from their symptom descriptions and treatment information, providing accurate and structured core data support for subsequent health risk projection. It should be noted that the symptom fluctuation coefficient is an indicator used to quantify the variation in the frequency and severity of symptoms in elderly individuals over a specific period of time, and its calculation method is as follows: The system retrieves all symptom records submitted by elderly individuals via the app within a preset time period. These symptom records include the symptom name, date of occurrence, and severity level. The preset time period is divided into several time segments, and the total number of times the symptoms occur and the severity score of each symptom are counted in each time segment. Calculate the symptom burden index for each time segment, where the symptom burden index is the sum of all symptom severity scores within that segment; The mean of the symptom load index for all time segments is calculated as the baseline load value; Calculate the sum of the absolute values ​​of the deviations between the symptom load index and the baseline load value for each time segment, and divide by the number of time segments to obtain the symptom fluctuation range; Divide the symptom fluctuation range by the baseline load value to obtain the normalized symptom fluctuation coefficient.

[0026] Reference Figure 3 As shown, the construction of the elderly health fault tree knowledge graph specifically includes: Collect clinical guidelines, expert experience, and historical case data in geriatric medicine, and extract disease names, complication names, symptom names, and medication names as knowledge nodes; The top node is the elderly's disability, hospitalization, and death; the middle node is each chronic disease and its complications; and the bottom node is the stage-specific health characteristics. Determine the logical transmission relationships between each node, including AND gate relationships and OR gate relationships; Based on the top node, middle node, bottom node, and the logical transmission relationship between them, a knowledge graph of fault tree for elderly health is constructed.

[0027] Understandably, health risks in the elderly are characterized by multifactoriality, complexity, and cumulative effects. A single symptom or abnormal medication behavior often makes it difficult to directly extrapolate its potential health consequences. This solution constructs a health fault tree knowledge graph for the elderly, organizing fragmented symptom descriptions and medication behaviors according to clinical logic. It establishes a causal transmission path from microscopic manifestations to macroscopic outcomes, enabling computers to understand "which symptom combinations might indicate a certain disease" and "which diseases might further lead to disability or hospitalization," thus providing a structured reasoning basis for subsequent health risk extrapolation. This knowledge graph adopts a three-layer node structure: the top node represents the final health outcome requiring warning, with disability, hospitalization, and death among the elderly; the intermediate nodes represent the direct medical causes leading to the top node; and the bottom nodes represent the raw information that can be collected from feedback from elderly individuals. The logical relationships between nodes include "AND gates" and "OR gates." An "AND gate" indicates that multiple bottom nodes must occur simultaneously to trigger a higher-level node (e.g., multiple symptoms must appear simultaneously to diagnose a disease), while an "OR gate" indicates that the occurrence of any bottom node can trigger a higher-level node (e.g., the appearance of any typical symptom can indicate the risk of a disease). Through this hierarchical logical structure, clinical medical knowledge is transformed into computer-analyzable reasoning rules, enabling the layer-by-layer transmission and quantitative assessment of health risks for elderly individuals.

[0028] Reference Figure 4 As shown, the calculation of the probability of an elderly individual experiencing a health endpoint event and the identification of key risk transmission pathways leading to an increased probability specifically includes: The current state of the base node is input into the fault tree knowledge graph as the phased health feature. If a certain symptom or behavior exists in the phased health feature, the state of the base node corresponding to the symptom or behavior is set to 1, otherwise it is set to 0. The upward method is used to calculate the probability of occurrence of each intermediate node by starting from the bottom node and proceeding upwards layer by layer according to the logic gate rules; Starting from the top node, the downward method is used to search for all activated propagation paths in the current state layer by layer according to the logic gate rules; The minimum cut set is output as the key risk transmission path, which is used to explain the main causes of the risk at the top node.

[0029] Understandably, traditional risk assessments only output risk probabilities and cannot explain the sources of risk, leading to a lack of targeted interventions and making it difficult for elderly individuals and their families to understand the causes of risk. This solution, based on a fault tree knowledge graph, calculates the probability of occurrence of each node layer by layer from the bottom node to the top node using an upward method. Specifically, the probability of occurrence of an AND gate node is the product of the probabilities of occurrence of all its child nodes, and the probability of occurrence of an OR gate node is 1 minus the product of the probabilities of non-occurrence of all its child nodes, until the probability of occurrence of the top node is calculated, achieving a quantitative and accurate assessment of health endpoint events. Simultaneously, a downward method searches backward from the top node for all activated transmission paths, selecting the minimum cut set containing the fewest bottom node combinations as the key risk transmission path, achieving a qualitative and accurate tracing of the risk's origin. The criteria for determining the activated transmission path are as follows: The complete path is defined as follows: all bottom nodes on the path are in state 1, and the path can be continuously triggered to the top node through logic gate rules. From all activated transmission paths, the minimum cut set that leads to an increase in the probability of the top node is selected. The minimum cut set contains the minimum number of bottom node combinations. The minimum cut set may contain one or more, all of which are output as key risk transmission paths. This method enables computers to perform risk deduction based on structured clinical logic and transforms abstract risk probabilities into concrete and understandable risk causes, improving their compliance with health intervention measures and truly realizing the computability, interpretability, and interventionability of risk assessment. It should be noted that the prior probability of the bottom node can be pre-assigned based on historical statistical data or expert experience. In this scheme, the state of the bottom node is 0 / 1, which means that the current individual has the feature. Its occurrence probability is used for the upward method calculation.

[0030] The generation of differentiated early warning information and intervention recommendations based on the probability of health endpoint events and key risk transmission pathways specifically includes: The risk level is determined based on the probability of a health endpoint event, and the risk level includes three levels: low risk, medium risk, and high risk. Based on the risk level and key risk transmission pathways, generate differentiated intervention recommendations: If the risk level is low, the existing frequency of periodic inquiries will be maintained without any additional intervention. If the risk level is medium and the key risk transmission path includes a medication adherence node, a medication reminder suggestion will be generated, and a follow-up reminder will be added in the next inquiry. If the risk level is medium and the key risk transmission path includes nodes of newly emerging symptoms, then a medical advice will be generated and follow-up questions will be added in the next inquiry. If the key risk transmission path of a medium-risk individual contains multiple types of nodes, then these will be merged to generate multiple corresponding intervention recommendations. If the risk level is high, an emergency warning message will be generated and pushed to the APP or consult a doctor. The warning message includes the risk level, key risk transmission path and recommended medical department.

[0031] It is understandable that the intervention needs and warning intensities corresponding to different health risk levels vary significantly. A single intervention and warning approach cannot adapt to the individualized health status of elderly individuals and may easily lead to insufficient or excessive intervention. This solution classifies health endpoint events into low, medium, and high risk levels, and simultaneously identifies the core causes of risk by combining key risk transmission pathways. Based on this, highly tailored differentiated warning information and intervention recommendations are generated: for low-risk individuals, a routine monitoring schedule is maintained to reduce unnecessary interference; for medium-risk individuals, medication reminders and medical advice are generated based on specific risk points such as medication adherence and new symptoms, and follow-up is strengthened through subsequent inquiries; for high-risk individuals, an emergency warning is triggered, and complete information including the risk level, core causes, and recommended medical departments is pushed to relevant terminals and physicians. This achieves a tiered health management from routine monitoring and targeted intervention to emergency warnings, making intervention recommendations more targeted and warning information more instructive, ensuring timely response to high-risk situations and low-intrusion monitoring.

[0032] The optimization and adjustment of the logic gate weights in the fault tree knowledge graph based on the comparison results specifically includes: Acquire the actual health event outcomes of elderly individuals, including positive and negative events, and record the specific time and type of the events. The predicted probabilities of all elderly individuals within the historical prediction period are paired with the actual results to construct positive event sample sets and negative event sample sets respectively. Calculate the average predicted probability of positive event samples. If the value is lower than the first threshold, it is determined that there is a risk of missed reports in the graph, and the weight coefficients of each logic gate in the corresponding key risk transmission path are increased. Calculate the average predicted probability of negative event samples. If the value is higher than the second threshold, it is determined that the graph has a risk of false alarm, and the weight coefficient of each logic gate in the corresponding key risk transmission path is weakened. The optimized and adjusted fault tree knowledge graph will be used as the benchmark for the next round of risk simulation.

[0033] Understandably, medical knowledge itself is constantly evolving, and the disease manifestations of different individuals also vary. Through feedback from actual health event outcomes, the logical relationship weights in the fault tree knowledge graph can be continuously corrected, making risk projection more and more accurate and achieving personalized adaptation. It should be noted that positive events refer to actual health endpoint events such as disability, hospitalization, or death; negative events refer to health endpoint events that did not occur within the prediction period. It should be noted that, for the average predicted probability of positive event samples, ideally, the predicted probability of an actual event should be close to 1. Therefore, the error of a positive event can be expressed as 1 minus the average predicted probability of a positive event. The larger this value, the lower the predicted probability of the graph for actual events, indicating a risk of underreporting. For the average predicted probability of negative event samples, ideally, the predicted probability of an event that did not actually occur should be close to 0. Therefore, the error of a negative event can be directly expressed as the average predicted probability of a negative event. The larger this value, the higher the predicted probability of the model for events that did not occur, indicating a risk of false alarms. It should be noted that, from the positive event sample set, samples with predicted probabilities below the first threshold are selected. For each selected missed case, the key risk transmission path leading to the missed case is extracted, and the weight coefficients of each logic gate in the path are enhanced by multiplying the original weights by (1 + enhancement factor). Weight enhancement means that the logical relationships on the path are given higher importance in subsequent risk deduction. From the negative event sample set, samples with predicted probabilities above the second threshold are selected. For each selected false positive sample, the key risk transmission path leading to the false positive is extracted, and the weight coefficients of each logic gate in the path are weakened by multiplying the original weights by (1 - weakening factor). At the same time, the path may have missing intervention nodes, that is, the actual intervention measures taken are not reflected in the fault tree path and need to be added in subsequent knowledge graph updates. It should be noted that the first threshold and the second threshold can be set according to the distribution characteristics of historical data. The lower quartile of the predicted probability of all positive event samples is taken as the first threshold, and the upper quartile of the predicted probability of all negative event samples is taken as the second threshold. The enhancement factor and the weakening factor are regarded as learnable parameters. The prediction error is minimized by the gradient descent algorithm, and the enhancement factor is iteratively updated until convergence, and then determined.

[0034] Furthermore, based on the same inventive concept as the aforementioned dynamic health risk assessment and early warning method for the elderly population, this solution proposes a dynamic health risk assessment and early warning system for the elderly population, comprising: The health record module is used to acquire the initial health test data of elderly individuals, compare the initial health test data with preset chronic disease diagnostic standards, and establish hierarchical initial health records based on the comparison results. The data acquisition module is used to collect symptom descriptions, treatment information, and test data from elderly individuals based on the initial health record hierarchy, through APP registration and regular testing. The health assessment and optimization module is used to extract phased health features based on symptom descriptions and treatment information provided by elderly individuals; construct an elderly health fault tree knowledge graph, with health endpoint events as top nodes, chronic diseases, acute symptoms, and abnormal physiological indicators as intermediate nodes, and phased health features as bottom nodes, connected by logic gates to form risk transmission paths; input the current state of the phased health features as bottom nodes into the fault tree knowledge graph to calculate the probability of an elderly individual experiencing a health endpoint event and identify key risk transmission paths that lead to an increase in this probability; generate differentiated early warning information and intervention suggestions based on the probability of health endpoint events and key risk transmission paths; obtain the actual health event results of elderly individuals and compare them with the probability of health endpoint events, optimizing and adjusting the weights of logic gates in the fault tree knowledge graph based on the comparison results; The health assessment and optimization module includes: The feature extraction unit is used to extract phased health features based on symptom descriptions and treatment information provided by elderly individuals. The graph construction unit is used to construct an elderly health fault tree knowledge graph. The fault tree knowledge graph has health endpoint events as top nodes, chronic diseases, acute symptoms and abnormal physiological indicators as intermediate nodes, and stage-specific health characteristics as bottom nodes. The nodes are connected by logic gates to form a risk transmission path. A health assessment unit is used to input the current state of the stage-based health characteristics as the base node into the fault tree knowledge graph, calculate the probability of an elderly individual experiencing a health endpoint event, and identify the key risk transmission path that leads to an increase in this probability. The early warning and recommendation unit is used to generate differentiated early warning information and intervention recommendations based on the probability of health endpoint events and key risk transmission paths. The optimization and adjustment unit is used to obtain the actual health event results of elderly individuals, compare them with the probability of health endpoint events, and optimize and adjust the logic gate weights in the fault tree knowledge graph based on the comparison results.

[0035] In summary, the advantages of this invention are: through hierarchical filing, differentiated data collection, fault tree knowledge graph reasoning, and dynamic weight iteration, it achieves accurate quantitative projection, interpretable source tracing, and tiered intervention of health risks in the elderly, effectively adapting to the entire disease course of the elderly population, and combining practicality, accuracy, and personalization.

[0036] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A method for dynamic assessment and early warning of health risks for the elderly population, characterized in that, include: Acquire the initial health test data of elderly individuals, compare the initial health test data with the preset chronic disease diagnostic criteria, and establish a hierarchical initial health record based on the comparison results; Based on the hierarchy of initial health records, information on symptoms, treatment, and test results is collected from elderly individuals through APP registration and regular testing. Based on symptom descriptions and treatment information provided by elderly individuals, phased health characteristics are extracted; A fault tree knowledge graph for elderly health is constructed. The fault tree knowledge graph has health endpoint events as top nodes, chronic diseases, acute symptoms and abnormal physiological indicators as intermediate nodes, and stage-specific health characteristics as bottom nodes. The nodes are connected by logic gates to form risk transmission paths. The current state of the base node is input into the fault tree knowledge graph as the phased health characteristics, and the probability of an elderly individual experiencing a health endpoint event is calculated, and the key risk transmission path that leads to the increase of this probability is identified. Based on the probability of health endpoint events and key risk transmission pathways, differentiated early warning information and intervention recommendations are generated; The actual health events that occur in elderly individuals are obtained and compared with the probability of health endpoint events. Based on the comparison results, the weights of logic gates in the fault tree knowledge graph are optimized and adjusted.

2. The method for dynamic assessment and early warning of health risks for the elderly population according to claim 1, characterized in that, The establishment of hierarchical initial health records specifically includes: If an elderly individual's initial health check data does not identify a chronic disease, a natural disease monitoring file should be established. Natural disease surveillance records use the initial health check data as a reference baseline, record only basic physiological indicators, and are marked as level one. If an elderly individual is diagnosed with a chronic disease based on their initial health check data, obtain physician diagnostic data and treatment plans, and establish a chronic disease management file. Chronic disease management records include diagnosis information, medication regimens, and treatment goals; the record level is marked as level two. We continuously monitor and obtain medical diagnosis information from elderly individuals, and dynamically adjust the file level and supplement relevant data based on the feedback.

3. The method for dynamic assessment and early warning of health risks for the elderly population according to claim 2, characterized in that, The hierarchical structure based on initial health records, through APP registration and regular testing, collects symptom descriptions, treatment information, and test data from elderly individuals, specifically including: Differentiated collection cycles and dimensions are set based on the archive hierarchy; By registering an app, the app pushes voice inquiry content that matches the age level to elderly individuals, and simultaneously collects symptom descriptions and treatment information from elderly individuals in response to the inquiries. Based on regular offline health check-ups, data on various physiological indicators and test results generated by the corresponding testing processes for elderly individuals are collected. The subjective feedback information collected by the APP is linked and integrated with the objective data obtained from regular testing, and then uniformly collected into the health records of the corresponding elderly individuals, which are stored as periodic health characteristic data.

4. The method for dynamic assessment and early warning of health risks for the elderly population according to claim 3, characterized in that, The extraction of phased health characteristics based on symptom descriptions and treatment information from elderly individuals specifically includes: For elderly individuals corresponding to Level 1 records, if new symptom keywords are reported, the symptom name, occurrence time, and duration are extracted as characteristics of the new symptom. For elderly individuals corresponding to Level 1 records, if they report no symptoms and no medical records in several consecutive inquiries, they are set to a continuous health status label. For elderly individuals corresponding to Level 2 records, the proportion of on-time medication responses within a preset time period is counted to extract characteristics of poor medication adherence. For elderly individuals corresponding to Level 2 records, the frequency and severity of various symptoms reported within a preset time period are statistically analyzed, the symptom fluctuation coefficient is calculated, and the characteristics of unstable symptom control are extracted. For elderly individuals corresponding to Level 2 records, when symptoms that have not been previously reported are recorded in several consecutive inquiry responses, and these symptoms match the symptoms in the list of known complications of the chronic disease, they are extracted as complication warning features.

5. The method for dynamic assessment and early warning of health risks for the elderly population according to claim 4, characterized in that, The construction of the elderly health fault tree knowledge graph specifically includes: Collect clinical guidelines, expert experience, and historical case data in geriatric medicine, and extract disease names, complication names, symptom names, and medication names as knowledge nodes; The top node is the elderly's disability, hospitalization, and death; the middle node is each chronic disease and its complications; and the bottom node is the stage-specific health characteristics. Determine the logical transmission relationships between each node, including AND gate relationships and OR gate relationships; Based on the top node, middle node, bottom node, and the logical transmission relationship between them, a knowledge graph of fault tree for elderly health is constructed.

6. The method for dynamic assessment and early warning of health risks for the elderly population according to claim 5, characterized in that, The calculation of the probability of an elderly individual experiencing a health endpoint event and the identification of key risk transmission pathways leading to an increased probability specifically include: The current state of the base node is input into the fault tree knowledge graph as the phased health feature. If a symptom or behavior exists in the phased health feature, the state of the base node corresponding to the symptom or behavior is set to 1, otherwise it is set to 0. The upward method is used to calculate the probability of occurrence of each intermediate node by starting from the bottom node and proceeding upwards layer by layer according to the logic gate rules; The downward method is used to search for all activated propagation paths in the current state, starting from the top node and following the logic gate rules layer by layer downward. The minimum cut set is output as the key risk transmission path, which is used to explain the main causes of the risk at the top node.

7. The method for dynamic assessment and early warning of health risks for the elderly population according to claim 6, characterized in that, The generation of differentiated early warning information and intervention recommendations based on the probability of health endpoint events and key risk transmission pathways specifically includes: The risk level is determined based on the probability of a health endpoint event, and the risk level includes three levels: low risk, medium risk, and high risk. Based on the risk level and key risk transmission pathways, generate differentiated intervention recommendations: If the risk level is low, the existing frequency of periodic inquiries will be maintained without any additional intervention. If the risk level is medium and the key risk transmission path includes a medication adherence node, a medication reminder suggestion will be generated, and a follow-up reminder will be added in the next inquiry. If the risk level is medium and the key risk transmission path includes nodes of newly emerging symptoms, then a medical advice will be generated and follow-up questions will be added in the next inquiry. If the key risk transmission path of a medium-risk individual contains multiple types of nodes, then these nodes will be merged to generate multiple corresponding intervention recommendations. If the risk level is high, an emergency warning message will be generated and pushed to the APP or consult a doctor. The warning message includes the risk level, key risk transmission path and recommended medical department.

8. A method for dynamic assessment and early warning of health risks for the elderly population according to claim 7, characterized in that, The optimization and adjustment of the logic gate weights in the fault tree knowledge graph based on the comparison results specifically includes: Acquire the actual health event outcomes of elderly individuals, including positive and negative events, and record the specific time and type of the events. The predicted probabilities of all elderly individuals within the historical prediction period are paired with the actual results to construct positive event sample sets and negative event sample sets respectively. Calculate the average predicted probability of positive event samples. If the value is lower than the first threshold, it is determined that there is a risk of missed reports in the graph, and the weight coefficients of each logic gate in the corresponding key risk transmission path are increased. Calculate the average predicted probability of negative event samples. If the value is higher than the second threshold, it is determined that the graph has a false alarm risk, and the weight coefficient of each logic gate in the corresponding key risk transmission path is weakened. The optimized and adjusted fault tree knowledge graph will be used as the benchmark for the next round of risk simulation.

9. A dynamic health risk assessment and early warning system for the elderly population, characterized in that, The method for dynamic assessment and early warning of health risks for the elderly population as described in any one of claims 1-8 includes: The health record module is used to acquire the initial health test data of elderly individuals, compare the initial health test data with preset chronic disease diagnostic standards, and establish hierarchical initial health records based on the comparison results. The data acquisition module is used to collect symptom descriptions, treatment information, and test data from elderly individuals based on the initial health record hierarchy, through APP registration and regular testing. The health assessment and optimization module is used to extract phased health features based on symptom descriptions and treatment information provided by elderly individuals; construct an elderly health fault tree knowledge graph, with health endpoint events as top nodes, chronic diseases, acute symptoms, and abnormal physiological indicators as intermediate nodes, and phased health features as bottom nodes, connected by logic gates to form risk transmission paths; input the current state of the phased health features as bottom nodes into the fault tree knowledge graph to calculate the probability of an elderly individual experiencing a health endpoint event and identify key risk transmission paths that lead to an increase in this probability; generate differentiated early warning information and intervention suggestions based on the probability of health endpoint events and key risk transmission paths; obtain the actual health event results of elderly individuals and compare them with the probability of health endpoint events, optimizing and adjusting the weights of logic gates in the fault tree knowledge graph based on the comparison results.

10. A dynamic health risk assessment and early warning system for the elderly population according to claim 9, characterized in that, The health assessment and optimization module includes: The feature extraction unit is used to extract phased health features based on symptom descriptions and treatment information provided by elderly individuals. The graph construction unit is used to construct an elderly health fault tree knowledge graph. The fault tree knowledge graph has health endpoint events as top nodes, chronic diseases, acute symptoms and abnormal physiological indicators as intermediate nodes, and stage-specific health characteristics as bottom nodes. The nodes are connected by logic gates to form a risk transmission path. A health assessment unit is used to input the current state of the stage-based health characteristics as the base node into the fault tree knowledge graph, calculate the probability of an elderly individual experiencing a health endpoint event, and identify the key risk transmission path that leads to an increase in this probability. The early warning and recommendation unit is used to generate differentiated early warning information and intervention recommendations based on the probability of health endpoint events and key risk transmission paths. The optimization and adjustment unit is used to obtain the actual health event results of elderly individuals, compare them with the probability of health endpoint events, and optimize and adjust the logic gate weights in the fault tree knowledge graph based on the comparison results.

Images