Multi-factor coupling root cause positioning method and system based on health problem forming path

Abstract

The invention discloses a multi-factor coupling root cause positioning method and system based on a health problem forming path, and relates to the technical field of medical health data processing, and the method comprises the specific steps: obtaining multi-source data through a medical health big data platform, carrying out the standardized integration and alignment, recognizing a health problem, and generating structured information; then modeling by stages to construct a health problem time sequence knowledge graph; anchoring data by using an algorithm and coupling and analyzing influence factors; mining markers, and calculating confidence to generate a root cause set; and finally, outputting a result according to confidence, associating information and linking online and offline intelligent diagnosis and treatment ecology. According to the method, multi-dimensional health data are integrated, a time sequence knowledge graph is constructed, and accurate anchoring of the data is achieved; health factor effects are determined through time sequence coupling analysis, early warning markers are mined, and root cause positioning reliability is improved; and multi-element association and quantitative standards are also established, multiple platforms are connected, accurate reference is provided for medical decision making, and development of a health diagnosis and treatment system is promoted.

Description

Technical Field

[0001] This invention relates to the field of medical and health data processing technology, specifically to a multi-factor coupled root cause localization method and system based on the formation path of health problems. Background Technology

[0002] With the deepening of digital transformation in the healthcare field, multi-source health data resources are continuously enriched, covering continuous time-series data generated by online device sensors, detection and diagnostic data accumulated during offline clinical diagnosis and treatment, lifestyle records reflecting an individual's long-term status, and health assessment information. At the same time, authoritative disease diagnostic standards are gradually being improved, integrating core contents such as various clinical pathway norms, symptom judgment criteria, and indicator reference ranges, laying the foundation for the scientific identification of health problems. As people's awareness of health management continues to improve, simply focusing on the surface symptoms of health problems can no longer meet the needs. Tracing the root causes of health problems, clarifying the complete evolution path, and then achieving early prevention and precise intervention have become the core demands of the healthcare field. This has also driven the development of root cause localization technology towards multi-data fusion and multi-factor correlation analysis.

[0003] Traditional root cause analysis techniques for health problems have many limitations and are difficult to adapt to the needs of tracing complex health issues. Some techniques focus only on a single type of data and fail to effectively integrate multi-source heterogeneous information, resulting in a one-sided understanding of the factors that contribute to health problems and an inability to fully cover all possible influencing factors. Most analytical methods lack a systematic consideration of the time dimension, ignoring the stage-by-stage evolution of health problems from initial causes to final manifestations, and making it difficult to capture the temporal characteristics and dynamic changes of the effects of factors. At the same time, traditional techniques fail to quantify the interaction relationships between various influencing factors and cannot accurately distinguish the synergistic, superimposed, or inhibitory effects of different factors. Root cause judgment lacks rigorous logic and data support. In addition, traditional early warning markers are mostly generalized designs and lack personalized verification steps, resulting in insufficient accuracy in early warning. Moreover, the analysis results are often independent of existing medical and health platforms, making it difficult to achieve data interoperability and linkage with the diagnosis and treatment ecosystem, thus limiting their application value in real-world scenarios. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a multi-factor coupled root cause localization method and system based on the formation path of health problems. This method integrates multi-dimensional health data through a medical and health big data sharing platform, identifies health problems after standardization and timestamp alignment, constructs a phased temporal knowledge graph, and uses a dedicated anchoring algorithm to achieve precise matching between data and graph nodes. It then conducts multi-factor temporal coupling analysis, identifies personalized dynamic early warning markers with dual verification by mining key feature patterns, and generates an ordered root cause set using a root cause confidence dynamic algorithm. This set is then standardized and integrated with multiple platforms and the intelligent diagnosis and treatment ecosystem. This method takes into account the interaction of multiple factors and the temporal evolution of health problems, improving the accuracy and systematic nature of root cause localization and providing scientific support for health management and clinical diagnosis and treatment.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: On the one hand, a multi-factor coupled root cause localization method based on the formation path of health problems, the specific steps of which are as follows:

[0006] S1, Data Integration and Problem Identification: By acquiring online device sensor time-series data, offline clinical test data, lifestyle data and health assessment data of the target object through the medical and health big data sharing platform, after standardization and timestamp alignment, combined with the disease diagnosis standard library of the health information system, the current health problems of the target object are identified and structured descriptive information is generated.

[0007] S2, Temporal Graph Construction: Based on the occurrence and development patterns of health problems, and combined with the historical temporal case data of the medical and health big data sharing platform, the identified health problems are modeled in stages to construct a health problem temporal knowledge graph containing multiple time-dimensional knowledge subgraphs;

[0008] S3, Data Anchoring and Coupling Analysis: The time-series-graph anchoring algorithm is applied to calculate the fusion anchoring strength between the target data and the nodes of the time-series knowledge graph. The target data is anchored to the corresponding nodes and related factors according to the time dimension. Based on the anchoring results, time-series coupling analysis is performed on multiple health influencing factors.

[0009] The target data includes online device sensing time-series data and offline clinical test data in step S1. The online device sensing time-series data includes heart rate variability, 24-hour ambulatory blood pressure, blood glucose fluctuation curve, exercise duration, exercise intensity and sleep structure data. The offline clinical test data includes laboratory test reports, imaging examination results, doctor's diagnosis records and medical records. All target data are accompanied by timestamps accurate to the minute and notes on the collection scenario.

[0010] S4, Marker Mining and Confidence Calculation: Search for recurring device data patterns and clinical feature patterns before key turning points of health problems in the time-series knowledge graph, define personalized dynamic early warning markers, apply the root cause confidence dynamic algorithm to calculate the confidence of candidate health root causes, and generate a root cause candidate set.

[0011] S5, Results Output and Ecosystem Linkage: Output the root cause localization results of health problems in order of confidence, link them with the corresponding time-series knowledge graph and personalized dynamic early warning marker information, and connect them with the medical and health big data sharing platform, health information system and cloud platform in a standardized format to link the online and offline intelligent diagnosis and treatment ecosystem.

[0012] Furthermore, the online device sensor time-series data obtained through the medical and health big data sharing platform includes heart rate variability, 24-hour ambulatory blood pressure, blood glucose fluctuation curve, exercise duration, exercise intensity and sleep structure data; offline clinical test data includes laboratory test reports, imaging examination results, doctor's diagnosis records and medical records; lifestyle data includes dietary preferences, smoking and drinking history, work and rest patterns and exercise habits records; and health assessment data includes physical examination reports, past medical history and chronic disease follow-up records.

[0013] Furthermore, the disease diagnostic criteria library of the health information system includes the World Health Organization disease classification standards, the chronic disease diagnosis guidelines issued by the National Health Commission, clinical pathway specifications, symptom judgment standards, laboratory indicator reference ranges, disease duration requirements, and imaging characteristic standards. The laboratory indicator reference ranges are divided into specific numerical intervals according to age, gender, and physiological stage. The symptom judgment standards clearly define the symptom manifestations, frequency of attacks, and duration requirements.

[0014] Furthermore, the formation path of the health problem temporal knowledge graph is divided into an initial triggering phase, a functional imbalance phase, and a health problem manifestation phase. The initial triggering phase corresponds to 9-12 months before the manifestation of the health problem, the functional imbalance phase corresponds to 3-8 months before the manifestation of the health problem, and the health problem manifestation phase corresponds to 0-2 months before the health problem is identified. The initial triggering phase is divided into time nodes by month, the functional imbalance phase by half-month, and the health problem manifestation phase by week. Each time node in the initial triggering phase is associated with lifestyle data, environmental data, and device sensor data. Each time node in the functional imbalance phase is associated with clinical test data, omics data, and immune factor data. Each time node in the health problem manifestation phase is associated with clinical diagnosis data, symptom record data, and core physiological indicator data.

[0015] Furthermore, the relationships between nodes in the time-series knowledge graph include causal relationships, correlation relationships, and temporal chronological relationships. The causal relationship satisfies that after factor A appears, factor B will inevitably appear within a 95% confidence interval, and B will still appear when there are no other interfering factors. The correlation relationship satisfies that the Pearson correlation coefficient between factors is ≥0.6 and the p-value is <0.05. The temporal chronological relationship satisfies that factor A appears at least one data collection cycle earlier than factor B, and there are no reverse occurrence cases. All relationship edges are accompanied by a correlation strength value, the causal relationship strength is 1.0, the correlation relationship strength is the corresponding correlation coefficient value, and the temporal chronological relationship strength is 0.8.

[0016] Furthermore, the mathematical expression of the time-series-graph anchoring algorithm is: ,in: Let be the fusion anchoring strength between the i-th type of target data and the j-th node of the temporal knowledge graph at time t; This is the time decay dynamic factor; Similarity is calculated by fusing target data and map features; For graph nodes Path importance weights; is the temporal collaborative gain factor; i represents the category number of the target data, i takes the value of 1 or 2, i=1 corresponds to the device sensing time series data, i=2 corresponds to the clinical test data; j represents the index of the node in the temporal knowledge graph associated with the i-th category of target data.

[0017] Furthermore, the specific steps for performing time-series coupling analysis on multiple health influencing factors based on anchoring results are as follows: First, determine the scope of health influencing factors participating in the coupling analysis, including lifestyle factors, environmental factors, immune factors, and metabolic indicators. Then, extract the fusion anchoring strength data of the corresponding time-series knowledge graph nodes for each health influencing factor. Select the anchoring strength of each type of influencing factor corresponding to each node in sequence according to the time dimension, and set a coupling effect judgment threshold. When multiple influencing factors act together, the increase in the anchoring strength of the corresponding node is greater than 20% of the sum of the anchoring strengths when each type of factor acts alone, and it is judged as a synergistic effect. When multiple influencing factors are superimposed... If the increase in anchorage strength of the corresponding node is greater than 50% of the maximum anchorage strength when each factor acts alone but less than 20% of the sum of the anchorage strengths when each factor acts alone, it is determined to be a superposition effect. If the anchorage strength of the corresponding node of other influencing factors decreases by more than 10% after the action of a certain influencing factor, it is determined to be an inhibitory effect. Then, the coefficients of various coupling effects are quantified. The coefficient of synergistic effect is 1.2-1.5, the coefficient of superposition effect is 1.0-1.2, and the coefficient of inhibitory effect is 0.5-0.9. Finally, the coupling effect type, coefficient and anchorage strength change data of each node are sorted in time order to complete the time-series coupling analysis and output the analysis results.

[0018] Furthermore, the personalized dynamic early warning markers must undergo two levels of verification. The first level is verified in the target object's own historical time-series data, requiring a co-occurrence frequency of ≥80% before key turning points and consecutive occurrences of ≥3 times. The second level is verified by retrieving anonymized historical case data of 500 similar health problems from a medical and health big data sharing platform, requiring an early warning accuracy rate of ≥75%. If both levels of verification are met, the marker is officially confirmed as a personalized dynamic early warning marker. If either verification fails, the feature pattern is searched again and the verification process is repeated, up to a maximum of 3 times.

[0019] Furthermore, the mathematical expression for the root cause confidence dynamic algorithm is: ,in: The final confidence level of the k-th candidate root cause; Adaptive weights for root cause path coverage; , , For dynamic balancing weights; Let be the fusion anchoring strength between the i-th type of target data and the j-th node of the temporal knowledge graph at time t; Let k be the coupling coefficient of the root cause k at the j-th map node of the i-th type of target data; The integral represents the multi-stage coupling effect of the root cause k. The time of initiation of the root cause; This is the current analysis time; The effectiveness of early warning based on dynamic markers corresponding to root cause k; denoted as the historical data matching degree of root cause k; i represents the category index of the target data, i can be 1 or 2, i=1 corresponds to device sensing time series data, i=2 corresponds to clinical test data; j represents the index of the node in the time series knowledge graph associated with the i-th category of target data. This represents the total number of nodes in the temporal knowledge graph that participate in the anchoring calculation of the i-th type of target data corresponding to root cause k.

[0020] On the other hand, a multi-factor coupled root cause localization system based on the pathogenesis of health problems includes:

[0021] Data integration and problem identification module: It is used to acquire online device sensor time-series data, offline clinical test data, lifestyle data and health assessment data of target objects through the medical and health big data sharing platform, perform standardized integration and timestamp alignment operations, and identify health problems and generate structured description information by combining the disease diagnosis standard library of the health information system.

[0022] The time-series graph construction module is used to build a health problem time-series knowledge graph that includes multiple time-dimensional knowledge subgraphs, based on the occurrence and development patterns of health problems and combined with historical time-series case data from the medical and health big data sharing platform.

[0023] Data Anchoring and Coupling Analysis Module: This module is used to calculate the fusion anchoring strength between target data and time-series knowledge graph nodes using the time-series-graph anchoring algorithm. It anchors the target data to the corresponding nodes and related factors according to the time dimension, and performs time-series coupling analysis on multiple health influencing factors based on the anchoring results.

[0024] The marker mining and confidence calculation module is used to search for recurring device data patterns and clinical feature patterns before key turning points of health problems in the time-series knowledge graph, define personalized dynamic early warning markers and complete two-level verification, apply the root cause confidence dynamic algorithm to calculate the confidence of candidate health root causes and generate a root cause candidate set.

[0025] Results Output and Ecosystem Linkage Module: This module outputs the root cause localization results of health problems in order of confidence, links them to the corresponding time-series knowledge graph and personalized dynamic early warning marker information, and connects them to the medical and health big data sharing platform, health information system and cloud platform in a standardized format, linking the online and offline intelligent diagnosis and treatment ecosystem.

[0026] Compared with existing technologies, this multi-factor coupled root cause localization method and system based on the formation path of health problems has the following advantages:

[0027] I. This invention integrates multi-dimensional health data, standardizes and aligns it with timestamps, and combines it with a comprehensive disease diagnostic standard library to accurately identify health problems. It constructs a temporal knowledge graph covering multiple time stages, achieving precise anchoring of target data and graph nodes. Based on temporal coupling analysis, it deeply analyzes the interaction of various health influencing factors, clarifies different types and intensities of effects, breaks through the limitations of single-factor analysis, and can completely trace the formation path of health problems from initial causes to final manifestations. By mining characteristic patterns before key turning points, it identifies personalized dynamic early warning markers that have undergone double verification. With the help of dynamic algorithms, it accurately calculates the confidence of root causes, improves the pertinence and reliability of root cause localization, provides a scientific basis for early intervention of health problems, and solves the shortcomings of traditional root cause analysis in lacking temporal correlation and multi-factor synergistic consideration, making the tracing of the root causes of health problems more systematic and logical.

[0028] II. This invention establishes a temporal knowledge graph and clarifies the multi-dimensional relationships and intensity quantification standards between nodes, making the root cause analysis process traceable and interpretable. Combined with a temporal-graph anchoring algorithm, it achieves efficient integration of data and knowledge graph. By distinguishing time nodes at different stages and associating data types, it enables a refined breakdown of the health problem formation process. Coupled analysis quantifies the interaction coefficients of various factors, providing quantitative support for root cause judgment. It connects to multiple platforms in a standardized format, linking online and offline intelligent diagnosis and treatment ecosystems. Root cause results are output in order of confidence level and associated with relevant graphs and marker information, providing comprehensive and accurate references for medical decision-making. This helps health management shift from passive response to proactive prevention, improving diagnosis and treatment efficiency and health management quality, meeting personalized diagnosis and treatment needs in different scenarios, and promoting the development of the health diagnosis and treatment system towards a more precise, efficient, and collaborative direction.

[0029] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0031] Figure 1 A step-by-step framework diagram for a multi-factor coupled root cause localization method based on the formation path of health problems;

[0032] Figure 2 This is a block diagram of a multi-factor coupled root cause localization system based on the path formation of health problems;

[0033] Figure 3 The flowchart shows the input and output steps of the multi-factor coupling root cause localization method based on the formation path of health problems. Detailed Implementation

[0034] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0035] Example 1:

[0036] Implementation examples of a multi-factor coupling root cause localization method based on the formation path of health problems.

[0037] S1, Data Integration and Problem Identification: Relevant data on women aged 45 are collected through a medical and health big data sharing platform. Online device sensor time-series data covers heart rate variability, 24-hour ambulatory blood pressure and blood glucose fluctuation curves, exercise duration, exercise intensity, and sleep structure data for the past 12 months. All data are accompanied by timestamps accurate to the minute and notes on the collection scenario. Offline clinical test data includes laboratory test reports, imaging examination results, doctor's diagnosis records, and medical records for the past 12 months. Lifestyle data covers dietary preferences, smoking and drinking history, work and rest patterns, and exercise habits. Health assessment data includes physical examination reports, past medical history, and chronic disease follow-up records for the past 3 years. All the above data were standardized and integrated into a unified data format, and timestamps were aligned to form a coherent and usable dataset from different sources and collection frequencies. This dataset was then matched and analyzed using a disease diagnostic standard library from the health information system. This library includes the World Health Organization's Classification of Diseases, guidelines for chronic disease diagnosis issued by the National Health Commission, clinical pathway standards, symptom assessment criteria, laboratory indicator reference ranges categorized by age, gender, and physiological stage, disease duration requirements, and imaging characteristic standards. This ensured that the identification of health problems conformed to authoritative standards. Ultimately, it was found that the woman's systolic blood pressure had repeatedly been between 120-139 mmHg and her diastolic blood pressure between 80-89 mmHg over the past four months, with occasional dizziness and headaches, meeting the criteria for prehypertension. Therefore, the prehypertension health problem was identified, and a structured descriptive information was generated, including the health problem name, discovery time, related symptoms, and key abnormal laboratory indicators. This provided accurate basic data for subsequent time-series modeling. Figure 1 As shown.

[0038] S2, Temporal Graph Construction: Based on the occurrence and development patterns of prehypertension, and combined with historical temporal case data of this type of health problem in the medical and health big data sharing platform, phased temporal modeling is performed on the identified prehypertension health problems to construct a temporal knowledge graph of health problems. This temporal knowledge graph contains three time-dimensional knowledge subgraphs: the initial triggering phase corresponds to 9-12 months before the manifestation of the health problem; the functional imbalance phase corresponds to 3-8 months before the manifestation of the health problem; and the health problem manifestation phase corresponds to 0-2 months before the health problem is identified, clearly presenting the complete development path of the health problem from triggering factors to manifestation. The initial triggering phase is divided into 12 time nodes by month, with each node associated with lifestyle data, environmental data, and device sensor data for that month; the functional imbalance phase is divided into 12 time nodes by half-month, with each node associated with clinical test data, omics data, and immune factor data for that half-month; and the health problem manifestation phase is divided into 8 time nodes by week, with each node associated with clinical diagnostic data, symptom record data, and core physiological indicator data for that week, ensuring that key influencing data of different stages accurately correspond to time nodes. Simultaneously, the relationships between the nodes in the graph are defined. The causal relationship satisfies that factor B will inevitably appear within a 95% confidence interval after factor A appears, and that B will still appear when there are no other interfering factors, with a relationship strength of 1.0. The correlation relationship satisfies that the Pearson correlation coefficient between factors is ≥0.6 and the p-value is <0.05, with the relationship strength being the corresponding correlation coefficient value. The temporal relationship satisfies that factor A appears at least one data collection period earlier than factor B and there are no reverse occurrence cases, with a relationship strength of 0.8. This clarifies the logic and tightness of the relationships between each factor, providing clear graph support for subsequent data anchoring.

[0039] S3, Data Anchoring and Coupling Analysis: The time-series-graph anchoring algorithm is applied to calculate the fusion anchoring strength between the target data and the nodes of the time-series knowledge graph. The mathematical expression of the time-series-graph anchoring algorithm is: ,in: Let be the fusion anchoring strength between the i-th type of target data and the j-th node of the temporal knowledge graph at time t; This is the time decay dynamic factor; Similarity is calculated by fusing target data and map features; For graph nodes Path importance weights; , where i represents the category number of the target data, i is 1 or 2, i=1 corresponds to device sensing time-series data, i=2 corresponds to clinical testing data; j represents the node number in the time-series knowledge graph associated with the i-th type of target data. The target data includes online device sensing time-series data and offline clinical testing data from step S1. The algorithm allows different types of target data to be accurately matched to the corresponding graph nodes and associated factors according to the time dimension, achieving deep binding between data and graph. Based on the anchoring results, a time-series coupling analysis is performed on multiple health influencing factors. First, the scope of influencing factors involved in the analysis is determined to include lifestyle factors, environmental factors, immune factors, and metabolic indicators. Then, the fusion anchoring strength data of the time-series knowledge graph nodes corresponding to each influencing factor is extracted. The anchoring strength of each type of influencing factor corresponding to each node is selected sequentially according to the time dimension. A coupling effect judgment threshold is set. The analysis shows that when high-salt diet and lack of exercise work together, the anchoring strength increase of a certain node in the functional imbalance stage is 28% of the total anchoring strength when each factor acts alone. A value greater than 20% is judged as a synergistic effect. The synergistic effect coefficient is... The value is 1.4; after long-term sleep deprivation and mental stress are combined, the anchoring strength of the corresponding node increases by 55% of the maximum anchoring strength when each factor acts alone and is less than 20% of the sum of the anchoring strengths when each factor acts alone. This is judged as a superposition effect, and the superposition effect coefficient is 1.1; after regular water intake, the anchoring strength of the node corresponding to high salt diet decreases by 13%, which is greater than 10% and is judged as an inhibitory effect. The inhibitory effect coefficient is 0.8; finally, the coupling effect type coefficient and anchoring strength change data of each node are sorted out in chronological order, clearly presenting the mutual influence logic of multiple factors at different time nodes, providing core analytical basis for root cause mining.

[0040] S4, Marker Mining and Confidence Calculation: The system searches the time-series knowledge graph for recurring device data patterns and clinical characteristic patterns before key turning points in prehypertension. Preliminary screening identifies patterns of systolic blood pressure fluctuating between 115-120 mmHg for four consecutive weeks, with less than 40 minutes of exercise per week and at least three instances of high-salt food intake per day. These are used as candidate personalized dynamic early warning markers. Two levels of validation are performed on these candidate markers. The first level validates them in the target subject's own historical time-series data, showing a co-occurrence frequency of 88% and five consecutive occurrences before key turning points, meeting the requirement of a co-occurrence frequency ≥80% and at least three consecutive occurrences. The second level validates them by retrieving anonymized historical case data from 500 prehypertension cases from a medical and health big data sharing platform. The marker's early warning accuracy is 83%, meeting the requirement of an early warning accuracy ≥75%. These two levels of validation ensure that the marker can accurately predict the development trend of health problems. The root cause confidence dynamic algorithm is applied, combined with information such as the matching degree of historical data on the effectiveness of marker warnings obtained from the coupling analysis in step S3, to calculate the confidence of candidate health root causes. The mathematical expression of the root cause confidence dynamic algorithm is as follows: ,in: The final confidence level of the k-th candidate root cause; Adaptive weights for root cause path coverage; , , For dynamic balancing weights; Let be the fusion anchoring strength between the i-th type of target data and the j-th node of the temporal knowledge graph at time t; Let k be the coupling coefficient of the root cause k at the j-th map node of the i-th type of target data; The integral represents the multi-stage coupling effect of the root cause k. The time of initiation of the root cause; This is the current analysis time; The effectiveness of early warning based on dynamic markers corresponding to root cause k; denoted as the historical data matching degree of root cause k; i represents the category index of the target data, i can be 1 or 2, i=1 corresponds to device sensing time series data, i=2 corresponds to clinical test data; j represents the index of the node in the time series knowledge graph associated with the i-th category of target data. This represents the total number of nodes in the time-series knowledge graph that participate in the anchoring calculation of the i-th type of target data corresponding to root cause k. Candidate root causes include high-salt diet, lack of exercise, long-term sleep deprivation, and mental stress. The calculated confidence scores are 0.93 for high-salt diet, 0.89 for lack of exercise, 0.76 for long-term sleep deprivation, and 0.72 for mental stress. A candidate set of root causes is generated, sorted from high to low confidence, to provide an accurate basis for the final root cause output.

[0041] S5, Output and Ecosystem Integration: The system outputs the root cause localization results for prehypertension, sorted by confidence level. High-salt diet has a confidence level of 0.93, lack of exercise 0.89, chronic sleep deprivation 0.76, and mental stress 0.72. Simultaneously, it links the results to corresponding health problem time-series knowledge graphs and personalized dynamic early warning markers. This standardized format is used to connect with the medical and health big data sharing platform's health information system and cloud platform, ensuring that each platform can quickly obtain unified and standardized root cause analysis data. It integrates an online and offline intelligent diagnosis and treatment ecosystem. Online, through a health management app, it pushes root cause analysis reports, personalized dietary adjustment suggestions to reduce high-salt food intake, exercise plans (3 times a week, 50 minutes each time, moderate-intensity exercise), and sleep adjustment plans to ensure 7 hours of sleep daily, along with stress relief guidance. Offline, it collaborates with community hospitals and cardiovascular specialist clinics to arrange regular blood pressure monitoring, nutritional consultations, exercise guidance, and psychological counseling services for the target population, forming a closed-loop health intervention process to help the target population improve their health in a targeted manner. Figure 3 As shown.

[0042] In summary, this multi-factor coupled root cause localization method based on the health problem formation path targets prehypertension as the root cause. It integrates multi-source health-related data through a medical and health big data sharing platform, and identifies health problems by standardizing the data and matching it with an authoritative diagnostic standard library. A phased temporal knowledge graph is constructed based on the development patterns of health problems, and a temporal-graph anchoring algorithm is used to achieve precise correlation between data and the graph. Furthermore, temporal coupling analysis clarifies the multi-factor interaction patterns. Personalized dynamic early warning markers are determined through two levels of validation, and the confidence of candidate root causes is quantified using a root cause confidence dynamic algorithm. Finally, the results are output sequentially and linked to the intelligent diagnosis and treatment ecosystem, fully aligning with the health problem formation path to achieve precise root cause localization and targeted intervention support.

[0043] Example 2:

[0044] Implementation example of a multi-factor coupled root cause localization system based on the formation path of health problems.

[0045] This system is applicable to the aforementioned multi-factor coupled root cause localization method based on the formation path of health problems. The system architecture includes five core functional modules, which work together to complete the root cause localization of health problems. Specific implementation steps are as follows: Figure 2 As shown:

[0046] Data integration and problem identification module:

[0047] This module, serving as the core of the system's data input and preprocessing, possesses multi-source data access and standardized processing capabilities. Through an interface connecting to a medical and health big data sharing platform, it automatically collects online device sensor time-series data of the target object, including heart rate variability, 24-hour ambulatory blood pressure and blood glucose fluctuation curves, exercise duration, exercise intensity, and sleep structure data, along with time stamps accurate to the minute and collection scenario notes. Offline data includes clinical testing data, laboratory test reports, imaging examination results, doctor's diagnostic records and medical records, lifestyle data such as dietary preferences, smoking and drinking history, sleep patterns and exercise habits, health assessment data, physical examination reports, past medical history, and chronic disease follow-up records, ensuring comprehensive acquisition of the target object's health-related data. The module incorporates a data standardization engine to unify the format of the collected heterogeneous data, remove outliers, and complete missing values. Furthermore, a timestamp alignment algorithm integrates data from different sources and collection frequencies into a unified dataset along the time dimension, ensuring data consistency and usability. Simultaneously, the module integrates a disease diagnostic standard library from the health information system. This standard library includes the World Health Organization's disease classification standards, the National Health Commission's guidelines for chronic disease diagnosis, clinical pathway specifications, symptom assessment standards, laboratory indicator reference ranges categorized by age, gender, and physiological stage, disease duration requirements, and imaging characteristic standards. By intelligently matching and comparing the integrated target data with the standard library, the module automatically identifies the target's health problems and generates a structured descriptive report containing information such as the name of the health problem, the discovery time, related symptoms, and abnormal indicators. This provides accurate and standardized basic data support for subsequent modules.

[0048] Time series graph construction module:

[0049] The module receives structured description reports from the data integration and problem identification module, as well as historical time-series case data of similar health problems provided by the medical and health big data sharing platform. Based on the occurrence and development patterns of health problems, it automatically performs phased time-series modeling to construct a health problem time-series knowledge graph. The module first clarifies the three core phases of the time-series graph: the initial trigger phase (9-12 months before health problem manifestation), the functional imbalance phase (3-8 months before health problem manifestation), and the health problem manifestation phase (0-2 months before health problem identification). It then divides the time nodes of each phase according to rules: the initial trigger phase is divided by month, the functional imbalance phase by half-month, and the health problem manifestation phase by week, making the timeline of health problem development clearly identifiable. Subsequently, based on the pathophysiological characteristics of each phase, the module automatically associates corresponding types of data with each time node: the initial trigger phase is associated with lifestyle data, environmental data, and device sensor data; the functional imbalance phase is associated with clinical testing data, omics data, and immune factor data; and the health problem manifestation phase is associated with clinical diagnostic data, symptom record data, and core physiological indicator data, ensuring accurate correspondence between data and nodes at each phase. Meanwhile, the module has a built-in relationship mining engine that automatically identifies causal, correlation, and temporal relationships between graph nodes. It determines the relationship type based on preset rules. For causal relationships, factor B must appear within a 95% confidence interval after factor A appears, and it must still appear when there is no interference. For correlation relationships, the Pearson correlation coefficient must be ≥0.6 and the p-value must be <0.05. For temporal relationships, factor A must appear at least one data collection period earlier than factor B and there must be no reverse cases. The module also labels the association strength value for each relationship edge, with a causal relationship strength of 1.0, a correlation relationship strength equal to the corresponding correlation coefficient value, and a temporal relationship strength of 0.8. Finally, it generates a health issue time-series knowledge graph with a clear structure and rigorous logic, providing solid graph support for data anchoring and coupling analysis.

[0050] Data anchoring and coupling analysis module:

[0051] This module includes a data anchoring unit and a coupling analysis unit, which work together to complete the association between target data and the graph, as well as multi-factor coupling analysis. The data anchoring unit has a built-in time-series-graph anchoring algorithm. It receives the time-series knowledge graph output by the time-series graph construction module and the target data online device sensor time-series data and offline clinical test data output by the data integration module. The algorithm calculates the fusion anchoring strength between the target data and each node of the time-series knowledge graph. The mathematical expression of the time-series-graph anchoring algorithm is: ,in: Let be the fusion anchoring strength between the i-th type of target data and the j-th node of the temporal knowledge graph at time t; This is the time decay dynamic factor; Similarity is calculated by fusing target data and map features; For graph nodes Path importance weights; The temporal synergistic gain factor is defined as follows: 'i' represents the category number of the target data, taking values ​​of 1 or 2. 'i=1' corresponds to device sensing time-series data, and 'i=2' corresponds to clinical testing data. 'j' represents the node number in the temporal knowledge graph associated with the i-th category of target data. Based on the time dimension, different types of target data are precisely anchored to corresponding graph nodes and associated factors, achieving a deep association between data and the knowledge graph, allowing subsequent analysis to be based on precisely matched data. The coupling analysis unit first determines the scope of health influencing factors participating in the coupling analysis: lifestyle factors, environmental factors, immune factors, and metabolic indicators. It obtains the fusion anchoring strength data of the graph nodes corresponding to each influencing factor from the data anchoring unit, and extracts the anchoring strength of various influencing factors corresponding to each node according to the time dimension. Subsequently, the unit analyzes the effects of multiple influencing factors based on preset coupling effect judgment thresholds. When multiple factors act together, the increase in anchoring strength is greater than 20% of the sum of the individual effects of each factor, it is judged as a synergistic effect. When the increase is greater than 50% of the maximum anchoring strength of each factor acting alone but less than 20% of the sum, it is judged as a superposition effect. When the anchoring strength of the corresponding nodes of other factors decreases by more than 10% after the action of one factor, it is judged as an inhibitory effect. At the same time, the unit quantifies the coefficients of various coupling effects: synergistic effect coefficient 1.2-1.5, superposition effect coefficient 1.0-1.2, and inhibitory effect coefficient 0.5-0.9. Finally, the coupling effect type coefficients and anchoring strength change data of each node are organized in chronological order, and the time-series coupling analysis results are output, clearly presenting the mutual influence patterns among multiple factors, providing key analytical basis for root cause mining.

[0052] Marker mining and confidence calculation module:

[0053] The module comprises a marker mining unit and a confidence calculation unit, enabling the identification of personalized dynamic early warning markers and the assessment of candidate root cause confidence. The marker mining unit, based on time-series knowledge graphs and coupling analysis results, automatically searches for recurring device data patterns and clinical characteristic patterns before key turning points in health problems, forming candidate personalized dynamic early warning markers. The unit then initiates a two-level verification process. The first level verifies the markers in the target object's own historical time-series data, checking if the co-occurrence frequency is ≥80% and the consecutive occurrences are ≥3 times. The second level retrieves anonymized historical case data of 500 similar health problems from a medical and health big data sharing platform to verify if the early warning accuracy of the candidate markers is ≥75%. These two levels of verification ensure that the final identified personalized dynamic early warning markers accurately reflect the development trend of health problems. The confidence calculation unit incorporates a dynamic root cause confidence algorithm, combining information such as data anchoring strength, coupling coefficient, marker early warning effectiveness, and historical data matching degree to calculate the confidence of candidate health root causes. The mathematical expression of the dynamic root cause confidence algorithm is: ,in: The final confidence level of the k-th candidate root cause; Adaptive weights for root cause path coverage; , , For dynamic balancing weights; Let be the fusion anchoring strength between the i-th type of target data and the j-th node of the temporal knowledge graph at time t; Let k be the coupling coefficient of the root cause k at the j-th map node of the i-th type of target data; The integral represents the multi-stage coupling effect of the root cause k. The time of initiation of the root cause; This is the current analysis time; The effectiveness of early warning based on dynamic markers corresponding to root cause k; denoted as the historical data matching degree of root cause k; i represents the category index of the target data, i can be 1 or 2, i=1 corresponds to device sensing time series data, i=2 corresponds to clinical test data; j represents the index of the node in the time series knowledge graph associated with the i-th category of target data. This represents the total number of nodes in the time-series knowledge graph that participate in the anchoring calculation of the i-th type of target data corresponding to root cause k. The root cause candidate set is generated by sorting the nodes by confidence level, so that the root cause results output later have a clear priority order, providing a core basis for the result output.

[0054] Results Output and Ecosystem Linkage Module:

[0055] The module receives the root cause candidate set time-series knowledge graph and personalized dynamic early warning marker information output by the marker mining and confidence calculation module. It sorts and organizes the data by confidence level to form a standardized root cause localization report, including root cause name, confidence level, correlation data, supporting coupling analysis, etc., ensuring clear, comprehensive, and standardized results. The module connects to the medical and health big data sharing platform, health information system, and cloud platform through standardized interfaces to achieve real-time synchronization of results data, ensuring that all relevant platforms can obtain consistent root cause analysis information in a timely manner. Simultaneously, the module links online and offline intelligent diagnosis and treatment ecosystems. Online, it connects to health management apps and remote medical platforms to push root cause analysis reports and personalized health intervention suggestions, including dietary, exercise, and lifestyle adjustments, to target individuals. Offline, it connects with community health service centers, specialist hospitals, outpatient clinics, and physical examination institutions to trigger subsequent health monitoring, expert consultation, intervention guidance, and other services for target individuals. This constructs a closed-loop intelligent diagnosis and treatment ecosystem encompassing data collection, root cause analysis, intervention execution, and feedback, facilitating accurate root cause localization and efficient intervention for health problems.

[0056] In summary, this system is compatible with the aforementioned root cause localization methods. Through the collaborative operation of five core modules, it comprehensively collects multi-source health data and completes standardized integration and health problem identification. Based on historical cases and disease patterns, it constructs a well-structured and clearly defined temporal knowledge graph. The temporal-graph anchoring algorithm and coupling analysis module are used to achieve data anchoring and multi-factor analysis. The marker mining and confidence calculation module outputs a precise set of root causes. The results are then output and connected to multiple platforms through the ecosystem linkage module, linking online and offline medical resources to build a closed-loop service. This provides efficient and standardized system support for root cause localization of health problems such as prehypertension, ensuring accurate and controllable localization and effective implementation of intervention measures.

[0057] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A multi-factor coupled root cause localization method based on the formation path of health problems, characterized in that, The specific steps of this method are as follows: S1, Data Integration and Problem Identification: By acquiring online device sensor time-series data, offline clinical test data, lifestyle data and health assessment data of the target object through the medical and health big data sharing platform, after standardization and timestamp alignment, combined with the disease diagnosis standard library of the health information system, the current health problems of the target object are identified and structured descriptive information is generated. S2, Temporal Graph Construction: Based on the occurrence and development patterns of health problems, and combined with the historical temporal case data of the medical and health big data sharing platform, the identified health problems are modeled in stages to construct a health problem temporal knowledge graph containing multiple time-dimensional knowledge subgraphs; S3, Data Anchoring and Coupling Analysis: The time-series-graph anchoring algorithm is applied to calculate the fusion anchoring strength between the target data and the nodes of the time-series knowledge graph. The target data is anchored to the corresponding nodes and related factors according to the time dimension. Based on the anchoring results, time-series coupling analysis is performed on multiple health influencing factors. S4, Marker Mining and Confidence Calculation: Search for recurring device data patterns and clinical feature patterns before key turning points of health problems in the time-series knowledge graph, define personalized dynamic early warning markers, apply the root cause confidence dynamic algorithm to calculate the confidence of candidate health root causes, and generate a root cause candidate set. S5, Results Output and Ecosystem Linkage: Output the root cause localization results of health problems in order of confidence, link them with the corresponding time-series knowledge graph and personalized dynamic early warning marker information, and connect them with the medical and health big data sharing platform, health information system and cloud platform in a standardized format to link the online and offline intelligent diagnosis and treatment ecosystem.

2. The multi-factor coupled root cause localization method based on the health problem formation path according to claim 1, characterized in that, In step S1, the online device sensor time-series data obtained through the medical and health big data sharing platform includes heart rate variability, 24-hour ambulatory blood pressure, blood glucose fluctuation curve, exercise duration, exercise intensity and sleep structure data; offline clinical test data includes laboratory test reports, imaging examination results, doctor's diagnosis records and medical records; lifestyle data includes dietary preferences, smoking and drinking history, work and rest patterns and exercise habits records; and health assessment data includes physical examination reports, past medical history and chronic disease follow-up records.

3. The multi-factor coupled root cause localization method based on the health problem formation path according to claim 1, characterized in that, In step S1, the disease diagnostic standard library of the health information system includes the World Health Organization disease classification standards, the chronic disease diagnosis guidelines issued by the National Health Commission, clinical pathway specifications, symptom judgment standards, laboratory indicator reference ranges, disease duration requirements, and imaging characteristic standards. The laboratory indicator reference ranges are divided into specific numerical intervals according to age, gender, and physiological stage. The symptom judgment standards specify the symptom manifestation form, frequency of onset, and duration requirements.

4. The multi-factor coupled root cause localization method based on the health problem formation path according to claim 1, characterized in that, In step S2, the formation path of the health problem temporal knowledge graph is divided into an initial triggering timeline stage, a functional imbalance timeline stage, and a health problem manifestation timeline stage. The initial triggering timeline stage corresponds to 9-12 months before the manifestation of the health problem, the functional imbalance timeline stage corresponds to 3-8 months before the manifestation of the health problem, and the health problem manifestation timeline stage corresponds to 0-2 months before the health problem is identified. The initial triggering stage is divided into time nodes by month, the functional imbalance stage is divided into time nodes by half-month, and the health problem manifestation stage is divided into time nodes by week. Each time node in the initial triggering stage is associated with lifestyle data, environmental data, and device sensor data. Each time node in the functional imbalance stage is associated with clinical test data, omics data, and immune factor data. Each time node in the health problem manifestation stage is associated with clinical diagnosis data, symptom record data, and core physiological indicator data.

5. The multi-factor coupled root cause localization method based on the health problem formation path according to claim 1, characterized in that, In step S2, the relationships between the nodes of the time-series knowledge graph include causal relationships, correlation relationships, and temporal relationships. The causal relationship satisfies that after factor A appears, factor B will inevitably appear within a 95% confidence interval, and B will still appear when there are no other interfering factors. The correlation relationship satisfies that the Pearson correlation coefficient between factors is ≥0.6 and the P value is <0.

05. The temporal relationship satisfies that the occurrence time of factor A is at least one data collection cycle earlier than that of factor B, and there are no reverse occurrence cases. All relationship edges are accompanied by a correlation strength value, the causal relationship strength is 1.0, the correlation relationship strength is the corresponding correlation coefficient value, and the temporal relationship strength is 0.

8.

6. The multi-factor coupled root cause localization method based on the health problem formation path according to claim 1, characterized in that, In step S3, the mathematical expression of the time-map anchoring algorithm is: ,in: Let be the fusion anchoring strength between the i-th type of target data and the j-th node of the temporal knowledge graph at time t; This is the time decay dynamic factor; Similarity is calculated by fusing target data and map features; For graph nodes Path importance weights; is the temporal collaborative gain factor; i represents the category number of the target data, i takes the value of 1 or 2, i=1 corresponds to the device sensing time series data, i=2 corresponds to the clinical test data; j represents the index of the node in the temporal knowledge graph associated with the i-th category of target data.

7. The multi-factor coupled root cause localization method based on the health problem formation path according to claim 1, characterized in that, In step S4, the specific steps for performing time-series coupling analysis on multiple health influencing factors based on anchoring results are as follows: First, determine the scope of health influencing factors participating in the coupling analysis, including lifestyle factors, environmental factors, immune factors, and metabolic index factors. Then, extract the fusion anchoring strength data of the time-series knowledge graph nodes corresponding to each health influencing factor. Select the anchoring strength of each type of influencing factor corresponding to each node in sequence according to the time dimension. Set the coupling effect judgment threshold. When multiple influencing factors work together, the increase in the anchoring strength of the corresponding node is greater than 20% of the total anchoring strength when each type of factor works alone, and it is judged as a synergistic effect. When the increase in anchoring strength of the corresponding node after the superposition of multiple influencing factors is greater than 50% of the maximum anchoring strength when each factor acts alone and less than 20% of the sum of the anchoring strengths when each factor acts alone, it is determined to be a superposition effect. When the anchoring strength of the nodes corresponding to other influencing factors decreases by more than 10% after a certain influencing factor takes effect, it is determined to be an inhibitory effect. Subsequently, the coefficients of various coupling effects were quantified. The coefficients of synergistic effects ranged from 1.2 to 1.5, the coefficients of superposition effects ranged from 1.0 to 1.2, and the coefficients of inhibition effects ranged from 0.5 to 0.

9. Finally, the coupling effect types, coefficients, and anchoring strength changes of each node were organized in chronological order to complete the time-series coupling analysis and output the analysis results.

8. The multi-factor coupled root cause localization method based on the health problem formation path according to claim 1, characterized in that, In step S4, the personalized dynamic early warning marker needs to undergo two levels of verification. The first level is verified in the target object's own historical time series data, which must meet the requirement that the co-occurrence frequency before the key turning point is ≥80% and the number of consecutive occurrences is ≥3 times. The second level is verified by retrieving anonymized historical case data of 500 similar health problems from the medical and health big data sharing platform, which must meet the requirement that the early warning accuracy rate is ≥75%. If both levels of verification meet the standards, it is officially determined as a personalized dynamic early warning marker. If either verification fails, the feature pattern is searched again and the verification process is repeated, up to a maximum of 3 times.

9. The multi-factor coupled root cause localization method based on the health problem formation path according to claim 1, characterized in that, In step S4, the mathematical expression of the root cause confidence dynamic algorithm is: ,in: The final confidence level of the k-th candidate root cause; Adaptive weights for root cause path coverage; , , For dynamic balancing weights; Let be the fusion anchoring strength between the i-th type of target data and the j-th node of the temporal knowledge graph at time t; Let k be the coupling coefficient of the root cause k at the j-th map node of the i-th type of target data; The integral represents the multi-stage coupling effect of the root cause k. The time of initiation of the root cause; This is the current analysis time; The effectiveness of early warning based on dynamic markers corresponding to root cause k; denoted as the historical data matching degree of root cause k; i represents the category index of the target data, i can be 1 or 2, i=1 corresponds to device sensing time series data, i=2 corresponds to clinical test data; j represents the index of the node in the time series knowledge graph associated with the i-th category of target data. This represents the total number of nodes in the temporal knowledge graph that participate in the anchoring calculation of the i-th type of target data corresponding to root cause k.

10. A multi-factor coupled root cause localization system based on the path formation of health problems, the system being applicable to the multi-factor coupled root cause localization method based on the path formation of health problems as described in any one of claims 1-9, characterized in that, The system includes: Data integration and problem identification module: It is used to acquire online device sensor time-series data, offline clinical test data, lifestyle data and health assessment data of target objects through the medical and health big data sharing platform, perform standardized integration and timestamp alignment operations, and identify health problems and generate structured description information by combining the disease diagnosis standard library of the health information system. The time-series graph construction module is used to build a health problem time-series knowledge graph that includes multiple time-dimensional knowledge subgraphs, based on the occurrence and development patterns of health problems and combined with historical time-series case data from the medical and health big data sharing platform. Data Anchoring and Coupling Analysis Module: This module is used to calculate the fusion anchoring strength between target data and time-series knowledge graph nodes using the time-series-graph anchoring algorithm. It anchors the target data to the corresponding nodes and related factors according to the time dimension, and performs time-series coupling analysis on multiple health influencing factors based on the anchoring results. The marker mining and confidence calculation module is used to search for recurring device data patterns and clinical feature patterns before key turning points of health problems in the time-series knowledge graph, define personalized dynamic early warning markers and complete two-level verification, apply the root cause confidence dynamic algorithm to calculate the confidence of candidate health root causes and generate a root cause candidate set. Results Output and Ecosystem Linkage Module: This module outputs the root cause localization results of health problems in order of confidence, links them to the corresponding time-series knowledge graph and personalized dynamic early warning marker information, and connects them to the medical and health big data sharing platform, health information system and cloud platform in a standardized format, linking the online and offline intelligent diagnosis and treatment ecosystem.

Images