Health data processing method, system, device, and program product

Abstract

The application is suitable for the field of artificial intelligence and causal inference technology, and provides a health data processing method, system, device and program product. The method comprises the following steps: acquiring multi-source health data of a user; constructing an individual-level causal graph of the user based on the multi-source health data, and identifying confounding factors; adjusting the confounding factors based on the individual-level causal graph, and performing counterfactual reasoning to calculate individualized evaluation results under different processing schemes, and generating processing simulation results; collecting actual processing decisions made according to the processing simulation results and health feedback data generated after the execution; and based on the health feedback data, performing online adaptive incremental updating on the individual-level causal graph. Through the establishment of a closed loop from feedback collection to model updating, the application realizes the self-adaptation, continuous optimization and high timeliness of the model, and improves the consistency and confidence of the evaluation results.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence and causal inference technology, and in particular relates to a health data processing method, system, device and program product. Background Technology

[0002] In the field of health information processing, accurate assessment of users' health-related indicators is a prerequisite for personalized information support. Existing technologies, such as some machine learning-based assessment models, have begun to explore the use of multi-source data for analysis. However, these technologies generally suffer from the following drawbacks: the relationship between the model and actual application is open-loop. These models are typically static after deployment. While their assessment results can provide a reference for decision-making, the actual effectiveness data of the processing solutions cannot be automatically and in real-time absorbed by the model for iterative optimization. When new application evidence emerges or user group characteristics drift, the static model gradually ages, and the quality of its assessment results declines. Updating the model often requires offline, full retraining, which is not only time-consuming and labor-intensive but also leads to service interruptions, failing to meet the high timeliness and continuity requirements of application scenarios. Summary of the Invention

[0003] This application provides a health data processing method, system, device, and program product, aiming to solve the technical problems in the prior art where the causal evaluation model is static and open-loop, unable to self-optimize, and gradually ages, leading to a decline in the quality of evaluation results, as well as offline model updates causing service interruptions, failing to meet the application scenarios' requirements for high timeliness and continuity.

[0004] In a first aspect, embodiments of this application provide a health data processing method, the method comprising: Acquire users' multi-source health data; Based on the multi-source health data, an individual-level causal graph of the user is constructed, and confounding factors in the individual-level causal graph are identified; The identified confounding factors are adjusted based on the individual-level causal graph, and counterfactual reasoning is performed based on the adjusted individual-level causal graph to calculate individualized evaluation results under different treatment schemes and generate treatment simulation results. Collect actual processing decisions made based on the processing simulation results and health feedback data generated after the actual processing decisions are executed; Based on the health feedback data, the individual-level causal graph is updated online adaptively and incrementally.

[0005] In one possible implementation of the first aspect, constructing the user's individual-level causal graph based on the multi-source health data includes: The continuous variables in the multi-source health data are discretized to obtain discrete variables; The discrete variables are subjected to feature filtering to obtain the filtered discrete variables; Conditional independence tests are performed on the selected discrete variables to construct the initial skeleton of the individual-level causal graph; The edges of the initial skeleton are oriented to form a directed acyclic graph representing the causal relationships of the user, and the directed acyclic graph is used as the individual-level causal graph.

[0006] In one possible implementation of the first aspect, the counterfactual reasoning based on the adjusted individual-level causal graph to calculate individualized evaluation results under different treatment schemes and generate treatment simulation results includes: For each processing scheme, the processing node corresponding to the processing scheme is set to the value of the processing scheme in the adjusted individual-level causal graph; Using do-calculus or structural causal model, calculate the posterior probability distribution of target result nodes that have a causal relationship with the processing node in the individual-level causal graph after the processing node is set; Perform causal path analysis on the individual-level causal graph to determine the feature contribution score; Based on the posterior probability distribution, determine the evaluation difference compared with the untreated scenario where the processing node takes the value of a preset baseline, and the confidence interval for the processing scheme. The assessment difference, the confidence interval, and the feature contribution score are used as the individualized assessment results.

[0007] In one possible implementation of the first aspect, the online adaptive incremental update includes: Quantify the information gain of the health feedback data; The local subgraph to be updated in the individual-level causal graph is determined based on the information gain. The causal relationship parameters between nodes in the updated local subgraph are updated using a time-weighted method.

[0008] In one possible implementation of the first aspect, quantifying the information gain of the health feedback data and determining the local subgraph to be updated in the individual-level causal graph based on the information gain includes: Calculate the Kullback-Leibler divergence of the probability distribution of the target result node associated with the health feedback data before and after the health feedback data is collected, and use the Kullback-Leibler divergence as the information gain; When the information gain is greater than a preset divergence threshold, the local subgraph to be updated in the individual-level causal graph is determined based on the information gain.

[0009] In one possible implementation of the first aspect, determining the local subgraph to be updated in the individual-level causal graph based on the information gain includes: Centered on the target result node associated with the health feedback data; A radius is determined to define the update range, and the radius is positively correlated with the information gain; Nodes and edges whose shortest path hop count to the center is less than or equal to the radius are designated as local subgraphs to be updated.

[0010] In one possible implementation of the first aspect, the time-weighted parameter update of the causal relationship parameters between nodes in the local subgraph to be updated includes: The time-weighted learning rate is determined based on the time interval between the current time and the time when the health feedback data was generated. The causal relationship parameters between nodes in the local subgraph to be updated are updated using the time-weighted learning rate.

[0011] Secondly, embodiments of this application provide a health data processing system, including: The data acquisition module is used to acquire users' multi-source health data; The causal graph construction module is used to construct an individual-level causal graph for the user based on the multi-source health data, and to identify confounding factors in the individual-level causal graph. The reasoning module is used to adjust the identified confounding factors based on the individual-level causal graph, and to perform counterfactual reasoning based on the adjusted individual-level causal graph to calculate individualized evaluation results under different processing schemes and generate processing simulation results. The feedback acquisition module is used to collect the actual processing decisions made based on the processing simulation results and the health feedback data generated after the actual processing decisions are executed. The incremental update module is used to perform online adaptive incremental updates to the individual-level causal graph based on the health feedback data.

[0012] Thirdly, embodiments of this application provide a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the health data processing method described in any one of the first aspects above.

[0013] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, implements the health data processing method described in any one of the first aspects.

[0014] Fifthly, embodiments of this application provide a computer program product that, when run on a computer device, causes the computer device to execute the health data processing method described in any one of the first aspects.

[0015] This application's embodiments, by constructing individual-level causal graphs and identifying and adjusting confounding factors, distinguish between causal and correlational relationships between variables, avoiding spurious associations caused by confounding factors, and making the final simulation results more robust and reliable. By performing counterfactual reasoning, not only is a risk score provided, but the differences in expected effects from different treatment options can also be clearly simulated and compared, providing a quantitative reference for professional decision-making. Furthermore, by establishing a closed loop from feedback collection to model updates and performing online adaptive incremental updates, the model can intelligently filter meaningful information and perform efficient, local iterations, thereby continuously learning and evolving from practical applications without interrupting service.

[0016] It is understood that the beneficial effects of the second to fifth aspects mentioned above can be found in the relevant descriptions in the first aspect mentioned above, and will not be repeated here. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a health data processing method provided in an embodiment of this application; Figure 2 This is a schematic diagram illustrating the construction of an individual-level causal graph in an embodiment of this application; Figure 3 This is a schematic diagram illustrating the principle of simulating the counterfactual processing scheme in the embodiments of this application; Figure 4 This is a schematic diagram of the data interaction of the incremental update module in an embodiment of this application; Figure 5 This is a hardware topology diagram of an edge computing deployment in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of the computer device provided in the embodiments of this application. Detailed Implementation

[0019] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0020] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0021] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0022] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0023] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0024] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0025] Figure 1A schematic flowchart of a health data processing method provided in an embodiment of this application is shown.

[0026] S101, acquires multi-source health data of users.

[0027] Multi-source health data refers to a collection of data related to a user's health status from different channels and in various formats, including but not limited to imaging data, gene sequencing data, biochemical indicators, and historical processing records. This data collectively constitutes a comprehensive description of a user's health status and forms the basis for personalized analysis.

[0028] In the embodiments of this application, by aggregating data from different sources such as image features, genetic information, biochemical indicators and historical processing records, a more accurate picture of a user's overall health can be drawn. This is the basis for improving model accuracy, which is different from traditional methods that rely on a single data source.

[0029] S102, construct the user's individual-level causal graph based on the multi-source health data, and identify confounding factors in the individual-level causal graph.

[0030] Individual-level causal graphs are directed acyclic graphs constructed for specific users. Their nodes represent health-related variables (such as genes, biochemical indicators, treatment plans, and outcomes), and directed edges represent causal relationships between variables. This graph aims to reveal deep causal connections between variables, rather than simple statistical correlations. Unlike general population models, individual-level causal graphs capture user-specific causal structures, providing a precise causal reasoning basis for personalized assessment.

[0031] Confounding factors are variables that affect both treatment and outcome nodes. If these variables are not controlled, they can lead to spurious associations between treatments and outcomes, biasing causal effect estimations. For example, smoking history influences both a user's preference for a particular treatment and the prognosis of the disease, making it a typical confounding factor. Identifying and adjusting for confounding factors in individual-level causal graphs is a crucial step in ensuring the reliability of subsequent counterfactual inference results.

[0032] In this embodiment, by constructing an individual-level causal graph, the intrinsic mechanisms of interaction between variables can be expressed in a structured form. Identifying and subsequently adjusting confounding factors aims to eliminate interfering factors that may lead to spurious associations, thereby solving the problem of unreliable evaluation results due to confounding bias in the prior art. Through the above operations, the true causal chain can be extracted from complex data relationships, rather than remaining at the level of superficial statistical correlation.

[0033] S103, the identified confounding factors are adjusted based on the individual-level causal graph, and counterfactual reasoning is performed based on the adjusted individual-level causal graph to calculate individualized evaluation results under different processing schemes and generate processing simulation results.

[0034] Adjusting for identified confounding factors refers to eliminating the interference of confounding factors on the causal effect between processing nodes and result nodes by using methods such as backdoor criteria or hierarchical analysis before performing counterfactual inference, thereby ensuring the unbiasedness of causal estimation.

[0035] Counterfactual reasoning refers to the reasoning process that simulates how outcome variables will change under assumed conditions (such as the application of a certain treatment).

[0036] Individualized evaluation results refer to quantitative indicators of the expected effects of different treatment schemes for a specific user, including but not limited to: treatment effect size, confidence interval, feature contribution score, etc.

[0037] Processing simulation results refers to organizing and presenting individualized evaluation results in an understandable form to assist decision-makers in selecting alternatives.

[0038] In this embodiment, this step involves providing a quantitative prediction of "what if...". It is not simply an assessment of current risk, but rather a simulation of how the expected outcome (the target outcome node) would change after different treatment options (i.e., interventions at the treatment nodes in the causal graph). This counterfactual reasoning provides decision-makers with a basis for quantitative comparisons between different options, achieving a leap from risk assessment to effect simulation.

[0039] S104, Collect the actual processing decisions made based on the processing simulation results and the health feedback data generated after the actual processing decisions are executed.

[0040] In this context, actual treatment decision refers to the specific treatment plan that professionals ultimately select and implement for the user after referring to the results of the treatment simulation, such as choosing a certain drug, surgical method, or intervention.

[0041] Health feedback data refers to subsequent data related to the user's health status obtained through follow-up, monitoring, or examination after the actual treatment decision is implemented. This includes, but is not limited to: the actual status of the target outcome (such as disease progression or survival status), new biochemical indicator test values, new imaging examination results, and adverse event records. Health feedback data is a key basis for verifying the accuracy of model predictions and driving iterative updates to the model.

[0042] S105, based on the health feedback data, perform online adaptive incremental updates to the individual-level causal graph.

[0043] Online adaptive incremental update refers to the ability to receive new feedback data in real time without interrupting system service, and dynamically adjust the scope and intensity of updates based on the value of the data, thereby performing local and efficient iterative optimization of the causal graph model. "Online" signifies uninterrupted service, "adaptive" represents the intelligence of the update, and "incremental" represents the efficiency of the update.

[0044] In this embodiment, unlike traditional offline, full-scale updates, the online update process is real-time, local, and intelligent. It can assess the information value of new feedback data and adaptively adjust the scope and intensity of updates accordingly, thereby efficiently integrating new knowledge into existing models without interrupting service. Through this closed-loop feedback and update mechanism, the model is continuously optimized over time and with data accumulation, ensuring its long-term effectiveness.

[0045] Optionally, S102 constructs an individual-level causal graph for the user based on the multi-source health data, and identifies confounding factors in the individual-level causal graph, including: Step a1: Discretize the continuous variables (such as blood test values) in the multi-source health data to obtain discrete variables.

[0046] In this embodiment, continuous variables in multi-source health data are discretized and transformed into categorical variables to facilitate subsequent structure learning. For example, continuous variables such as lesion size in imaging features and detection values ​​in biochemical indicators are discretized using equal-frequency binning or binning methods based on clinical thresholds. For instance, tumor volume is divided into three categories: "less than 1 cm", "1 cm-3 cm", and "greater than 3 cm", and CEA index is divided into three categories: "below normal", "normal", and "above normal".

[0047] Step a2: Perform feature filtering on the discrete variables to obtain the filtered discrete variables.

[0048] In this embodiment, feature selection is performed on discrete variables to remove redundant or irrelevant variables, thereby reducing computational complexity and noise. The system employs a feature selection method based on mutual information or an L1 regularization method to remove redundant variables whose correlation with the target result node is lower than a preset threshold (e.g., 0.05), as well as noise variables that have no significant correlation with any other variables.

[0049] Step a3: Perform conditional independence tests on the selected discrete variables to construct the initial skeleton of the individual-level causal graph.

[0050] In this embodiment, a series of conditional independence tests are performed on the selected discrete variables to construct an initial skeleton representing the relationships between the variables. The system uses a PC algorithm to perform pairwise conditional independence tests on the selected variable set. For any two variables X and Y, given a variable set Z, if X and Y are conditionally independent, it is determined that there is no edge between X and Y; otherwise, the edge is retained. By traversing all variable pairs, an undirected graph representing the relationships between the variables is finally obtained as the initial skeleton. The conditional independence test can use a chi-square test or a G2 test based on mutual information, and the significance level can be set to 0.05.

[0051] Step a4: Orient the edges of the initial skeleton to form a directed acyclic graph representing the causal relationships of the user, and use the directed acyclic graph as the individual-level causal graph.

[0052] In this embodiment, specific orientation rules are used to orient the edges of the initial skeleton, forming a directed acyclic graph (DAG) representing user causal relationships, which serves as the final individual-level causal graph. The system uses V-structure orientation rules and a scoring function-based orientation method to assign directions to the edges in the initial skeleton. A V-structure refers to a structure of the form X→Z←Y, where there is no edge between X and Y, and X and Y are related when Z is conditional. By identifying such structures and combining them with scoring functions such as the Bayesian information criterion for global optimization, the resulting directed acyclic graph is the individual-level causal graph. The orientation process follows acyclic constraints to ensure that there are no directed cycles in the graph.

[0053] Optionally, S103 adjusts the identified confounding factors based on the individual-level causal graph, and performs counterfactual inference based on the adjusted individual-level causal graph to calculate individualized evaluation results under different treatment schemes, generating treatment simulation results, including: Step b1: For each processing scheme, set the processing node corresponding to the processing scheme to the value of the processing scheme in the adjusted individual-level causal graph.

[0054] In this context, a treatment node refers to a node in an individual-level causal graph that represents an interventionizable variable, such as a medication regimen or treatment method. Setting a treatment node to the value of a treatment regimen is called an intervention operation in causal inference, denoted by do(T=t). Its purpose is to simulate a scenario where a certain treatment is actively applied, rather than passively observed.

[0055] In this embodiment, the system, after adjusting for confounding factors, for each candidate treatment (such as gefitinib treatment or conventional chemotherapy), forcibly sets the value of the treatment node T to the value corresponding to that treatment. This operation is achieved by cutting off all causal edges pointing to the treatment node, so that the value of the treatment node is no longer affected by other variables, thereby simulating the effect of actively selecting the treatment in practical applications. For example, for the gefitinib treatment, the system executes do(T=gefitinib), fixing the treatment node to that value.

[0056] Step b2: Using do-calculus or structural causal model, calculate the posterior probability distribution of the target result node that has a causal relationship with the processing node in the individual-level causal graph after the processing node is set.

[0057] In this context, the target outcome node refers to the node representing the final outcome of concern in the individual-level causal graph, such as disease progression time, survival status, and degree of symptom relief. The posterior probability distribution refers to the probability distribution of the target outcome node taking various possible values ​​under given intervention conditions, reflecting the expected effect of the treatment plan.

[0058] In the embodiments of this application, the intervention distribution is calculated using do-calculus or structural causal modeling. Do-calculus is a classic method in the field of causal inference, which obtains an unbiased causal effect estimate by eliminating confounding biases; structural causal modeling obtains the counterfactual distribution of the target outcome node by modifying the structural equation of the treatment node and forward propagating. Both of these methods are mature existing technologies, and their specific calculation processes will not be elaborated here. For example, the system uses do-calculus to calculate that under gefitinib intervention, the posterior probability of the 1-year progression-free survival node is 0.7 for a value of no progression and 0.3 for a value of progression.

[0059] Step b3: Perform causal path analysis on the individual-level causal graph to determine the feature contribution score.

[0060] Among them, the feature contribution score refers to the quantitative value of the influence of each upstream feature node on the target outcome node in the individual-level causal graph, which is used to explain which factors contribute the most to the prediction result. Causal path analysis refers to tracing the causal path from the feature node to the outcome node along the directed edges in the causal graph and calculating the magnitude of the effect transmitted along the path.

[0061] In this embodiment, a causal path-based attribution method is used to calculate feature contribution scores. Commonly used methods include average causal effect decomposition and path tracing algorithms. By decomposing the total causal effect along the causal path into the contributions of each path, the contribution score of each feature node to the outcome node is obtained. For example, the system performs path analysis on the individual-level causal graph and calculates that the contribution score of EGFR mutation status to 1-year progression-free survival is 0.6, the contribution score of nodule volume is 0.2, and the contribution score of smoking history is 0.2, indicating that EGFR mutation status is the most important factor affecting the user's prognosis. This score, together with the assessment difference and confidence interval, constitutes the individualized assessment result, enhancing the interpretability of the result.

[0062] Step b4: Based on the posterior probability distribution, determine the evaluation difference compared with the untreated scenario where the processing node takes the value of the preset baseline, and the confidence interval for the processing scheme.

[0063] The assessment difference refers to the difference between the treatment plan and the untreated baseline scenario in terms of the expected effect at the target outcome node. It is usually expressed as the average causal effect, quantifying the expected benefits or risks brought by the treatment. The confidence interval is an estimate of the reliability range of the assessment difference, reflecting the uncertainty of the prediction results.

[0064] In this embodiment, the system compares the posterior probability distribution after intervention with the posterior probability distribution of the baseline scenario without intervention (e.g., do(T=0), indicating no treatment or a standard control regimen), and calculates the difference between the two as the assessment difference. For example, for the gefitinib treatment regimen, the system calculates a difference of 0.3 (0.7-0.4) in its 1-year progression-free survival probability compared to the baseline scenario, indicating that this regimen can improve the progression-free survival probability by 30% compared to the baseline. Simultaneously, the system calculates the confidence interval of this assessment difference, such as [0.22, 0.38], using the Bootstrap resampling method or Bayesian posterior-based interval estimation, to reflect the reliability of the prediction. The assessment difference and the confidence interval together constitute the core quantitative indicators of the individualized assessment results.

[0065] Step b5: The assessment difference, the confidence interval, and the feature contribution score are used as the individualized assessment results.

[0066] In this embodiment, the posterior probability distribution calculated in step b2, the feature contribution score determined in step b3, and the assessment difference and confidence interval calculated in step b4 are integrated to form a structured individualized assessment result. This result contains three core elements: the assessment difference quantifies the expected effect of the treatment regimen compared to the baseline; the confidence interval reflects the range of uncertainty in the predicted result; and the feature contribution score explains the key factors affecting the result. The system encapsulates these results into a processing simulation result report and pushes it to professionals via API or a visual interface as a quantitative reference for clinical decision-making. For example, for the gefitinib treatment regimen, the system outputs an assessment difference of 0.3, a confidence interval of [0.22, 0.38], and a feature contribution score of "EGFR mutation status: 0.6; nodule volume: 0.2; smoking history: 0.2," assisting professionals in understanding the predictive basis and making reasonable decisions.

[0067] Optional, S105 online adaptive incremental updates include: Step c1: Quantify the information gain of the health feedback data.

[0068] Information gain refers to the amount of information value that newly acquired health feedback data brings to the model, and is used to determine whether the data is worth using for model updates.

[0069] Specifically: Calculate the Kullback-Leibler divergence of the probability distribution of the target result node associated with the health feedback data before and after the health feedback data is collected, and use the Kullback-Leibler divergence as the information gain.

[0070] The Kullback-Leibler divergence (KL divergence) measures the difference between two probability distributions; a larger value indicates that the new data provides more information.

[0071] In this embodiment, the Kullback-Leibler divergence between the probability distribution of the target result node associated with the health feedback data before and after the data collection is calculated, and this divergence value is directly used as the information gain. The principle is that if new data drastically changes our perception of the result (probability distribution), then the KL divergence of this data is large, and the information gain is high. Furthermore, a preset divergence threshold is set; subsequent update steps are only triggered when the calculated information gain exceeds this threshold. This threshold-based triggering mechanism effectively filters out redundant data with insufficient information, avoids frequent invalid calculations, and ensures efficient utilization of system resources.

[0072] Step c2: Determine the local subgraph to be updated in the individual-level causal graph based on the information gain.

[0073] In this context, a local subgraph refers to a subgraph within an individual-level causal graph that comprises nodes and edges closely related to health feedback data and requiring parameter updates. By limiting the update scope to a local subgraph, rather than performing a global update on the entire causal graph, both the efficiency of incremental updates and service continuity are achieved.

[0074] Specifically: taking the target result node associated with the health feedback data as the center; determining a radius to define the update range, the radius being positively correlated with the information gain; and taking nodes and edges whose shortest path hop count to the center is less than or equal to the radius as local subgraphs to be updated.

[0075] In this embodiment, the process centers on the target result node associated with the health feedback data. First, a radius is determined to define the update range. The size of this radius is positively correlated with the information gain calculated in the previous step; that is, the larger the information gain, the larger the radius. Then, in the individual-level causal graph, the system delineates all nodes and edges whose shortest path hop count to the central node is less than or equal to this radius as the local subgraph to be updated. In this way, the abstract value of information gain is mapped to a specific, dynamically changing update region in the graph, achieving adaptive adjustment of the update range.

[0076] Step c3 involves updating the causal relationship parameters between nodes in the local subgraph to be updated using a time-weighted method.

[0077] Among them, causal relationship parameters refer to the quantitative indicators carried by each directed edge in the individual-level causal graph, used to represent the strength of the influence of the cause node on the result node, such as the magnitude of the causal effect, conditional probability, or regression coefficient. Time-series weighting refers to assigning different update weights according to the freshness of the data, with newer data having higher weights, enabling the model to dynamically adapt to the latest application practice changes.

[0078] Specifically: a time-weighted learning rate is determined based on the time interval between the current time and the time when the health feedback data was generated; the causal relationship parameters between nodes in the local subgraph to be updated are updated using the time-weighted learning rate.

[0079] Among them, the time-weighted learning rate refers to the update weight coefficient that is dynamically adjusted according to the freshness of the data. The longer the time interval, the smaller the learning rate, so that newer data occupies a higher weight in the update.

[0080] In this embodiment, a time-weighted learning rate is determined based on the time interval between the current time and the time the feedback data was generated, according to a preset monotonically decreasing function (e.g., an exponential decay function). This time-weighted learning rate decreases as the time interval increases. This function ensures that the longer the time interval (i.e., the older the data), the smaller the calculated learning rate. Then, the system uses this time-weighted learning rate to update the causal relationship parameters between nodes in the local subgraph. This ensures that the model can dynamically adapt to the latest application practice changes and maintain its timeliness.

[0081] In this embodiment, the information gain of the received health feedback data is first quantified, that is, the value of this new data to the model is assessed. Next, based on the calculated information gain, the range of local subgraphs in the individual-level causal graph that need to be updated is intelligently determined; the greater the information gain, the wider the potential impact. Finally, only within the determined local subgraph, a time-weighted parameter update is performed on the causal relationship parameters between the nodes to be updated, meaning newer data has a higher weight in the update. This achieves a precise, efficient, and adaptive update process, avoiding unnecessary recalculation of the entire model.

[0082] This application also provides a health data processing system. The system includes a data acquisition module, a causal graph construction module, an inference module, a feedback collection module, and an incremental update module. The data acquisition module performs the data acquisition steps described in the preceding methods. The causal graph construction module performs the steps of constructing an individual-level causal graph and identifying confounding factors. The inference module performs the steps of adjusting confounding factors and performing counterfactual inference. The feedback collection module performs the steps of collecting actual processing decisions and health feedback data. The incremental update module performs the core online adaptive incremental update step. These modules work together to realize the complete closed-loop processing flow described in this application.

[0083] The technical solution of this application will be described in more detail below with reference to the accompanying drawings and specific implementation scenarios.

[0084] The method of this application is specifically implemented in a scenario involving lung health data processing. First, the system acquires a user's multi-source health data through standard interfaces in the information system, such as FHIR (Fast Medical Interoperability Resource) and DICOM (Digital Imaging and Communications in Medicine). This data includes nodule volume extracted from chest CT images (a continuous variable), EGFR gene mutation status from gene sequencing reports (a categorical variable), CEA biochemical index values ​​from blood tests (a continuous variable), and the user's previous chemotherapy records. This data collectively forms the basis for personalized analysis.

[0085] Next, the system enters the cause-effect graph construction phase. For example... Figure 2 As shown, the process of constructing an individual-level causal graph uses multi-source health data such as image features and gene mutations as input, employs a PC algorithm for causal structure learning, constructs an initial skeleton through conditional independence testing and orients the edges, and combines propensity score analysis to identify confounding factors, ultimately outputting an individual-level causal graph (i.e., ...). Figure 2 The diagram shows an individual causal graph, which uses nodes and edges to represent causal relationships between variables and labels confounding factors.

[0086] After constructing an individual-level causal graph, the system performs counterfactual simulations. For example... Figure 3 As shown, counterfactual reasoning, through intervention settings on processing nodes, uses DO-calculus to calculate the posterior probability distribution of the target outcome node, and performs causal path analysis to determine the feature contribution score, ultimately outputting the evaluation difference and confidence interval. Suppose a comparison is needed between gefitinib treatment and traditional chemotherapy. After adjusting for the confounding factor of smoking history, the system forcibly sets the treatment node to gefitinib and chemotherapy respectively in the causal graph. Then, DO-calculus is used to calculate the posterior probability of the target outcome node, 1-year progression-free survival (PFS), under these two interventions. The simulation results may show that with the gefitinib regimen, the risk of progression within 1 year is 0.3, while with the chemotherapy regimen it is 0.6. Simultaneously, the system will also output the confidence interval for each regimen and the contribution score of the key features leading to this difference, forming a detailed processing simulation result report.

[0087] The simulation results were sent to the professionals' workstations. After reviewing the report, the professionals ultimately selected the lower-risk gefitinib regimen for the user. Following this, the system entered the feedback collection phase, automatically collecting data at predetermined time points (e.g., one year later) to confirm that the user had not experienced disease progression, and recording this event as a health feedback data point.

[0088] Finally, the system triggers an online adaptive incremental update mechanism. After receiving the feedback on the aforementioned outcome event, the incremental update module first quantifies the information gain of the feedback data. Specifically, the module calculates the Kullback-Leibler divergence of the probability distribution of the target outcome node's 1-year PFS before and after the feedback was collected, and uses this divergence value as the information gain. The calculation formula is as follows: (Formula 1) in, For new feedback events The resulting information gain; The Kullback-Leibler divergence function; Target result node before receiving new evidence The prior probability distribution; In conjunction with new evidence The target result node after The posterior probability distribution; This represents the target outcome node in the cause-effect graph (e.g., a state of no progress for 1 year). For newly received health feedback events. When the calculated information gain is greater than a preset divergence threshold. At that time, the system will initiate subsequent update steps.

[0089] Once the update is triggered, the system dynamically calculates a radius to define the update range based on information gain, with the target result node as the center. This radius is positively correlated with information gain. The calculation formula is as follows: (Formula 2) in, The influence domain search radius is dynamically calculated (based on the shortest path hop count in the graph). This serves as a basic search radius, which can be set to 1. The information gain is calculated using Formula 1; and All of these are adjustable hyperparameters. This is the radius expansion coefficient, which controls the degree to which the information gain amplifies the search radius; it can be set to 0.5. This is the gain scaling factor, which adjusts the sensitivity of the information gain to the logarithmic function; it can be set to 1. It is the natural logarithm function; This is a rounding function. The system will round up if the shortest path hop count to the center node is less than or equal to the radius. The nodes and edges are used as the local subgraph to be updated. .

[0090] Finally, in the defined local subgraph Within the system, the causal relationship parameters between nodes are updated using a time-weighted approach. The system first determines the time-weighted learning rate based on a preset exponential decay function and the time interval between the current time and the time when the feedback data was generated. Then, this learning rate is used to update any edge in the subgraph. causal relationship parameters The update is performed using the following formula: (Formula 3) in, and The edges are shown below: before and after the update. causal relationship parameters; Based solely on new evidence The estimated parameters; It is a time-weighted learning rate, and its specific form can be as follows: ,in, It is the base learning rate. It is the time decay coefficient. and These are the current time and the timestamp of the new evidence, respectively. The entire update process only applies to a local subgraph, taking only 0.8 seconds, during which the system continues to provide services, achieving zero-downtime updates.

[0091] In another embodiment, the method of this application can employ different algorithm models to adapt to different application scenarios. For example, in a scenario involving the processing of liver-related health data, the system acquires the user's abdominal MRI images (extracting the maximum diameter of a specific region), TP53 gene mutation status, alpha-fetoprotein (AFP) levels, and historical PD-1 medication records. In this scenario, the causal graph construction module can employ the Causal Forest algorithm. This algorithm can directly estimate the heterogeneity treatment effect (ITE), and is particularly suitable for discovering different responses of different individuals to the same treatment regimen. By analyzing the data, the model may discover that for this specific user, the PD-1 combined with targeted therapy regimen has a positive causal effect on the decrease in AFP levels and tumor shrinkage. Based on this, the inference module can simulate and compare the combination regimen with the PD-1 monotherapy regimen, predicting that after 6 months, the tumor shrinkage rate under the combination regimen can reach 65%, while the monotherapy regimen is only 40%. When the professional adopts the combination regimen, the system collects feedback data showing an actual shrinkage rate of 62% after 6 months. The incremental update module can utilize this feedback to incrementally update the causal forest model. For example, by evaluating information gain, it can identify the set of decision trees most relevant to the feedback and use a time-weighted mechanism to locally adjust the leaf node values ​​of these trees. This process demonstrates that the framework of the method in this application can be applied to different algorithmic models.

[0092] Furthermore, the method of this application can be deployed on different hardware architectures to meet diverse application needs. For example... Figure 5As shown, the method of this application can be deployed in an edge computing architecture, which includes core components such as an edge computing box, data source input, result output, and a cloud system. In one implementation of edge computing deployment, the method of this application is encapsulated as software and deployed on a high-performance edge computing box. This edge box is physically deployed within the application department or on a mobile terminal. The user's raw health data, such as newly generated CT image files and gene sequencing VCF files, can be directly input into the edge box. The edge box can execute the entire computational process from data acquisition, causal discovery to treatment scheme simulation locally. The calculated simulation report can be directly pushed to the workstations of local professionals, enabling real-time decision support. For example, in the test, a single complete causal discovery + treatment scheme simulation process took only 1.2 seconds on the edge box, and the CPU utilization was less than 50%, demonstrating that its computational latency and resource consumption are suitable for real-time computing scenarios such as bedside. Simultaneously, the edge box can communicate asynchronously with the backend system deployed in the cloud via an internal 5G network or local area network for data backup and global model analysis, demonstrating that the method of this application can be deployed under different system architectures.

[0093] To more clearly illustrate the details of data interaction during the update process of this application, please refer to [link / reference]. Figure 4 .like Figure 4 As shown, the incremental update module interacts with the FHIR server via a gRPC communication link, and with the Neo4j graph database (i.e., Figure 4The Neo4j graph (shown) interacts with each other via an MQTT communication link, with the entire interaction process completed within 1 second. In one specific implementation, after the inference module generates the processing simulation results, the system encapsulates them into an Observation resource conforming to the FHIR standard and writes it to the FHIR server of the Information System (HIS) via an HTTP POST request. Simultaneously, a result summary is pushed to the front end of the professional workstation via API calls. When the professional performs processing and records the outcome event, this event is also standardized into an FHIR resource and stored. The incremental update module deployed in the backend subscribes to the update events of the FHIR server in real time via a gRPC communication link. When a new outcome event FHIR resource is created, the FHIR server immediately sends a notification to the incremental update module via gRPC. Upon receiving the notification, the incremental update module pulls detailed feedback data and initiates its internal multi-stage adaptive update algorithm. After completing information gain calculation, local subgraph delineation, and temporal weighting parameter calculation, a set of concise update instructions is generated. Subsequently, the module publishes these update instructions to the Neo4j graph database, which serves as the backend for causal graph storage, via an MQTT communication link. The Neo4j database executes commands to atomically update the attributes of specified nodes and edge weights in the graph. The entire signaling interaction and computation update process is completed within 1 second. The above process illustrates a closed-loop data interaction scheme to achieve the online adaptive incremental update of this application.

[0094] It should be noted that the preset divergence threshold, basic search radius, radius expansion coefficient, gain scaling coefficient, basic learning rate, time decay coefficient, and time-weighted learning rate mentioned in the embodiments of this application are all exemplary values ​​for implementing the technical solution of this application. In practical applications, those skilled in the art can reasonably set or dynamically adjust these parameters according to the specific health data type, application scenario, computing resources, and accuracy requirements to optimize model performance. The adjustability of the above parameters does not affect the scope of the core technical solution claimed in this application.

[0095] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0096] It should be understood that the information interaction and execution process between the above modules are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0097] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0098] This application also provides a computer device, which includes: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the steps in any of the above method embodiments.

[0099] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0100] This application provides a computer program product that, when run on a computer device, enables the computer device to execute the steps described in the various method embodiments above.

[0101] Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Figure 6 As shown, the computer device of this embodiment includes: at least one processor 60 ( Figure 6 (Only one is shown in the diagram), memory 61, and computer program 62 stored in said memory 61 and executable on said at least one processor 60, wherein said processor 60 executes said computer program 62 to implement the steps in any of the above embodiments of the health data processing methods.

[0102] The computer device may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art will understand that... Figure 6 The examples of computer devices are merely examples and do not constitute a limitation on computer devices. They may include more or fewer components than shown in the illustration, or combinations of certain components, or different components. For example, they may also include input / output devices, network access devices, etc.

[0103] The processor 60 may be a Central Processing Unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor.

[0104] In some embodiments, the memory 61 may be an internal storage unit of the computer device, such as a hard drive or memory. In other embodiments, the memory 61 may be an external storage device of the computer device, such as a plug-in hard drive, smart media card (SMC), secure digital card (SD), flash card, etc. Furthermore, the memory 61 may include both internal and external storage units of the computer device. The memory 61 is used to store the operating system, applications, bootloader, data, and other programs, such as the program code of the computer program. The memory 61 can also be used to temporarily store data that has been output or will be output.

[0105] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include at least: any entity or system capable of carrying computer program code to a system / computer device, a recording medium, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, such as a USB flash drive, a portable hard drive, a magnetic disk, or an optical disk.

[0106] In the above embodiments, the descriptions of each embodiment have different focuses. For parts that are not described in detail or recorded in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0107] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0108] In the embodiments provided in this application, it should be understood that the disclosed systems / computer devices and methods can be implemented in other ways. For example, the system / computer device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0109] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0110] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method for processing health data, characterized in that, include: Acquire users' multi-source health data; Based on the multi-source health data, an individual-level causal graph of the user is constructed, and confounding factors in the individual-level causal graph are identified; The identified confounding factors are adjusted based on the individual-level causal graph, and counterfactual reasoning is performed based on the adjusted individual-level causal graph to calculate individualized evaluation results under different treatment schemes and generate treatment simulation results. Collect actual processing decisions made based on the processing simulation results and health feedback data generated after the actual processing decisions are executed; Based on the health feedback data, the individual-level causal graph is updated online adaptively and incrementally.

2. The method according to claim 1, characterized in that, The construction of the user's individual-level causal graph based on the multi-source health data includes: The continuous variables in the multi-source health data are discretized to obtain discrete variables; The discrete variables are subjected to feature filtering to obtain the filtered discrete variables; Conditional independence tests are performed on the selected discrete variables to construct the initial skeleton of the individual-level causal graph; The edges of the initial skeleton are oriented to form a directed acyclic graph representing the causal relationships of the user, and the directed acyclic graph is used as the individual-level causal graph.

3. The method according to claim 1, characterized in that, The process of performing counterfactual reasoning based on the adjusted individual-level causal graph to calculate individualized evaluation results under different treatment schemes and generate treatment simulation results includes: For each processing scheme, the processing node corresponding to the processing scheme is set to the value of the processing scheme in the adjusted individual-level causal graph; Using do-calculus or structural causal model, calculate the posterior probability distribution of target result nodes that have a causal relationship with the processing node in the individual-level causal graph after the processing node is set; Perform causal path analysis on the individual-level causal graph to determine the feature contribution score; Based on the posterior probability distribution, determine the evaluation difference compared with the untreated scenario where the processing node takes the value of a preset baseline, and the confidence interval for the processing scheme. The assessment difference, the confidence interval, and the feature contribution score are used as the individualized assessment results.

4. The method according to claim 3, characterized in that, The online adaptive incremental update includes: Quantify the information gain of the health feedback data; The local subgraph to be updated in the individual-level causal graph is determined based on the information gain. The causal relationship parameters between nodes in the updated local subgraph are updated using a time-weighted method.

5. The method according to claim 4, characterized in that, The process of quantifying the information gain of the health feedback data and determining the local subgraph to be updated in the individual-level causal graph based on the information gain includes: Calculate the Kullback-Leibler divergence of the probability distribution of the target result node associated with the health feedback data before and after the health feedback data is collected, and use the Kullback-Leibler divergence as the information gain; When the information gain is greater than a preset divergence threshold, the local subgraph to be updated in the individual-level causal graph is determined based on the information gain.

6. The method according to claim 4, characterized in that, The step of determining the local subgraph to be updated in the individual-level causal graph based on the information gain includes: Centered on the target result node associated with the health feedback data; A radius is determined to define the update range, and the radius is positively correlated with the information gain; Nodes and edges whose shortest path hop count to the center is less than or equal to the radius are designated as local subgraphs to be updated.

7. The method according to claim 4, characterized in that, The time-weighted parameter update of the causal relationship parameters between nodes in the local subgraph to be updated includes: The time-weighted learning rate is determined based on the time interval between the current time and the time when the health feedback data was generated. The causal relationship parameters between nodes in the local subgraph to be updated are updated using the time-weighted learning rate.

8. A health data processing system, characterized in that, include: The data acquisition module is used to acquire users' multi-source health data; The causal graph construction module is used to construct an individual-level causal graph for the user based on the multi-source health data, and to identify confounding factors in the individual-level causal graph. The reasoning module is used to adjust the identified confounding factors based on the individual-level causal graph, and to perform counterfactual reasoning based on the adjusted individual-level causal graph to calculate individualized evaluation results under different processing schemes and generate processing simulation results. The feedback acquisition module is used to collect the actual processing decisions made based on the processing simulation results and the health feedback data generated after the actual processing decisions are executed. The incremental update module is used to perform online adaptive incremental updates to the individual-level causal graph based on the health feedback data.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 7.

10. A computer program product, characterized in that, Includes a computer program that, when run, implements the method as described in any one of claims 1 to 7.

Images