A big data-based whole-course management system and method

Abstract

The present application belongs to the technical field of whole course management, and particularly relates to a whole course management system and method based on big data. The present application can process text information according to a target disease query preset differential threshold through disease sensing processing, realizes the differentiation of the strictness of processing different clinical risk diseases in system design, provides a basic mechanism for balancing the efficiency and safety of data processing, takes the update amplitude exceeding the preset threshold as a quantitative judgment condition for triggering the overall reconstruction of the model, provides a clear and executable technical basis for whether the model is locally fine-tuned or completely innovated under different degrees of patient state changes, simulates and deduces the alternative schemes through the calling of the patient digital twin model, and selects the optimal scheme according to the quantitative score sorting to make the path adjustment change from relying on qualitative experience to a systematic optimization method based on quantitative simulation.

Description

Technical Field

[0001] This invention belongs to the field of whole-course disease management technology, specifically relating to a whole-course disease management system and method based on big data. Background Technology

[0002] Whole-course disease management refers to the continuous, collaborative, and personalized health management of patients throughout the entire cycle from disease prevention, early diagnosis, treatment intervention to rehabilitation follow-up, relying on information technology.

[0003] Existing technologies typically use uniform information extraction models and fixed confidence thresholds to process all text, making it difficult to differentiate the processing based on the varying accuracy requirements of different diseases. For example, processing medical records for "common cold" and "myocardial infarction" using the same standards can easily lead to unnecessary inefficiency in some low-risk diseases and introduce risks in high-risk diseases due to insufficiently stringent standards. Secondly, while existing technologies can build models using patient data, they lack a clear and quantifiable set of rules to determine when to completely overhaul the model rather than making minor adjustments. This can result in delayed model updates when the patient's condition undergoes fundamental changes. Furthermore, when management approaches need to be adjusted due to changes in the patient's condition, existing technologies often rely on expert experience or simple rules to select from a limited number of options, lacking a systematic method that can quickly and quantitatively simulate and compare the effects of multiple alternative adjustment plans based on the patient's current specific model. Summary of the Invention

[0004] The purpose of this invention is to provide a big data-based full-course management system and method. This system can process text information through disease-aware processing, based on preset differential thresholds for target diseases, thereby differentiating the severity of treatment for diseases with different clinical risks. This provides a fundamental mechanism for balancing the efficiency and security of data processing. Using "update magnitude exceeding a preset threshold" as a quantitative judgment condition to trigger a comprehensive model reconstruction provides a clear and executable technical basis for determining whether to perform local fine-tuning or complete overhaul of the model under different degrees of patient status changes. By calling the patient's digital twin model to simulate and deduce alternative solutions, and ranking them according to quantitative scores to select the optimal solution, the decision-making process transforms path adjustment from relying on qualitative experience to a systematic optimization method based on quantitative simulation.

[0005] The specific technical solution adopted by this invention is as follows: A big data-based approach to full-course disease management includes: Acquire multimodal time-series health data of patients throughout their entire disease course; The process of performing disease-aware processing on unstructured text data in the multimodal time-series health data includes: determining the target disease currently relevant to the patient; processing the unstructured text data through an information extraction model to obtain at least one medical entity and its corresponding confidence score; obtaining an automatic adoption threshold and a manual review threshold based on the target disease, wherein the automatic adoption threshold is greater than the manual review threshold; comparing the confidence score with the automatic adoption threshold and the manual review threshold; when the confidence score is greater than or equal to the automatic adoption threshold, directly adopting the medical entity as structured data; when the confidence score is less than the automatic adoption threshold but greater than or equal to the manual review threshold, marking the medical entity as requiring manual review; and when the confidence score is less than the manual review threshold, marking the medical entity as requiring detailed manual review. Based on the adopted structured data, construct and update a digital twin model that reflects the individual health status of the patient; Based on the digital twin model, the risk of patients at different stages of the disease is assessed, and a personalized full-course management path is generated. The personalized, full-course disease management pathway is executed, and the digital twin model is optimized based on the execution feedback. In a preferred embodiment, determining the patient's currently relevant target disease includes at least one of the following methods: Receive the disease identification designated by medical staff; Extract the diagnosed disease codes from the patient's electronic health record; Disease prediction is performed on the multimodal time-series health data using a disease classification model.

[0006] In a preferred embodiment, the step of obtaining the corresponding automatic adoption threshold and manual review threshold based on the target disease includes: Query the pre-configured disease-to-threshold mapping table to directly obtain a unique set of threshold parameters corresponding to the target disease. The threshold parameters include the automatic adoption threshold and the manual review threshold.

[0007] In a preferred embodiment, the pre-configured disease-to-threshold mapping table is set based on at least one of the following factors: The clinical risk level resulting from the misidentification of common medical entities under this disease; Statistical characteristics of the complexity and ambiguity of medical texts related to this disease; The distribution of accuracy rates for identifying medical entities related to this disease by the information extraction model in historical data.

[0008] In a preferred embodiment, an optimization step for automatically adopting a threshold is also included: The accuracy rate of automatically adopted decisions for the target disease within a preset time period was statistically analyzed. Based on a comparison between the accuracy of the automatic adoption decision and the preset target accuracy, the automatic adoption threshold is adjusted. Specifically, when the accuracy rate of automatically adopted decisions is lower than the preset target accuracy rate, the automatic adoption threshold is increased; when the accuracy rate of automatically adopted decisions is higher than the preset target accuracy rate, the automatic adoption threshold is decreased.

[0009] In a preferred embodiment, the adjustment of the automatic adoption threshold is based on the magnitude and persistent trend of the deviation between the accuracy of the automatic adoption decision and the preset target accuracy. When the accuracy of automatically adopted decisions continues to be lower than the target accuracy and the deviation widens, the automatic adoption threshold is increased by a significant amount. When the accuracy of automatically adopted decisions fluctuates around the target accuracy, the automatic adoption threshold is fine-tuned by a small margin.

[0010] In a preferred embodiment, the construction and updating of the digital twin model includes: Based on the adopted structured data, key feature parameters related to the patient's physiological state are extracted; Establish dynamic correlations among the key feature parameters to form a digital mapping of the individual patient's physiological state; Newly adopted structured data is continuously input into the digital mapping to update the key feature parameters and their correlations; When the update magnitude caused by new data exceeds a preset update threshold, a complete reconstruction of the digital mapping is triggered.

[0011] In a preferred embodiment, after generating a personalized full-course management pathway, the process further includes a dynamic adjustment step for the pathway: During the implementation of the pathway, the deviation between the patient's real-time health data and the expected goals of the pathway is continuously monitored; When the deviation exceeds the adaptive adjustment threshold, an instantaneous simulation of the digital twin model is triggered to simulate the effects of different adjustment schemes. Based on the simulation results, the goals, intensity, or frequency of interventions are dynamically adjusted in subsequent stages of the personalized full-course management pathway.

[0012] This invention also provides a big data-based full-course management system, which uses the above-mentioned big data-based full-course management method, including: The data acquisition module is used to acquire multimodal time-series health data of patients throughout the entire course of their disease; The data processing module is used to perform disease-aware processing on the unstructured text data in the multimodal time-series health data, including: determining the target disease currently related to the patient; processing the unstructured text data through an information extraction model to obtain at least one medical entity and its corresponding confidence score; obtaining the corresponding automatic adoption threshold and manual review threshold based on the target disease, wherein the automatic adoption threshold is greater than the manual review threshold; comparing the confidence score with the automatic adoption threshold and the manual review threshold; when the confidence score is greater than or equal to the automatic adoption threshold, directly adopting the medical entity as structured data; when the confidence score is less than the automatic adoption threshold but greater than or equal to the manual review threshold, marking the medical entity as requiring manual review; when the confidence score is less than the manual review threshold, marking the medical entity as requiring detailed manual review. The digital twin module is used to build and update a digital twin model that reflects the individual health status of a patient based on adopted structured data; The pathway management module is used to assess the patient's risk at different stages of the disease based on the digital twin model and generate a personalized full-course management pathway. An optimization module is used to execute the personalized full-course disease management path and optimize the digital twin model based on execution feedback.

[0013] And, an electronic device, the electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the big data-based full-course management method according to any one of claims 1 to 8.

[0014] The technical effects achieved by this invention are as follows: This invention, through disease-aware processing, can process text information based on preset differentiated thresholds for target diseases, thereby differentiating the severity of treatment for diseases with different clinical risks in the system design, providing a fundamental mechanism for balancing the efficiency and security of data processing. Secondly, in the maintenance of the digital twin model, "update magnitude exceeding a preset threshold" serves as a quantitative judgment condition triggering a comprehensive model reconstruction. This provides a clear and actionable technical basis for whether to make localized fine-tuning or complete overhaul of the model under different degrees of patient status changes. Furthermore, in the dynamic adjustment of the management path, by calling the patient's digital twin model to simulate and extrapolate alternative plans, and ranking them according to quantitative scores to select the optimal plan, the path adjustment shifts from relying on qualitative experience to a systematic optimization method based on quantitative simulation. This enables key aspects of the entire disease management process to possess clearer logic, more quantitative judgment, and more systematic decision-making capabilities. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of the method flow of the present invention; Figure 2 This is a schematic diagram of the system modules of the present invention; Figure 3 This is a schematic diagram of the electronic device structure of the present invention. Detailed Implementation

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0017] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0018] Secondly, the term "an embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in a preferred embodiment" appearing in different places throughout this specification does not necessarily refer to the same embodiment, nor is it a single embodiment or an embodiment selectively excluded from other embodiments.

[0019] Please see Figure 1 As shown, this invention provides a method for full-course disease management based on big data, including: S1. Obtain multimodal time-series health data of the patient throughout the entire course of the disease; In step S1, multimodal time-series health data throughout the patient's entire disease course is acquired, laying the foundation for subsequent analysis. The entire disease course refers to the complete health event chain from the disease risk period, early screening, diagnosis, treatment, rehabilitation to long-term follow-up. Multimodal refers to diverse data types, including structured diagnostic and treatment data from hospital information systems, continuous physiological signals from wearable devices, and unstructured text from patient self-reports. Time-series refers to these data being generated continuously or periodically along a timeline.

[0020] Specifically, collecting a complete set of data reflecting a patient's health status from multiple dimensions and in chronological order, which depicts the dynamic development of the disease, is the data foundation required to build and continuously update a patient's personal digital twin model.

[0021] S2. Perform disease-aware processing on the unstructured text data in the multimodal time-series health data, including: determining the target disease currently related to the patient; processing the unstructured text data through an information extraction model to obtain at least one medical entity and its corresponding confidence score; obtaining the corresponding automatic adoption threshold and manual review threshold based on the target disease, wherein the automatic adoption threshold is greater than the manual review threshold; comparing the confidence score with the automatic adoption threshold and the manual review threshold; when the confidence score is greater than or equal to the automatic adoption threshold, directly adopting the medical entity as structured data; when the confidence score is less than the automatic adoption threshold but greater than or equal to the manual review threshold, marking the medical entity as requiring manual review; when the confidence score is less than the manual review threshold, marking the medical entity as requiring detailed manual review.

[0022] In step S2, unstructured text data in the multimodal time-series health data undergoes disease-aware processing. Based on preset differential processing thresholds for each disease, intelligent hierarchical processing of text information is achieved, balancing extraction efficiency and clinical safety from a mechanistic perspective. Unstructured text data refers to text information without a predefined fixed format or data model. For example, a patient's description of symptoms and feelings such as "I've been feeling dizzy and weak for the past week, especially after meals" is an example. Disease-aware processing refers to data processing strategies adjusted and optimized according to the specific disease type (disease) the patient suffers from. The target disease refers to the specific disease that the patient currently needs to focus on managing and analyzing. It is determined through at least one of the following logics: extracting the primary diagnosis from the patient's current medical record; selecting the most active disease from the patient's historically diagnosed disease list; or receiving a disease identifier specified by medical staff. For example, for a patient with diabetes and hypertension, if the core of the current medical visit and data analysis is diabetes management, then diabetes is the current target disease. Information extraction models are conventional techniques used in this field to extract structured information from text. Their purpose is to automatically identify and extract meaningful medical information from unstructured text data. Unstructured text data refers to text information without a predefined fixed format or data model, such as doctors' medical records and patients' self-reported text. Medical entities refer to specific medical concept objects extracted from text, including but not limited to: diseases and diagnoses, symptoms and signs, examinations and tests, and drugs. Confidence score refers to a numerical value given by the information extraction model when outputting a medical entity, indicating the reliability of the extraction result. The higher the score, the more confident the model is in its extraction. Automatic adoption threshold refers to a high confidence threshold pre-set for the "target disease." When the confidence score of the model's extracted result is greater than or equal to this threshold, the system considers it highly reliable and can automatically convert it into structured data without manual intervention. Structured data refers to data formats that can be directly recognized, classified, and calculated by computers. For example, extracting "dizziness" from the patient's complaint of "dizziness for three days" as a symptom entity and storing it in the "complaint symptoms" field of a database. The manual review threshold refers to a low confidence threshold set in advance for the "target disease". It defines the boundary that needs to attract human attention. The automatic adoption threshold is greater than the manual review threshold.

[0023] Specifically, this processing is completed within the text intelligent governance engine. This engine first identifies the disease requiring immediate attention (target disease) using a disease identifier. Then, the text is fed into a medical entity recognizer for processing. This medical entity recognizer, or information extraction model, is implemented using natural language processing technology. It employs a known deep learning-based named entity recognition model, such as a pre-trained language model based on the Transformer architecture, like BERT. This model undergoes domain adaptation and fine-tuning on a large corpus of medical texts to enable it to recognize and standardize medical entities. The model outputs a confidence score between 0 and 1 for each identified entity. Next, the engine's policy controller queries a pre-defined disease-threshold mapping table based on the target disease to obtain the corresponding automatic adoption threshold and manual review threshold. These thresholds are pre-set based on the performance of the information extraction model, clinical risk level, and text feature complexity in historical data for the target disease, after statistical analysis, and support dynamic optimization. Finally, the engine's routing distributor automatically determines the processing path based on the threshold range in which the entity's confidence score falls: It compares the confidence score with the automatic adoption threshold and the manual review threshold. When the confidence score is greater than or equal to the automatic adoption threshold, the medical entity is directly adopted as structured data. When the confidence score is less than the automatic adoption threshold but greater than or equal to the manual review threshold, the medical entity is marked as requiring manual review, prompting professionals to conduct rapid checks and confirmations to balance efficiency and accuracy. When the confidence score is less than the manual review threshold, the medical entity is marked as requiring detailed manual review, demanding more effort from professionals for careful judgment and correction. The correct results confirmed after manual review and detailed review are used as high-quality labeled data for incremental training and confidence calibration of the information extraction model, and can be used to iteratively optimize the thresholds for relevant diseases. The entire process ensures that the structured data entering subsequent processes has both high automation and high reliability. Therefore, differentiated information confidence judgment standards are adopted according to the specific diseases of patients to achieve a balance between safety and efficiency.

[0024] The determination of the patient's current relevant target disease includes at least one of the following methods: Receive the disease identification designated by medical staff; Extract the diagnosed disease codes from the patient's electronic health record; Disease prediction is performed on the multimodal time-series health data using a disease classification model.

[0025] Specifically, a hierarchical decision-making logic is employed to determine the target disease, ensuring the accuracy and clinical relevance of the results: First, the disease identifier specified by medical staff in real-time through the interactive interface is prioritized to fully respect clinical decision-making; if not, the primary diagnosis code is extracted from the patient's electronic health record for this visit; if no clear primary diagnosis is found, the active diagnosis from the most recent record is selected; if none of the above can be obtained, a disease classification model is invoked to analyze and predict the patient's recent data. This disease classifier is a model obtained through supervised learning using a large amount of patient historical data labeled with diseases. It can extract fusion features from the patient's recent key structured data, semantic features of the textual complaints and present medical history, and summary indicators of specific physiological signals, and map them to the most probable disease category. For example, for a patient with diabetes and hypertension, if the core of the current visit and data analysis is diabetes management, then diabetes is the current target disease. This hierarchical decision-making logic forms a complete, reliable, and automated closed loop for determining the target disease, from clear clinical specification to extraction from existing records and intelligent prediction.

[0026] Secondly, the automatic adoption threshold and manual review threshold obtained based on the target disease include: Query the pre-configured disease-to-threshold mapping table to directly obtain a unique set of threshold parameters corresponding to the target disease. The threshold parameters include the automatic adoption threshold and the manual review threshold.

[0027] Specifically, a mapping table between diseases and thresholds is maintained. Once a patient's current relevant target disease is determined, the system uses this target disease as the key to query the mapping table and obtain a unique set of automatically adopted thresholds and manually reviewed thresholds preset for that disease. This mapping table is pre-configured by domain experts and data scientists based on factors such as the historical model performance and clinical risks of each disease.

[0028] The pre-configured disease-to-threshold mapping table is set based on at least one of the following factors: The clinical risk level resulting from the misidentification of common medical entities under this disease; Statistical characteristics of the complexity and ambiguity of medical texts related to this disease; The distribution of accuracy rates for identifying medical entities related to this disease by the information extraction model in historical data.

[0029] Specifically, the threshold setting combines clinical medicine, natural language processing, and data analysis. For example, for diseases with high clinical risk levels, an extremely high automatic adoption threshold is set to ensure that key symptoms are not incorrectly adopted. For diseases with high complexity and ambiguity in expression, quantitative assessment is performed by analyzing the contextual variability of relevant terms and the number of synonyms in historical text corpora, and a more conservative threshold is set accordingly. Simultaneously, the historical accuracy distribution is continuously monitored, and more aggressive thresholds are set for diseases with stable model performance to improve efficiency.

[0030] Then, it also includes optimization steps for the automatic adoption threshold: The accuracy rate of automatically adopted decisions for the target disease within a preset time period was statistically analyzed. Based on a comparison between the accuracy of the automatic adoption decision and the preset target accuracy, the automatic adoption threshold is adjusted. Specifically, when the accuracy rate of automatically adopted decisions is lower than the preset target accuracy rate, the automatic adoption threshold is increased; when the accuracy rate of automatically adopted decisions is higher than the preset target accuracy rate, the automatic adoption threshold is decreased.

[0031] Specifically, a performance monitoring and optimization service is run regularly. To obtain the accuracy of automated decision adoption, automatically adopted medical entities are periodically sampled, and these sampled entities are then submitted to medical staff for result verification. This service statistically analyzes the sampled and verified automatically adopted entities for a specific target disease over a past period, calculating the proportion of those proven correct, which is used as the accuracy rate of automated decision adoption for that period. This accuracy rate is compared with a preset, desired target accuracy rate. This comparison and adjustment process constitutes a standard control loop: the accuracy rate of automated decision adoption serves as the feedback signal, the automated adoption threshold is the control object, and the target accuracy rate is the setpoint. If the accuracy rate of automated decision adoption is low, it indicates that the current automated adoption standard is too lenient; therefore, the automated adoption threshold is increased to make the standard more stringent, thereby improving the reliability of future data. Conversely, the automated adoption threshold is decreased to improve automation efficiency. This facilitates maintaining the performance within the ideal range over the long term.

[0032] The adjustment of the automatic adoption threshold is based on the magnitude and continuous trend of the deviation between the accuracy of the automatic adoption decision and the preset target accuracy. When the accuracy of automatically adopted decisions continues to be lower than the target accuracy and the deviation widens, the automatic adoption threshold is increased by a significant amount. When the accuracy of automatically adopted decisions fluctuates around the target accuracy, the automatic adoption threshold is fine-tuned by a small margin.

[0033] Specifically, the performance monitoring and optimization service internally maintains a historical sequence of the accuracy of automatic decision adoption for each disease. It analyzes not only current deviations but also trends in these deviations. When it detects that the accuracy of automatic decision adoption is consistently below the target and the deviation is widening, it indicates deteriorating system performance, requiring strong intervention. Therefore, significant adjustments are made, such as setting the threshold adjustment step size to a larger base value to quickly correct the deviation. When the accuracy of automatic decision adoption fluctuates normally around the target value, only minor adjustments are made, such as using a smaller base step size, to avoid oscillations caused by overreacting to noise and maintain stability. This makes the optimization process smoother and more effective.

[0034] S3. Based on the adopted structured data, construct and update a digital twin model that reflects the individual health status of the patient; In step S3, based on the adopted structured data, a digital twin model reflecting the individual patient's health status is constructed and updated. Quantitative rules are used to determine when a complete model reconstruction is necessary, providing a clear technical basis for its precise evolution. The adopted structured data refers to the structured data obtained in step S2, the specific steps of which are as follows: Based on the adopted structured data, key feature parameters related to the patient's physiological state are extracted; Establish dynamic correlations among the key feature parameters to form a digital mapping of the individual patient's physiological state; Newly adopted structured data is continuously input into the digital mapping to update the key feature parameters and their correlations; When the update magnitude caused by new data exceeds a preset update threshold, a complete reconstruction of the digital mapping is triggered.

[0035] Specifically, using the structured data adopted in step S2, a personalized, computable, and simulable virtual model—a digital twin model—is created for the patient. This process is completed within the digital twin engine. First, the engine extracts a set of key feature parameters characterizing the patient's core physiological state from the patient's temporal structured data using a feature extractor. These parameters form the foundation for subsequent modeling. Then, through the engine's model builder, a suitable, parameterizable computational model is selected as a template based on domain knowledge of the target disease. For example, for chronic disease management, a pharmacokinetic / pharmacodynamic model based on physiology can be used as the basic template. Relationships are established using the patient's historical data, and through model fitting, personalized parameters in the template model are determined, resulting in a workable digital mapping specific to the patient. This mapping is continuously fine-tuned by the engine's online learner as new data is input to track the slow changes in the patient's physiological state. The update magnitude of the model is quantified through continuous monitoring. The update magnitude is measured by calculating the deviation between the prediction error of the digital twin model on newly incorporated structured data and the baseline value of the prediction error of the model on the historical validation dataset before the update. For example, when monitoring the model's prediction error on new data, if this monitoring indicator exceeds a preset update threshold for several consecutive periods, it indicates that the patient's condition may have undergone significant and rapid changes. At this point, simple fine-tuning is insufficient to maintain the model's accuracy. The engine triggers the model refactorer, which uses all data from the most recent period to re-execute the complete model building process, thereby upgrading the model version to maintain an accurate portrayal of the patient's current condition.

[0036] S4. Based on the digital twin model, assess the patient's risk at different stages of the disease and generate a personalized full-course management path; In step S4, based on the digital twin model, the patient's risk at different stages of the disease is assessed, and a personalized full-course management path is generated. Multiple adjustment schemes are simulated and deduced by calling the digital twin model, and the optimal path is selected based on quantitative scores, thus achieving dynamic and quantitative adjustment of the path. A full-course management path refers to a systematic management plan tailored to the patient, covering the entire process from disease onset to outcome.

[0037] Specifically, a digital twin model is used to quantitatively assess and plan for the patient's future health status. This assessment is phased, with different risk assessment logics built-in for different stages such as the acute phase, stable phase, and recovery phase. This risk assessment logic is implemented through specific algorithms. For example, for the stable phase, a gradient boosting tree model is used, with the key feature parameters output by the digital twin model as input and the output being a quantitative score of the control indicator achievement rate and the risk of long-term complications. For the recovery phase, a logistic regression model is used, outputting a probability score of functional recovery progress. The patient's current state is input into their digital twin model, and the above-mentioned built-in risk assessment logic is run, outputting a quantitative risk score and key indicator predictions as the assessment result. Based on this assessment result, a clinical knowledge base is accessed. This knowledge base stores disease management rules based on medical guidelines in the form of production rules, such as rules like "If the long-term complication risk score is greater than threshold A, then it is recommended to strengthen monitoring plan B." Through a rule engine, the assessment result is matched and inferred against the rules in the knowledge base to generate a preliminary set of intervention recommendations encompassing treatment, monitoring, rehabilitation, and follow-up dimensions. Finally, the proposed set of pathways is orchestrated and time-planned by a pathway optimizer that considers patient preferences, availability of medical resources, and time constraints. This pathway optimizer is implemented based on a constraint satisfaction problem solver, linear programming, or a heuristic rule engine to generate an executable, customized, full-course disease management pathway that is most feasible and yields the best expected results under given constraints.

[0038] The process of generating a personalized full-course disease management pathway also includes a dynamic adjustment step: During the implementation of the pathway, the deviation between the patient's real-time health data and the expected goals of the pathway is continuously monitored; When the deviation exceeds the adaptive adjustment threshold, an instantaneous simulation of the digital twin model is triggered to simulate the effects of different adjustment schemes. Based on the simulation results, the goals, intensity, or frequency of interventions are dynamically adjusted in subsequent stages of the personalized full-course management pathway.

[0039] Specifically, by defining the dynamism and adaptability of the entire disease management pathway, static plans are transformed into intelligent navigation. During pathway execution, the execution monitor compares the patient's actual data with the pre-set goals for each stage of the pathway in real time. If a significant and persistent deviation occurs, the execution monitor immediately sends an alert to the pathway adjustment decision-maker. Upon receiving the alert, the decision-maker quickly generates several feasible alternative adjustment plans based on the current situation. Then, the decision-maker calls the patient's digital twin model in parallel, sets different virtual input parameters for each plan, and commands the model to quickly simulate the evolution trajectory of key health indicators over a period of time after the implementation of the plan. Next, the evaluation function built into the decision-maker scores and ranks all simulation results based on principles such as "maximizing efficacy and minimizing risk," selecting the optimal plan.

[0040] The evaluation function operates as follows: First, it calculates the efficacy improvement score and risk increase score for each alternative adjustment plan based on the simulation results. The efficacy improvement score is quantified based on the closeness of the key health indicators output by the simulation to the ideal clinical goal; the risk increase score is quantified based on abnormal indicators or complication risk signals that appear during the simulation. Then, these two scores are assigned different weights according to their clinical importance and weighted together to calculate the total score for each plan. Finally, the total scores of all plans are compared and ranked, and the plan with the highest overall score is selected as the optimal adjustment plan.

[0041] Based on the parameters of this optimal solution, the relevant task instructions in the subsequent management path are modified, such as changing the medication dosage for next week, changing daily monitoring to three times a week, and adjusting the stage goal from "achieving blood glucose targets" to "avoiding hypoglycemia." This facilitates proactive and forward-looking optimization of the management plan based on real-time changes in the patient's condition.

[0042] S5. Execute the personalized full-course disease management path and optimize the digital twin model based on the execution feedback; In step S5, the personalized full-course management path is executed, and the digital twin model is optimized based on execution feedback. The digital twin model is continuously calibrated, forming a self-learning, continuously improving performance enhancement loop. Data generated during the execution process is used to optimize the digital twin model, facilitating continuous accumulation and improvement of full-course management during use.

[0043] Specifically, the generated personalized management pathways are synchronously distributed and driven through doctor workstations and patient terminal applications. During this process, the system continuously collects all new data related to pathway execution, particularly recording the actual health outcomes of patients following each intervention. These newly generated execution records, containing the correspondence between "interventions and actual outcomes," along with contextual information such as the time of occurrence, constitute feedback information for the system's learning.

[0044] This feedback is periodically collected and transmitted to the digital twin model building and maintenance process involved in step S3. The digital twin engine uses this feedback to calibrate the patient's personalized digital twin model, enabling the model to more accurately reflect the patient's true response characteristics to current interventions. Through this mechanism, a continuous optimization cycle is established: the model becomes more accurate with feedback, leading to more effective management paths; the execution of these new paths generates further feedback for further model optimization.

[0045] Please see Figure 2 A big data-based full-course management system, using the aforementioned big data-based full-course management method, includes: The data acquisition module is used to acquire multimodal time-series health data of patients throughout the entire course of their disease; The data processing module is used to perform disease-aware processing on the unstructured text data in the multimodal time-series health data, including: determining the target disease currently related to the patient; processing the unstructured text data through an information extraction model to obtain at least one medical entity and its corresponding confidence score; obtaining the corresponding automatic adoption threshold and manual review threshold based on the target disease, wherein the automatic adoption threshold is greater than the manual review threshold; comparing the confidence score with the automatic adoption threshold and the manual review threshold; when the confidence score is greater than or equal to the automatic adoption threshold, directly adopting the medical entity as structured data; when the confidence score is less than the automatic adoption threshold but greater than or equal to the manual review threshold, marking the medical entity as requiring manual review; when the confidence score is less than the manual review threshold, marking the medical entity as requiring detailed manual review. The digital twin module is used to build and update a digital twin model that reflects the individual health status of a patient based on adopted structured data; The pathway management module is used to assess the patient's risk at different stages of the disease based on the digital twin model and generate a personalized full-course management pathway. An optimization module is used to execute the personalized full-course disease management path and optimize the digital twin model based on execution feedback.

[0046] In the above, the data acquisition module is responsible for acquiring multimodal time-series health data throughout the patient's entire disease course, providing the system with a complete set of health data covering the patient's entire life cycle, multiple dimensions, and time sequence, laying the foundation for accurate analysis; the data processing module is responsible for performing disease-aware processing on the unstructured text data in the multimodal time-series health data, including: determining the target disease currently related to the patient; processing the unstructured text data through an information extraction model to obtain at least one medical entity and its corresponding confidence score; obtaining the corresponding automatic adoption threshold and manual review threshold based on the target disease, wherein the automatic adoption threshold is greater than the manual review threshold; comparing the confidence score with the automatic adoption threshold and the manual review threshold, and when the confidence score is greater than or equal to the automatic adoption threshold, directly adopting the medical entity as structured data; when the confidence score is less than the automatic adoption threshold but greater than or equal to the manual review threshold, marking the medical entity as requiring manual review; when When the confidence score is lower than the manual review threshold, the medical entity is marked as requiring detailed manual review. Based on a pre-defined differentiated processing threshold for each disease, intelligent hierarchical processing of text information is achieved, balancing extraction efficiency and clinical safety from a mechanistic perspective. The digital twin module is responsible for constructing and updating a digital twin model reflecting the individual patient's health status based on the adopted structured data. Quantitative rules determine when a complete model reconstruction is necessary, providing a clear technical basis for its precise evolution. The path management module is responsible for assessing the patient's risk at different stages of the disease based on the digital twin model and generating a personalized full-course management path. By calling the digital twin model to simulate and deduce various adjustment schemes, and selecting the best based on quantitative scores, dynamic and quantitative adjustments to the path are achieved. The optimization module is responsible for executing the personalized full-course management path and optimizing the digital twin model based on execution feedback, continuously calibrating the digital twin model, and driving the system to form a self-learning, continuously performance-enhancing closed loop.

[0047] Please see Figure 3 An electronic device, characterized in that: the electronic device comprises: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the big data-based full-course management method according to any one of claims 1 to 8.

[0048] The processor of the aforementioned electronic device can be a high-performance central processing unit (CPU) or graphics processing unit (GPU), and the memory can include storage devices such as random access memory (RAM), read-only memory (ROM), solid-state drive (SSD), or hard disk drive. In addition, the electronic device may also include an arithmetic unit, input devices, output devices, and a network interface. The arithmetic unit can be a logic unit used to perform various arithmetic and logical operations to assist the processor in completing complex data processing tasks. Input devices can include keyboards, mice, touch screens, etc., used to receive user input instructions and data. Output devices can include displays, printers, etc., used to display processing results and output reports. The network interface is used to enable network communication between the electronic device and other systems or devices for data exchange and remote monitoring.

[0049] To enable those skilled in the art to better understand the present invention, the following description is based on a hypothetical treatment scenario of a type 2 diabetes patient.

[0050] Patient Wang, 50 years old, was previously diagnosed with type 2 diabetes. Recently, he proactively reported through a health management application: "I have been feeling thirsty for the past few days, drinking more water than usual, needing to get up to urinate at night, and feeling lethargic during the day." In response to Mr. Wang's situation, the system of this invention began to operate. First, the system simultaneously acquired the diabetes diagnosis information recorded in Mr. Wang's electronic health record, as well as the self-report text he had just submitted.

[0051] In the text processing stage, the system identified "type 2 diabetes" as the target disease to be managed. The self-reported text was then sent to a medical entity recognizer for analysis. The medical entity recognizer identified three key symptom entities: "thirst," "polyuria," and "fatigue," and assigned confidence scores of 0.94, 0.89, and 0.92, respectively. The system then retrieved the preset disease-threshold mapping table based on the target disease and the preset processing standard for that disease: an automatic adoption threshold of 0.90 and a manual review threshold of 0.80. Based on this standard, the system automatically adopted the high-confidence entities "thirst" and "fatigue" as structured data, while temporarily storing the slightly lower-confidence entity "polyuria" and generating a pending task to prompt medical staff for quick verification. After reviewing the text context on the system interface, the medical staff confirmed the entity's validity and subsequently adopted it.

[0052] The system then integrated structured data such as Wang's recent blood glucose monitoring records and adopted symptom information. Based on this data, the digital twin engine used a fitting algorithm to build a personalized blood glucose response prediction model for Wang. This model can approximately simulate Wang's physiological characteristics; for example, it can estimate his possible postprandial blood glucose response to diet and exercise.

[0053] Based on this personalized blood glucose response prediction model, the system assessed Wang's current condition. The model simulation showed that his recent blood glucose control was poor, indicating a moderate risk of short-term hyperglycemia. Combining the assessment results with management rules from the clinical knowledge base, the system generated a preliminary management plan for Wang for the following week, including recommendations for a follow-up visit to the outpatient clinic, adjustments to recent monitoring frequency, and receiving relevant health education information through the application.

[0054] Mr. Wang began implementing the plan. Two days later, the system, through continuously collected monitoring data, discovered that his self-reported postprandial blood glucose levels had not improved as expected, showing a persistent deviation from the planned target. At this point, the system's dynamic adjustment mechanism was triggered. Based on the current situation, the adjustment decision module generated two alternative response plans: the first plan recommended immediately initiating an online consultation; the second plan was to immediately send the patient a set of enhanced dietary control guidelines and advise him to return to the outpatient clinic the following day as originally planned, bringing detailed blood glucose records.

[0055] To proactively assess which option was superior, the system simultaneously invoked Wang's digital twin model to rapidly simulate the potential effects of both options over the next few days. The simulation results showed that the second option performed better overall in terms of both expected blood sugar improvement and ease of implementation. Therefore, the system automatically adopted the second option and dynamically adjusted Wang's subsequent management tasks accordingly: detailed dietary adjustment suggestions were immediately sent to his mobile device, and the online follow-up check, originally scheduled for three days later, was moved forward to the following day, coinciding with his outpatient visit.

[0056] The new blood glucose monitoring data generated after this adjustment, and the correlation between "dietary guidance and blood glucose changes," were recorded by the system as an intervention feedback. This feedback information will be used subsequently to fine-tune and calibrate the parameters of Wang's personal digital twin model, enabling the model to more accurately reflect his individual characteristics and provide a more reliable basis for future decision-making.

[0057] The above is merely an illustrative application scenario used to illustrate the synergistic relationship between the various stages of the method of this invention. Those skilled in the art will understand that the full-course disease management method described in this invention can be realized by employing appropriate natural language processing models, computational modeling tools, and optimization algorithms.

[0058] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, apparatus, article, or method that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, apparatus, article, or method. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, apparatus, article, or method that includes that element.

[0059] The above description is merely a preferred embodiment of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention. Structures, devices, and operating methods not specifically described or explained in this invention are implemented according to conventional methods in the art unless otherwise specified or limited.

Claims

1. A big data-based method for full-course disease management, characterized in that: include: Acquire multimodal time-series health data of patients throughout their entire disease course; The process of performing disease-aware processing on unstructured text data in the multimodal time-series health data includes: determining the target disease currently relevant to the patient; processing the unstructured text data through an information extraction model to obtain at least one medical entity and its corresponding confidence score; obtaining an automatic adoption threshold and a manual review threshold based on the target disease, wherein the automatic adoption threshold is greater than the manual review threshold; comparing the confidence score with the automatic adoption threshold and the manual review threshold; when the confidence score is greater than or equal to the automatic adoption threshold, directly adopting the medical entity as structured data; when the confidence score is less than the automatic adoption threshold but greater than or equal to the manual review threshold, marking the medical entity as requiring manual review; and when the confidence score is less than the manual review threshold, marking the medical entity as requiring detailed manual review. Based on the adopted structured data, construct and update a digital twin model that reflects the individual health status of the patient; Based on the digital twin model, the risk of patients at different stages of the disease is assessed, and a personalized full-course management path is generated. The personalized full-course disease management path is executed, and the digital twin model is optimized based on the execution feedback.

2. The method for full-course disease management based on big data according to claim 1, characterized in that: The determination of the patient's current relevant target disease includes at least one of the following methods: Receive the disease identification designated by medical staff; Extract the diagnosed disease codes from the patient's electronic health record; Disease prediction is performed on the multimodal time-series health data using a disease classification model.

3. The method for full-course disease management based on big data according to claim 1, characterized in that: The automatic adoption threshold and manual review threshold obtained based on the target disease include: Query the pre-configured disease-to-threshold mapping table to directly obtain a unique set of threshold parameters corresponding to the target disease. The threshold parameters include the automatic adoption threshold and the manual review threshold.

4. The method for full-course disease management based on big data according to claim 3, characterized in that: The pre-configured disease-to-threshold mapping table is set based on at least one of the following factors: The clinical risk level resulting from the misidentification of common medical entities under this disease; Statistical characteristics of the complexity and ambiguity of medical texts related to this disease; The distribution of accuracy rates for identifying medical entities related to this disease by the information extraction model in historical data.

5. The method for full-course disease management based on big data according to claim 1, characterized in that: It also includes optimization steps for the automatic adoption threshold: The accuracy rate of automatically adopted decisions for the target disease within a preset time period was statistically analyzed. Based on a comparison between the accuracy of the automatic adoption decision and the preset target accuracy, the automatic adoption threshold is adjusted. Specifically, when the accuracy rate of automatically adopted decisions is lower than the preset target accuracy rate, the automatic adoption threshold is increased; when the accuracy rate of automatically adopted decisions is higher than the preset target accuracy rate, the automatic adoption threshold is decreased.

6. The method for full-course disease management based on big data according to claim 5, characterized in that: The adjustment of the automatic adoption threshold is based on the magnitude and continuous trend of the deviation between the accuracy of the automatic adoption decision and the preset target accuracy. When the accuracy of automatically adopted decisions continues to be lower than the target accuracy and the deviation widens, the automatic adoption threshold is increased by a significant amount. When the accuracy of automatically adopted decisions fluctuates around the target accuracy, the automatic adoption threshold is fine-tuned by a small margin.

7. The method for full-course disease management based on big data according to claim 1, characterized in that: The construction and updating of the digital twin model includes: Based on the adopted structured data, key feature parameters related to the patient's physiological state are extracted; Establish dynamic correlations among the key feature parameters to form a digital mapping of the individual patient's physiological state; Newly adopted structured data is continuously input into the digital mapping to update the key feature parameters and their correlations; When the update magnitude caused by new data exceeds a preset update threshold, a complete reconstruction of the digital mapping is triggered.

8. The method for full-course disease management based on big data according to claim 1, characterized in that: After generating a personalized full-course disease management pathway, the process also includes a dynamic adjustment step: During the implementation of the pathway, the deviation between the patient's real-time health data and the expected goals of the pathway is continuously monitored; When the deviation exceeds the adaptive adjustment threshold, an instantaneous simulation of the digital twin model is triggered to simulate the effects of different adjustment schemes. Based on the simulation results, the goals, intensity, or frequency of interventions are dynamically adjusted in subsequent stages of the personalized full-course management pathway.

9. A big data-based full-course disease management system, characterized in that: The method for full-course disease management based on big data as described in any one of claims 1 to 8 includes: The data acquisition module is used to acquire multimodal time-series health data of patients throughout the entire course of their disease; The data processing module is used to perform disease-aware processing on the unstructured text data in the multimodal time-series health data, including: determining the target disease currently related to the patient; processing the unstructured text data through an information extraction model to obtain at least one medical entity and its corresponding confidence score; obtaining the corresponding automatic adoption threshold and manual review threshold based on the target disease, wherein the automatic adoption threshold is greater than the manual review threshold; comparing the confidence score with the automatic adoption threshold and the manual review threshold; when the confidence score is greater than or equal to the automatic adoption threshold, directly adopting the medical entity as structured data; when the confidence score is less than the automatic adoption threshold but greater than or equal to the manual review threshold, marking the medical entity as requiring manual review; when the confidence score is less than the manual review threshold, marking the medical entity as requiring detailed manual review. The digital twin module is used to build and update a digital twin model that reflects the individual health status of a patient based on adopted structured data; The pathway management module is used to assess the patient's risk at different stages of the disease based on the digital twin model and generate a personalized full-course management pathway. An optimization module is used to execute the personalized full-course disease management path and optimize the digital twin model based on execution feedback.

10. An electronic device, characterized in that: The electronic device includes: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores a computer program that can be executed by the at least one processor, which is then executed by the at least one processor to enable the at least one processor to perform the big data-based full-course management method according to any one of claims 1 to 8.

Images