Cloud platform-based chronic disease health data management system

Abstract

The present application relates to the technical field of chronic disease health management, in particular to a chronic disease health data management system based on a cloud platform, comprising: collecting physiological index time series data, medication compliance records and text of subjective feelings of signs through a data collection module in combination with a user terminal device and active reporting; a state analysis module reconstructs the multi-dimensional health state of the data set to generate accurate patient health state characteristics; a scheme generation module calls a cloud platform optimization model to generate individualized health management schemes and high-risk warning signs; a scheme correction module corrects the scheme in combination with subjective feelings of signs and historical behavior patterns to improve adaptability; and a warning intervention module generates medication reminders, health monitoring suggestions and other instructions. The system can accurately capture patient health dynamics, improve the executability of management schemes, and achieve precise and individualized health management for patients with chronic diseases.

Description

Technical Field

[0001] This invention relates to the field of chronic disease health management technology, and in particular to a cloud-based chronic disease health data management system. Background Technology

[0002] Currently, chronic disease health data management largely relies on cloud platforms to deploy basic data processing functions. It acquires patients' physiological indicator data through user terminal collection devices and generates basic health management suggestions by combining simple data analysis models. Some systems also include medication records. However, the overall data collection dimensions are relatively singular, focusing mainly on objective physiological indicators, and failing to achieve integrated collection and in-depth processing of multiple types of data.

[0003] In existing technologies, health status analysis often employs a single-dimensional data interpretation approach, lacking comprehensive consideration of temporal changes in physiological indicators, medication adherence, and subjective feelings. This leads to discrepancies between the generated health status characteristics and the patient's actual health condition. Furthermore, the generated health management plans are mostly generic templates, failing to be tailored to individual patient subjective feelings and historical behavioral habits, making it difficult to adapt to individual differences. Moreover, early warning and intervention instructions often focus solely on abnormal physiological indicators, lacking comprehensiveness and specificity. While existing technologies have implemented cloud-based management and medication reminders for chronic diseases, they suffer from the following shortcomings:

[0004] The system mostly uses fixed thresholds to judge abnormal physiological indicators, and has not established a dynamic response model between drug concentration and physiological indicators. It cannot distinguish between drug effects and confounding factors such as diet and exercise, resulting in biased health status assessment.

[0005] Current adherence management only records medication behavior without analyzing the correlation between adherence patterns and changes in physiological indicators, and lacks a targeted improvement strategy generation mechanism. Existing plans are mostly generated using standardized templates without adaptive adjustments based on patients' subjective feelings and historical behavioral patterns, resulting in poor plan feasibility and high patient resistance.

[0006] The core need of chronic disease management lies in achieving precise and individualized health interventions. However, existing technologies are unable to effectively integrate and deeply analyze multi-dimensional data, cannot accurately depict the health status of patients, and lack the ability to dynamically adapt management plans, making it difficult to meet the personalized health management needs of patients and effectively solve the problem of poor management results caused by incomplete data and poor plan adaptability. Summary of the Invention

[0007] The purpose of this invention is to address the shortcomings of existing technologies by proposing a cloud-based chronic disease health data management system.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a cloud-based chronic disease health data management system, comprising:

[0009] The data acquisition module collects individual dynamic data sets of chronic disease patients through acquisition devices deployed on user terminals and active reporting by users. The individual dynamic data sets include time-series data of physiological indicators, medication adherence records, and subjective feelings of physical signs.

[0010] The status analysis module performs multidimensional health status reconstruction processing on the individual dynamic data set to generate health status characteristics of the chronic disease patients;

[0011] The solution generation module calls the patient management optimization model deployed on the cloud platform to perform personalized intervention analysis on the health status characteristics, and generates individualized health management solutions and high-risk warning indicators for the patients with chronic diseases.

[0012] The plan correction module performs patient adaptation adjustment and correction processing on the individualized health management plan to generate a corrected health management plan. The patient adaptation adjustment and correction processing is based on the correlation between the subjective feeling text of the vital signs and the historical behavior pattern.

[0013] The early warning and intervention module generates a set of early warning and intervention instructions based on the high-risk early warning identifier. The set of early warning and intervention instructions includes medication reminder plans, health monitoring suggestion plans, and health management suggestion plans.

[0014] As a further aspect of the present invention, the individual dynamic data set is subjected to multidimensional health status reconstruction processing to generate health status characteristics of the chronic disease patients, including:

[0015] The health status characteristics include the trend of indicator fluctuations, compliance scores, and a set of keywords related to abnormal sensations;

[0016] The time-series data of the physiological indicators are divided into multiple monitoring cycle subsequences according to a preset time granularity, and each monitoring cycle subsequence corresponds to a continuous monitoring period.

[0017] For each of the monitoring period subsequences, the following processing is performed:

[0018] The medication behavior trajectory of the chronic disease patient is constructed by combining the medication adherence record, and the medication behavior trajectory includes the deviation sequence of actual medication time, dosage and preset plan;

[0019] The medication behavior trajectory is correlated with the monitoring period subsequence to generate the health status reconstruction result for the current monitoring period. The health status reconstruction result includes the indicator change curve affected by medication behavior.

[0020] Pattern recognition processing is performed on the health status reconstruction results from multiple consecutive monitoring periods to calculate the fluctuation trend of the indicators, compliance scores, and a set of keywords related to abnormal perceptions; among which,

[0021] The fluctuation trend of the indicator refers to the direction and amplitude of change of key physiological indicators over a long period of time.

[0022] The compliance score is calculated based on the deviation sequence of the actual medication time and dosage from the preset plan.

[0023] The set of keywords for abnormal sensations is obtained by extracting them from the subjective sensation text of the physical signs through natural language processing.

[0024] As a further aspect of the present invention, the medication behavior trajectory is correlated with the monitoring period subsequence to generate a health status reconstruction result for the current monitoring period, including:

[0025] Based on the preset pharmacokinetic model and the actual medication time and dosage, the theoretical blood drug concentration curve during the current monitoring period is calculated;

[0026] The preset pharmacokinetic model is a one-compartment or two-compartment model, and the parameters include: standard parameters obtained from the drug instructions or clinical pharmacology databases as typical population values ​​of Ka, Ke, and Vd; and individualized adjustments to Ke based on the patient's age, weight, and liver and kidney function.

[0027]

[0028] A one-compartment model was used for oral medication selection.

[0029]

[0030] Intravenous administration was performed using a two-compartment model;

[0031] For new drugs not included in the database, the system uses a population pharmacokinetic model and Bayesian feedback to iteratively update individual parameters based on 2-3 blood drug concentration monitoring data of the patient.

[0032] By comparing the changes of key indicators in the time series data of the physiological indicators with the theoretical blood drug concentration curve, a time-aligned comparison is made to establish an indicator-blood drug concentration response model.

[0033] Based on the aforementioned index-blood drug concentration response model, the index change component directly driven by medication behavior and the index fluctuation component caused by other factors are separated.

[0034] By integrating the indicator change component and the indicator fluctuation component, and combining them with the qualitative description extracted from the subjective feeling text of the vital signs, the health status reconstruction result containing the correlation between quantitative indicators and qualitative descriptions is generated.

[0035] As a further aspect of the present invention, the step of calling the patient management optimization model deployed on the cloud platform to perform personalized intervention analysis on the health status characteristics, and generating an individualized health management plan and high-risk warning indicator for the chronic disease patient, includes:

[0036] The fluctuation trend of the indicator is input into the trend assessment layer of the patient management optimization model, and the health status evolution path and key intervention time window are determined through the health status trend prediction module.

[0037] The compliance score is input into the behavior analysis layer of the patient management optimization model, and behavior pattern clustering calculation is performed to generate compliance improvement strategies and an assessment of the difficulty of strategy implementation.

[0038] The abnormal sensation keyword set is input into the symptom analysis layer of the patient management optimization model, and potential health concerns and potential triggers are retrieved based on the knowledge graph matching algorithm;

[0039] By integrating the key intervention time window, the compliance improvement strategy, and the potential health concerns, a personalized management needs index for the chronic disease patient is generated, and an individualized health management plan is determined based on the matching results of the personalized management needs index and the preset intervention strategy library.

[0040] Based on the health status evolution path, the difficulty assessment of the strategy implementation, and the severity classification of the potential triggers, high-risk warnings requiring immediate intervention, medium-risk warnings requiring regular monitoring, and low-risk warnings requiring continued observation are identified.

[0041] As a further aspect of the present invention, the individualized health management plan is subjected to patient adaptation adjustment and modification processing to generate a modified health management plan, including:

[0042] Extract patient preference statements and behavioral constraint descriptions from the subjective feelings of the physical signs, and identify fixed patterns and variable links in the patient's lifestyle habits;

[0043] The reminder time and frequency in the medication reminder scheme are calibrated based on the identified fixed pattern to generate a medication reminder correction scheme that conforms to the patient's circadian rhythm.

[0044] Based on the behavioral constraint description, the feasibility of the recommendations in the health management proposal is screened, and adjustment recommendations that patients cannot implement or strongly resist are eliminated.

[0045] By incorporating the aforementioned flexible steps, progressive adjustment alternatives are embedded into the health management recommendation plan to form a tiered behavior change plan;

[0046] The revised health management plan is generated by integrating the medication reminder revision plan, the health management recommendation plan after feasibility screening, and the step-by-step behavior change plan.

[0047] As a further aspect of the present invention, the step of calibrating the reminder time and frequency in the medication reminder scheme according to the identified fixed pattern to generate a medication reminder correction scheme that conforms to the patient's circadian rhythm includes:

[0048] Analyze the patient’s daily meal, sleep and work patterns from the subjective feelings of the physical signs or historical behavioral data.

[0049] The matching degree between the preset medication time points in the medication reminder scheme and the patient's regular time period is calculated to identify conflicting time points and highly matching time points;

[0050] The medication reminders at the conflicting time points are adjusted to the adjacent, highly matched time points with low patient activity, and the adjusted time intervals are ensured to meet the minimum interval requirements for medication administration.

[0051] The adjusted daily medication time sequence is smoothed to ensure that the reminder intervals are evenly distributed, and the medication reminder correction scheme is generated.

[0052] Adjust medication reminders from conflicting time points to adjacent, highly matched time points, including activity quantification based on patient historical behavioral data to calculate activity scores for each time point:

[0053]

[0054] in The average number of steps. For terminal interaction frequency, ; A value less than 0.3 is considered low activity. The minimum interval constraint is set according to the drug instructions. Antihypertensive drugs Hours, hypoglycemic drugs Hours, anticoagulant Within hours, the system automatically matches drugs using their codes; the optimization goal is to maximize the overall matching accuracy. The constraints are Integer programming is used to solve the problem, and the coefficient of variation of the adjusted time series is checked for smoothness. Less than 0.2, that is

[0055]

[0056] Ensure the spacing is uniform.

[0057] As a further aspect of the present invention, a set of warning and intervention instructions is generated based on the high-risk warning identifier, including:

[0058] For high-risk warnings requiring immediate intervention, an instant alarm command and emergency contact plan are generated. The instant alarm command is triggered simultaneously through strong terminal reminder and manual review on the platform side.

[0059] To address the high-risk warnings that require regular monitoring, a structured health monitoring recommendation scheme is developed, which includes a list of follow-up examination items, recommended follow-up examination time ranges, and guidance on interpreting results.

[0060] Based on the compliance improvement strategy and the step-by-step behavior change plan, a personalized health management recommendation plan is generated, which includes specific quantitative goals and achievement paths for diet, exercise, and monitoring.

[0061] The real-time alarm instructions, emergency contact plan, structured health monitoring suggestion plan, and health management suggestion plan are prioritized and packaged according to the warning level and execution time to generate the warning and intervention instruction set.

[0062] As a further aspect of the present invention, the construction of the structured health monitoring recommendation scheme includes:

[0063] Based on the potential health concerns and potential triggers, corresponding clinical examination indicators and functional assessment items are mapped from the medical knowledge base;

[0064] Based on the aforementioned key intervention time windows, a recommended implementation time interval is set for each of the aforementioned clinical examination indicators and functional assessment items;

[0065] Configure a result interpretation template for each of the aforementioned clinical examination indicators and functional assessment items. The result interpretation template includes the normal value range, the meaning of abnormal values, and suggestions for the next steps.

[0066] The mapped clinical examination indicators and functional assessment items, the corresponding recommended execution time intervals, and the result interpretation templates are linked and integrated to generate a structured follow-up plan document.

[0067] As a further aspect of the present invention, the changes in key indicators in the time-series data of the physiological indicators are compared with the theoretical blood drug concentration curve in a time-aligned manner to establish an indicator-blood drug concentration response model, including:

[0068] Obtain the peak and trough time points of the theoretical blood drug concentration curve;

[0069] In the time series data of physiological indicators, the physiological indicator observation values ​​corresponding to the peak time point and the trough time point are identified. The physiological indicator observation values ​​within a set time window before and after the peak time point are marked as the first data set, and the physiological indicator observation values ​​within the set time window before and after the trough time point are marked as the second data set.

[0070] The mean value of the physiological indicators observed in the first dataset is calculated as the high concentration correlation value, and the mean value of the physiological indicators observed in the second dataset is calculated as the low concentration correlation value.

[0071] Based on the difference between the high concentration correlation value and the low concentration correlation value, and the difference between the peak value and the trough value of the theoretical blood drug concentration curve, the change in physiological indicators caused by a unit change in blood drug concentration is calculated as the core response coefficient.

[0072] A time delay factor is introduced, which is obtained by fitting the offset of the peak value of physiological index change relative to the peak time point of blood drug concentration.

[0073] By combining the core response coefficient and the time delay factor, a mathematical model is constructed to reflect the delayed response of physiological indicators to changes in theoretical blood drug concentration, which serves as the indicator-blood drug concentration response model.

[0074] As a further aspect of the present invention, the compliance score is input into the behavioral analysis layer of the patient management optimization model, behavioral pattern clustering calculation is performed, and compliance improvement strategies and strategy execution difficulty assessments are generated, including:

[0075] A cluster analysis dataset was constructed by extracting adherence score sequences and corresponding demographic attributes and behavioral log data from historical data containing multiple patients.

[0076] Based on the changing patterns of the adherence score sequence, patients in the cluster analysis dataset are grouped into unsupervised clusters to identify different categories of medication adherence behavior patterns.

[0077] For each category of behavioral pattern, high-frequency behavioral characteristics and environmental factors that lead to medication adherence behavior patterns are extracted from the demographic attributes and behavioral log data and used as behavioral pattern labels.

[0078] The current adherence scores, demographic attributes, and behavioral log data of chronic disease patients are input into the behavior analysis layer, matched to the closest category of behavior patterns, and a targeted list of behavior change suggestions is generated based on the matched behavior pattern labels, as part of the adherence improvement strategy.

[0079] Based on the overlap between the current chronic disease patients' behavioral log data and the high-frequency features in the matched behavioral pattern labels, as well as the average group compliance improvement difficulty corresponding to the behavioral pattern labels, the expected difficulty level of the current patients in implementing the compliance improvement strategy is quantified, which serves as an assessment of the difficulty of implementing the strategy.

[0080] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0081] This approach reconstructs a multidimensional health status profile of an individual's dynamic data set, including time-series physiological indicators, medication adherence records, and subjective feelings about physical signs. By integrating multi-type and multi-dimensional dynamic patient data, it overcomes the limitations of conventional techniques that rely solely on single physiological indicators for health status analysis. It comprehensively captures the dynamic changes in a patient's health status, taking into account both objective physiological data and subjective feelings. This eliminates biases caused by interpreting single data points, allowing health status profiles to better reflect the patient's actual health condition and achieving accurate characterization of the patient's health status, thus avoiding misjudgments due to incomplete data.

[0082] Based on the correlation between subjective feelings about physical signs and historical behavioral patterns, individualized health management plans are adjusted and revised to adapt to the patient, generating a revised health management plan. Unlike conventional technologies where general health management plans lack individual adaptability, this approach, by linking patient subjective feelings with historical behavioral patterns, allows for dynamic adjustments to plan details. This ensures the plan aligns with the patient's behavioral habits and subjective tolerance, preventing a disconnect between the plan and the patient's actual behavior, improving plan feasibility, reducing inadequate implementation due to plan discomfort, and making the adjusted plan more tailored to the individual patient's health needs, thus achieving precise implementation of individualized management. Attached Figure Description

[0083] Figure 1 This is a sequence diagram of the cloud-based chronic disease health data management system described in this invention.

[0084] Figure 2 Flowchart for multidimensional health status reconstruction processing;

[0085] Figure 3 A flowchart for generating health status reconstruction results for correlation analysis. Detailed Implementation

[0086] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0087] In the description of this invention, it should be understood that the terms "length," "width," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicating orientation or positional relationships, are based on the orientation or positional relationships shown in the accompanying drawings and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the invention. Furthermore, in the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0088] See Figure 1 This invention provides a cloud-based chronic disease health data management system, specifically comprising:

[0089] The system comprises a data acquisition module, a status analysis module, a treatment plan generation module, a treatment plan revision module, and an early warning and intervention module. The data acquisition module automatically collects time-series physiological indicator data from chronic disease patients through acquisition devices deployed on user terminals. It also receives medication adherence records and subjective feelings about physical signs proactively reported by patients through the user terminal application interface, collectively forming an individual dynamic data set. The status analysis module receives this individual dynamic data set and performs multi-dimensional health status reconstruction processing to generate health status characteristics that comprehensively reflect the patient's condition. The treatment plan generation module calls a pre-trained patient management optimization model deployed on a cloud platform. This model performs personalized intervention analysis on the input health status characteristics, outputting an individualized health management plan for the chronic disease patient and generating a high-risk early warning indicator. The treatment plan revision module performs patient-adaptive adjustments to the aforementioned individualized health management plan. This process is based on historical behavioral patterns and patient preferences analyzed from the subjective feelings about physical signs, ultimately generating a revised health management plan. Based on the high-risk warning indicators generated by the scheme generation module, the early warning and intervention module generates a series of specific warning and intervention instructions, including medication reminder schemes, health monitoring suggestion schemes, and health management suggestion schemes, and sends them to user terminals or relevant healthcare providers.

[0090] In one embodiment of the present invention, the state analysis module performs multidimensional health state reconstruction processing on an individual dynamic data set to generate health state characteristics of patients with chronic diseases. These health state characteristics include indicator fluctuation trends, compliance scores, and a set of keywords related to abnormal sensations. (See also...) Figure 2During processing, the time-series data of physiological indicators are divided into multiple continuous monitoring cycle subsequences according to a preset time granularity, with each monitoring cycle subsequence corresponding to a continuous monitoring period. For each monitoring cycle subsequence, association analysis is performed. A medication behavior trajectory for chronic disease patients is constructed by combining medication adherence records. This trajectory includes the deviation sequence of actual medication time, dosage, and preset regimen. Association analysis is then performed on this medication behavior trajectory and the current monitoring cycle subsequence to generate a health status reconstruction result for the current monitoring cycle. This reconstruction result includes curves showing the changes in indicators affected by medication behavior. Pattern recognition processing is performed on the health status reconstruction results of multiple consecutive monitoring cycles to calculate indicator fluctuation trends, adherence scores, and a set of keywords for abnormal sensations. Specifically, the indicator fluctuation trend represents the direction and amplitude of change of key physiological indicators over a long period; the adherence score is calculated based on the deviation sequence of actual medication time, dosage, and preset regimen; and the set of keywords for abnormal sensations is extracted through natural language processing of subjective feelings about physical signs.

[0091] One specific step in the correlation analysis of medication behavior trajectories and monitoring cycle subsequences involves establishing an indicator-blood drug concentration response model. This step includes: obtaining the peak and trough time points of the theoretical blood drug concentration curve; identifying the physiological indicator observations corresponding to these peak and trough time points in the physiological indicator time series data; labeling the physiological indicator observations within a set time window before and after the peak time point as the first data set; and labeling the physiological indicator observations within the same set time window before and after the trough time point as the second data set; calculating the mean of the physiological indicator observations in the first data set as the high-concentration correlation value, and calculating the mean of the physiological indicator observations in the second data set as the low-concentration correlation value; and calculating the change in physiological indicators caused by a unit change in blood drug concentration based on the difference between the high-concentration and low-concentration correlation values, and the difference between the peak and trough values ​​of the theoretical blood drug concentration curve, using this value as the core response coefficient. A time delay factor, tau, is introduced and obtained through the following steps:

[0092] collection Data from each dosing cycle was used to identify the peak plasma drug concentration time point in each cycle. Peak time of physiological indicators ;

[0093] Calculate offset ;

[0094] The maximum likelihood estimation method is used to fit tau:

[0095]

[0096] Summation traversal to ;

[0097] like Large variation (standard deviation) (minutes), a mixed-effects model was used to separate intra- and between-individual variability, and the individual predicted value was taken as the patient's predicted value. ;

[0098] The time delay factor is obtained by fitting the offset of the peak value of the physiological indicator change relative to the peak value of the blood drug concentration. Combining the core response coefficient and the time delay factor, a mathematical model reflecting the delayed response of the physiological indicator to changes in theoretical blood drug concentration is constructed, which serves as the indicator-blood drug concentration response model.

[0099] In its implementation, the state analysis module performs multidimensional health status reconstruction processing on the individual dynamic data set. The processing object is the raw set received from the data acquisition module, which contains continuous physiological indicator time-series data, medication adherence records, and subjective feelings of vital signs. The processing aims to generate a comprehensive and structured output of health status characteristics, including indicator fluctuation trends, adherence scores, and a set of keywords related to abnormal feelings. The processing begins with the periodic division of the physiological indicator time-series data. In the specific implementation, the system uses the natural day as the preset time granularity, dividing the continuous physiological indicator time-series data such as blood pressure and blood glucose into multiple monitoring period subsequences, each corresponding to a complete calendar day. For each monitoring period subsequence obtained, the state analysis module initiates a correlation analysis process. The correlation analysis process first integrates medication adherence records, which are derived from medication logs manually reported by patients through the terminal application. The integration process constructs a medication behavior trajectory reflecting the actual medication details of the day, including the deviation sequence of actual medication time, dosage, and the doctor's preset plan. In some embodiments, for hypertensive patients, the preset plan is to take one antihypertensive drug at 8:00 am and 8:00 pm every day, and the medication behavior trajectory is recorded as "take one drug at 9:05 am and take one drug at 8:30 pm", thus forming a deviation sequence from the preset time [+65 minutes, +30 minutes].

[0100] In practical implementation, generating health status reconstruction results requires deep correlation between medication behavior trajectories and monitoring cycle subsequences. The correlation analysis process invokes a pre-defined pharmacokinetic model. Based on the actual medication time and dosage in the medication behavior trajectory, the model calculates the theoretical blood drug concentration curve for the corresponding time period of the current monitoring cycle subsequence. Next, the changes in key indicators in the physiological time series data are compared with the theoretical blood drug concentration curve to establish an indicator-blood drug concentration response relationship model. Establishing this model involves multiple steps. In practice, the system first identifies the peak and trough times of blood drug concentration from the theoretical blood drug concentration curve. Subsequently, in the time-series data of physiological indicators, blood pressure observations corresponding to the peak and trough times of blood drug concentration were located. Based on pharmacokinetic studies, the peak time for blood drug concentration is typically 1-2 hours after administration, and a 30-minute window can cover the plateau period. Alternatively, based on the dynamic adjustment of the specific drug, all blood pressure observations within a 30-minute window before and after the peak time of blood drug concentration were labeled as the first data set, and all blood pressure observations within a 30-minute window before and after the trough time of blood drug concentration were labeled as the second data set. The arithmetic mean of all blood pressure observations in the first data set was calculated, and the result was used as the high-concentration correlation value. The arithmetic mean of all blood pressure observations in the second data set was calculated, and the result was used as the low-concentration correlation value.

[0101] Understandably, building a model requires quantifying the response relationship. Based on the difference between high-concentration and low-concentration correlation values, and the concentration difference between the peak and trough values ​​of the theoretical blood drug concentration curve, the system calculates the change in physiological indicators caused by a unit change in blood drug concentration. This change is defined as the response coefficient. The formula for calculating the response coefficient is:

[0102]

[0103] Where: symbol Represents the response coefficient, symbol Indicates high concentration correlation value, symbol Indicates low concentration correlation value, symbol The peak value of the theoretical blood drug concentration curve is represented by the symbol. This represents the trough value of the theoretical blood drug concentration curve. A time delay factor is introduced, which is obtained by fitting the average offset of the peak blood pressure change time relative to the peak blood drug concentration time point. Combining the calculated response coefficient and the time delay factor, the system constructs a mathematical model reflecting the lagged response of blood pressure values ​​to changes in theoretical blood drug concentration; this mathematical model is the indicator-blood drug concentration response relationship model. In specific implementation, based on the indicator-blood drug concentration response relationship model, the system can separate the indicator change component directly driven by medication behavior and the indicator fluctuation component caused by other factors from the blood pressure fluctuations observed in the monitoring period subsequence. The indicator change component and the indicator fluctuation component are fused, and keywords extracted from the subjective feeling text of vital signs on the same day are associated to generate the health status reconstruction result corresponding to the current monitoring period subsequence. The health status reconstruction result includes the indicator change curve affected by medication behavior. Optionally, the system performs pattern recognition processing on the health status reconstruction results of multiple consecutive days. Pattern recognition processing calculates the indicator fluctuation trend and compliance score over a long period, and extracts a set of abnormal feeling keywords by performing natural language processing on the accumulated subjective feeling text of vital signs over multiple days.

[0104] In one embodiment of the present invention, the state analysis module performs correlation analysis on the medication behavior trajectory and the monitoring cycle subsequence to generate a health status reconstruction result for the current monitoring cycle. The process includes: (See attached document). Figure 3 Based on a pre-defined pharmacokinetic model and actual medication time and dosage, the theoretical blood drug concentration curve for the current monitoring period is calculated. The changes in key indicators in the time-series physiological data are compared with this theoretical blood drug concentration curve to establish an indicator-blood drug concentration response model. Based on this model, the component of indicator change directly driven by medication behavior and the component of indicator fluctuation caused by other factors are separated from the observed physiological indicator changes. The separated indicator change and fluctuation components are fused and combined with qualitative descriptions extracted from subjective sensory reports to generate a health status reconstruction result for the current monitoring period. This result includes the correlation between quantitative indicators and qualitative descriptions.

[0105] In practice, the state analysis module generates the health status reconstruction results for the current monitoring period. This process begins with a correlation analysis of the medication behavior trajectory and the monitoring period sub-sequence. The correlation analysis calls a built-in pharmacokinetic model. In practice, this model is a mathematical model pre-constructed based on the pharmacokinetic parameters of a specific drug. The model calculates the theoretical blood drug concentration curve for the corresponding time period of the current monitoring period sub-sequence based on the input actual medication time and dosage. Specifically, for once-daily hypoglycemic drugs, the system calculates the theoretical blood drug concentration change curve from the time of medication administration ("10:00 AM") to the dosage ("one tablet") recorded in the medication behavior trajectory, combined with the drug's half-life and volume of distribution parameters.

[0106] In some embodiments, after obtaining the theoretical blood drug concentration curve, the system performs a time alignment comparison operation, which places the changes of key indicators in the time series data of physiological indicators on the same time axis as the theoretical blood drug concentration curve. It can be understood that time alignment comparison requires precise matching of the timestamps of the blood glucose monitoring value sequence with the time points of the theoretical blood drug concentration curve to identify the blood glucose observations corresponding to the rising, peak, and falling phases of blood drug concentration. Based on the results of the time alignment comparison, the system performs the step of establishing an indicator-blood drug concentration response relationship model. The model aims to quantify the dynamic characteristics of the impact of blood drug concentration changes on physiological indicators. In some embodiments, the model building process analyzes the lag and magnitude ratio of blood glucose changes relative to blood drug concentration changes. This relationship can be characterized by a mathematical expression that includes delay and linear response. Optionally, the expression describing the dynamic relationship within a single monitoring period is as follows:

[0107]

[0108] Where: symbol This represents the change in blood glucose level observed at time t, with the symbol […]. The sensitivity coefficient representing the response of blood glucose to changes in blood drug concentration, with the symbol... Indicates time Theoretical change in blood drug concentration at a point, symbol This represents the time delay between a change in blood drug concentration and the resulting observable change in blood glucose. Model parameters and It is obtained by fitting the data of the current monitoring period subsequence.

[0109] In practical implementation, based on the established indicator-pharmaceutical concentration response model, the system decomposes the time-series data of raw physiological indicators in the monitoring period subsequence. This decomposition process separates the indicator change component directly driven by medication behavior and the indicator fluctuation component caused by other factors from the observed overall indicator fluctuations. The indicator change component is calculated by the model based on the theoretical pharmaceutical concentration curve, reflecting the theoretical indicator change trajectory under ideal conditions caused solely by drug action. The indicator fluctuation component, on the other hand, is obtained by subtracting the indicator change component from the raw observation data. This component includes the influence of other confounding factors such as diet, exercise, and emotions. Optionally, after component separation, the system performs fusion and correlation operations, fusing the indicator change component and the indicator fluctuation component, and simultaneously introducing qualitative descriptive keywords extracted from the subjective feelings of patients reported during the current monitoring period. In practice, for a diabetic patient, the system generates a health status reconstruction result that includes quantitative curves and qualitative text. The health status reconstruction result may be presented as a blood glucose curve marked "abnormally high blood glucose peak after lunch". The curve also distinguishes the expected change part from the effect of the drug and the abnormal fluctuation part caused by external factors such as diet with different colors or line types.

[0110] In one embodiment of the present invention, the solution generation module invokes a patient management optimization model deployed on a cloud platform to perform personalized intervention analysis on health status characteristics, generating individualized health management plans and high-risk warning indicators. During processing, the indicator fluctuation trend is input into the trend assessment layer of the patient management optimization model, and the health status trend prediction module within this layer determines the patient's health status evolution path and key intervention time windows. The compliance score is input into the behavior analysis layer of the patient management optimization model, performing behavior pattern clustering calculations to generate compliance improvement strategies and an assessment of the difficulty of strategy implementation. The abnormal perception keyword set is input into the symptom analysis layer of the patient management optimization model, and potential health concerns and potential triggers are retrieved based on a knowledge graph matching algorithm. By integrating the key intervention time window, compliance improvement strategies, and potential health concerns, a personalized management needs index for the chronic disease patient is generated. Based on the matching results of this personalized management needs index with the preset intervention strategy library in the cloud platform, the final individualized health management plan is determined. Based on the evolution path of health status, the assessment of the difficulty of strategy implementation, and the severity classification of potential triggers, high-risk warnings requiring immediate intervention, medium-risk warnings requiring regular monitoring, and low-risk warnings requiring continued observation were identified. These were determined by fitting historical data from 100 patients, with actual compliance improvement as the dependent variable and overlap and group difficulty as independent variables, minimizing prediction error.

[0111] When performing behavioral pattern clustering calculations at the behavior analysis layer to generate adherence improvement strategies and assess the difficulty of strategy implementation, the process specifically includes: extracting adherence score sequences containing multiple patients and corresponding demographic attributes and behavioral log data from historical data to construct a clustering analysis dataset. Based on the changing patterns of the adherence score sequences, unsupervised clustering is performed on the patients in the clustering analysis dataset to identify different categories of medication adherence behavior patterns. For each category of behavior pattern, high-frequency behavioral features and environmental factors leading to this type of medication adherence behavior pattern are extracted from demographic attributes and behavioral log data, and these features are used as labels for that behavior pattern. The current chronic disease patients' adherence scores, demographic attributes, and behavioral log data are input into the behavior analysis layer, matched to the closest category of behavior pattern, and a targeted list of behavior change suggestions is generated based on the matched behavior pattern labels. This list is a component of the adherence improvement strategy. Based on the overlap between the current chronic disease patient's behavior log data and the high-frequency features in the matched behavior pattern labels, as well as the average compliance improvement difficulty of the group corresponding to the behavior pattern label, the expected difficulty level of the current patient in implementing the compliance improvement strategy is quantified. This difficulty level serves as an assessment of the difficulty of strategy implementation.

[0112] In practical implementation, the solution generation module invokes the patient management optimization model deployed on the cloud platform to perform personalized intervention analysis on health status characteristics. At the start of processing, the solution generation module receives health status characteristics from the status analysis module, including indicator fluctuation trends, compliance scores, and a set of abnormal sensation keywords. The solution generation module inputs the indicator fluctuation trends into the trend assessment layer of the patient management optimization model. The trend assessment layer processes the indicator fluctuation trends through an internally integrated health status trend prediction module, which determines the health status evolution path and key intervention time windows for patients with chronic diseases based on a time-series prediction algorithm. The solution generation module inputs the compliance scores into the behavior analysis layer of the patient management optimization model. The behavior analysis layer performs behavior pattern clustering calculations, generating compliance improvement strategies and an assessment of the difficulty of strategy implementation. The solution generation module inputs the set of abnormal sensation keywords into the symptom analysis layer of the patient management optimization model. The symptom analysis layer uses a knowledge graph matching algorithm to retrieve potential health concerns and potential triggers associated with the abnormal sensation keyword set from the medical knowledge graph. The treatment plan generation module integrates key intervention time windows, adherence improvement strategies, and potential health concerns to generate a personalized management needs index for patients with chronic diseases. This index is a structured vector. Individualized health management plans are determined based on the matching results between the personalized management needs index and the pre-set intervention strategy library in the cloud platform. Based on the health status evolution path, strategy implementation difficulty assessment, and severity grading of potential triggers, the module identifies high-risk warnings requiring immediate intervention, medium-risk warnings requiring regular monitoring, and low-risk warnings requiring continued observation.

[0113] In practical implementation, the behavior analysis layer performs behavior pattern clustering calculations in multiple steps. It extracts adherence score sequences and corresponding demographic attributes and behavioral log data from historical data stored on the cloud platform to construct a clustering analysis dataset. Based on the changing patterns of the adherence score sequences, unsupervised clustering is performed on the patients in the clustering analysis dataset. This unsupervised clustering grouping uses a clustering algorithm to identify different categories of medication adherence behavior patterns. For each category of medication adherence behavior pattern, high-frequency behavioral features and environmental factors leading to the medication adherence behavior pattern are extracted from the demographic attributes and behavioral log data. These high-frequency behavioral features and environmental factors together constitute the behavior pattern label. In some embodiments, the unsupervised clustering grouping identifies three types of medication adherence behavior patterns: "morning missed dose type," "midday delayed dose type," and "nighttime regular dose type." For the "morning missed dose type," behavioral pattern labels such as "rushing out of home in the morning" and "irregular breakfast" are extracted. The current adherence scores, demographic attributes, and behavioral log data of chronic disease patients are input into the behavior analysis layer. The behavior analysis layer matches them to the closest category of behavior patterns and generates a targeted list of behavior change suggestions based on the matched behavior pattern labels. The list of behavior change suggestions is a component of the adherence improvement strategy.

[0114] It is understandable that generating a strategy execution difficulty assessment requires quantitative analysis. Based on the overlap between the current chronic disease patient's behavioral log data and the high-frequency features in the matched behavioral pattern labels, and the average group adherence improvement difficulty corresponding to the behavioral pattern labels, the expected difficulty level for the current patient to implement the adherence improvement strategy is quantified. In some embodiments, the expected difficulty level is derived through a calculation model using the formula:

[0115]

[0116] Where: symbol The symbol represents the quantitative value of the strategy execution difficulty assessment. This indicates the degree of overlap between the current chronic disease patient's behavioral log data and the high-frequency features in the matched behavioral pattern labels. Calculated through feature matching ratio; symbol The symbol represents the average difficulty of improving compliance among groups corresponding to behavioral pattern labels; it is a constant statistically derived from historical improvement data. and It involves adjusting the weights to satisfy... The formula calculates The values ​​are mapped to three difficulty levels: high, medium, and low.

[0117] The difficulty P in improving the average compliance of the group was statistically determined through the following steps:

[0118] All patient cases exhibiting this behavioral pattern were extracted from the historical database. Cases showing substantial improvement in adherence scores were selected, and those with an improvement greater than 20% and a duration greater than 30 days were categorized as conversion cases. For each conversion case, the number of days required for improvement was calculated. Take all The median is used as the typical improvement time for this type of pattern; the improvement success rate is calculated. ;

[0119] Difficulty level:

[0120]

[0121] The observation period is capped at 90 days. The value ranges from 0 to 1, with larger values ​​indicating higher difficulty. If the number of cases for a certain pattern is less than 10, the expert Delphi method is used to assign values, and the median is taken as the median. , marked as expert experience value.

[0122] The rules for mapping E values ​​to high, medium, and low difficulty levels are as follows:

[0123] Low risk ( The patient characteristics and the pattern are highly overlapping, and the pattern is easy to improve, so improvement is expected to be easy;

[0124] Medium risk ( Patient characteristics partially overlap with the pattern or the pattern is of moderate difficulty to improve, requiring the development of targeted strategies;

[0125] High risk ( Patients with low overlap with the pattern or whose pattern is difficult to improve require gradual intervention and enhanced follow-up.

[0126] Optionally, the compliance improvement strategies output by the behavior analysis layer include specific recommendations, with an assessment of the difficulty of strategy implementation attached as a rating label.

[0127] In practice, the behavioral pattern clustering calculation process relies on an internally maintained behavioral pattern feature table, which stores the typical features and association information of various behavioral patterns. Refer to Table 1 for a partial view of the behavioral pattern feature table:

[0128] Table 1: Characteristics of Medication Adherence Behavioral Patterns

[0129]

[0130] Understandably, when performing matching, the behavior analysis layer compares the current patient's compliance score sequence with the "typical compliance score sequence characteristics" in the table to determine the closest behavior pattern category. After matching the "lunchtime delay type" category, the behavior analysis layer generates behavior change suggestions such as "setting flexible reminders related to lunch events" based on "high-frequency behavior characteristics" and "environmental factor characteristics," and uses the "group average improvement difficulty coefficient" of 0.55 in the calculation of strategy execution difficulty assessment. In some embodiments, the behavior pattern characteristic table is periodically updated with new historical data. Finally, the compliance improvement strategy and strategy execution difficulty assessment output by the behavior analysis layer, along with the outputs of the trend assessment layer and the symptom analysis layer, are used by the solution generation module for fusion and decision-making.

[0131] In one embodiment of the present invention, the plan modification module performs patient-adaptive adjustment and modification processing on the individualized health management plan to generate a modified health management plan. The processing includes: extracting patient preference statements and behavioral constraint descriptions from the patient's subjective feelings of vital signs, and identifying fixed patterns and flexible elements in the patient's lifestyle. Based on the identified fixed patterns, the medication reminder plan in the individualized health management plan is calibrated, adjusting its reminder time and frequency to generate a modified medication reminder plan that conforms to the patient's circadian rhythm. Based on the behavioral constraint descriptions, the health management suggestions in the individualized health management plan are screened for feasibility, eliminating adjustment suggestions that the patient cannot implement or strongly resists. Combined with the identified flexible elements, progressive adjustment alternatives are embedded in the health management suggestion plan to form a step-by-step behavior change plan. The modified medication reminder plan, the feasibility-screened health management suggestion plan, and the step-by-step behavior change plan are integrated to generate the modified health management plan.

[0132] The process of calibrating the reminder time and frequency in the medication reminder plan based on the identified fixed patterns to generate a revised medication reminder plan includes: analyzing the patient's daily meal, sleep, and work patterns from subjective perception data or historical behavioral data. Based on the Chinese Pharmacopoeia and drug instructions, the values ​​are: 4 hours for antihypertensive drugs, 6 hours for hypoglycemic drugs, and 12 hours for anticoagulants. The system automatically matches the preset medication time points in the medication reminder plan with the patient's regular time periods through drug coding, calculating the matching degree and identifying conflicting and highly matched time points. Medication reminders at conflicting time points are adjusted to adjacent, highly matched time points with lower patient activity levels, ensuring that the adjusted time interval meets the minimum interval requirement for medication administration. The smoothness of the adjusted full-day medication time point sequence is verified to ensure uniform distribution of reminder intervals, ultimately generating the revised medication reminder plan.

[0133] In its implementation, the plan modification module performs patient-adaptive adjustments to the individualized health management plan. This process begins with parsing the received individualized health management plan, which includes initial versions of medication reminder plans, health monitoring suggestions, and health management suggestions. The plan modification module extracts patient preferences and behavioral constraints from the text of subjective feelings about vital signs. Examples of patient preferences include "habitually going to bed late" and "needs a nap after lunch," while examples of behavioral constraints include "allergic to seafood" and "knee discomfort preventing running." Based on these extracted preferences and behavioral constraints, the plan modification module identifies fixed patterns and adaptable elements in the patient's lifestyle. Fixed patterns refer to daily, stable, and unchangeable behavioral periods, while adaptable elements refer to behavioral habits that can be adjusted.

[0134] In some embodiments, the protocol correction module calibrates the reminder time and frequency in the medication reminder protocol based on the identified fixed patterns. The calibration operation generates a medication reminder correction protocol that conforms to the patient's daily rhythm. The calibration operation includes several sub-steps. The protocol correction module analyzes the patient's daily regular time periods for meals, sleep, and work from subjective feelings of vital signs or historical behavioral data. In a specific implementation, through analysis of a week's behavioral log, the fixed patterns are determined to be "breakfast time 07:00-07:30", "lunch break time 12:30-13:00", and "bedtime 23:30". The protocol correction module calculates the matching degree between the preset medication time points in the medication reminder protocol and the patient's regular time periods. The matching degree calculation identifies conflicting time points and highly matching time points. It can be understood that the matching degree calculation is based on the degree of time overlap and the patient's activity status for evaluation. Optionally, the preset medication time "12:00" overlaps with the start of the fixed pattern "lunch break time 12:30-13:00", which may interfere with the lunch break, and is therefore identified as a conflict time point; while the preset medication time "07:30" matches the end point of "breakfast time 07:00-07:30", and is identified as a high-match time point.

[0135] In practice, the protocol correction module adjusts medication reminders at conflicting time points to adjacent, highly matched time points with low patient activity. This adjustment ensures the adjusted time interval meets the minimum medication interval requirement. The module performs a smoothness check on the adjusted full-day medication time sequence, ensuring even distribution of reminder intervals and a relative standard deviation of less than 20%, avoiding excessively scattered or concentrated reminder times. This ultimately generates a medication reminder correction plan. In some embodiments, the medication reminder at the conflicting time point "12:00" is adjusted to a highly matched and low-activity time point at "13:05" (after lunch break). The adjusted time is approximately 5.5 hours from the post-breakfast medication time "07:30," meeting the minimum interval requirement of 4 hours, and approximately 4.9 hours from the pre-dinner medication time "18:00," passing the smoothness check. It can be understood that the adjustment amount can be calculated using an optimization function. The goal of this function is to maximize the match between the adjusted time points and the fixed lifestyle pattern while satisfying the minimum time interval constraint. An example for calculating the adjustment of a single time point is as follows:

[0136]

[0137] Where: symbol Indicates the preset original medication time point, symbol Indicates the adjusted target medication time point, symbol This represents the time offset calculated based on the matching degree of lifestyle patterns and the minimum interval constraint. During calibration, the scheme correction module generates a matching analysis table to assist decision-making; see Table 2.

[0138] Table 2: Matching Analysis of Patients' Daily Routine and Medication Reminder Time

[0139]

[0140] The plan revision module performs feasibility screening on the recommendations in the health management plan based on the behavioral constraint description, eliminating adjustments that patients cannot implement or strongly resist. In practice, if a health management plan includes a recommendation to "run for 30 minutes daily," but the behavioral constraint description includes "knee discomfort," the plan revision module marks this recommendation as "unfeasible" and removes it. Combining identified workarounds, the plan revision module embeds progressive adjustment alternatives into the health management plan, creating a step-by-step behavior change plan. For example, the aggressive recommendation to "reduce daily sodium intake to below 5g" is replaced with a step-by-step behavior change plan: "Record current intake in the first week, attempt to reduce to 7g in the second week, and target 5g in the third week." Finally, the plan revision module integrates the medication reminder revision plan, the feasibility-screened health management plan, and the step-by-step behavior change plan to generate the revised health management plan.

[0141] In one embodiment of the invention, the early warning and intervention module generates a set of early warning and intervention instructions based on high-risk early warning indicators. For high-risk early warnings requiring immediate intervention, an instant alarm instruction and an emergency contact plan are generated. The instant alarm instruction is triggered simultaneously through a strong reminder on the user terminal and manual review on the cloud platform. For medium-risk early warnings requiring periodic monitoring, a structured health monitoring suggestion plan is constructed. This structured health monitoring suggestion plan includes a list of review items, a recommended review time range, and guidance on interpreting results. Based on the compliance improvement strategy and tiered behavior change plan generated by the plan correction module, a personalized health management suggestion plan is generated. This plan includes specific quantitative goals and achievement paths for aspects such as diet, exercise, and monitoring. The instant alarm instruction, emergency contact plan, structured health monitoring suggestion plan, and health management suggestion plan are prioritized and packaged according to early warning level and execution timeliness to generate the final set of early warning and intervention instructions.

[0142] The process of constructing a structured health monitoring recommendation plan includes: mapping corresponding clinical examination indicators and functional assessment items from a medical knowledge base connected to the cloud platform based on potential health concerns and potential triggers identified by the plan generation module; setting a recommended execution time interval for each clinical examination indicator and functional assessment item based on the key intervention time window determined by the plan generation module; configuring a result interpretation template for each clinical examination indicator and functional assessment item, which includes the normal range, the meaning of abnormal values, and next action suggestions; and linking and integrating the mapped clinical examination indicators and functional assessment items, the corresponding recommended execution time intervals, and the result interpretation templates to generate a structured follow-up plan document.

[0143] In practical implementation, the early warning and intervention module generates a set of early warning and intervention instructions based on high-risk early warning indicators. This set includes medication reminder plans, health monitoring recommendations, and health management recommendations. For high-risk early warnings requiring immediate intervention, the module generates instant alarm instructions and emergency contact plans. The instant alarm instructions trigger a forced pop-up notification and continuous sound alarm through the user's terminal application for strong reminders. Simultaneously, a review task containing real-time patient data and historical trends is pushed to the monitoring interface of medical staff on the cloud platform to achieve simultaneous triggering of manual review. In some embodiments, for high-risk early warnings determined by the system to be persistently elevated blood glucose levels accompanied by the keyword "chest pain," the instant alarm instructions activate the terminal alarm and simultaneously create an emergency contact plan. The emergency contact plan includes instructions to automatically dial emergency services and to send early warning information to preset family contacts.

[0144] The early warning and intervention module constructs a structured health monitoring recommendation plan for medium-risk warnings requiring regular monitoring. This plan is a standardized document containing a list of follow-up examination items, recommended follow-up timeframes, and results interpretation guidelines. The process of constructing the structured health monitoring recommendation plan begins with the early warning and intervention module mapping corresponding clinical examination indicators and functional assessment items from a medical knowledge base connected to the cloud platform, based on potential health concerns and potential triggers obtained from the plan generation module. In specific implementation, for medium-risk warning patients identified as having "increased risk of heart failure" with the potential trigger being "volume overload," the early warning and intervention module maps clinical examination indicators and functional assessment items such as "serum B-type natriuretic peptide (BNP) test," "echocardiography," and "daily weight variability assessment" from the medical knowledge base. The early warning and intervention module, combined with the key intervention time windows obtained from the plan generation module, sets a recommended execution time interval for each clinical examination indicator and functional assessment item. This recommended execution time interval defines the recommended time range for performing each examination, and its calculation relies on an algorithm based on the warning generation time, which outputs a specific time window. Here is an example of how to determine the recommended execution time interval:

[0145]

[0146] Where: symbol Indicates the calculated recommended execution time interval, symbol Indicates the base time point for the generation of the high-risk warning sign, symbol This represents the earliest execution time offset based on the urgency of the examination item and clinical guidelines, with the symbol... This indicates the latest execution time offset determined based on an assessment of patient compliance history. For example, for "echocardiography," if... , If so, it is recommended that the execution time range be from the 7th to the 30th day after the warning is generated.

[0147] In some embodiments, the early warning intervention module configures a result interpretation template for each clinical examination indicator and functional assessment item. The result interpretation template includes the normal value range, the meaning of abnormal values, and suggestions for next steps. Optionally, the result interpretation template configured for "serum B-type natriuretic peptide (BNP) detection" will explicitly list "normal value <100 pg / mL" and state that "if the result is between 100-300 pg / mL, it suggests a mild risk of heart failure, and it is recommended to optimize the diuretic regimen; if it is >300 pg / mL, it is recommended to have a follow-up consultation with a cardiologist as soon as possible." The early warning intervention module integrates and correlates the mapped clinical examination indicators and functional assessment items, the corresponding recommended execution time intervals, and the result interpretation templates to generate a structured follow-up plan document. It can be understood that the early warning intervention module generates personalized health management recommendations based on the compliance improvement strategies and step-by-step behavior change plans obtained from the plan generation module and plan revision module. In practical implementation, if the adherence improvement strategy includes "associating medication with morning coffee," and the tiered behavior change plan includes "starting with 2000 steps per day and increasing by 500 steps per week," then the generated health management recommendation plan will clearly define the specific quantitative goals and achievement path for "taking medication immediately after drinking coffee every morning and completing the daily walking goal after taking the medication, with a first-week goal of 2000 steps." The early warning and intervention module will prioritize and package the immediate alarm instructions, emergency contact plans, structured health monitoring recommendation plans, and health management recommendation plans according to the warning level and execution time. The priority rule is that high-risk warning instructions take precedence over medium-risk warning instructions, and emergency instructions take precedence over routine instructions. Finally, a complete set of warning and intervention instructions will be generated and issued. The output results of this system need to be reviewed by a licensed physician and should not be used as a direct basis for clinical diagnosis or treatment decisions.

[0148] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A cloud-based chronic disease health data management system, characterized in that: The system includes: The data acquisition module collects individual dynamic data sets of chronic disease patients through acquisition devices deployed on user terminals and active reporting by users. The individual dynamic data sets include time-series data of physiological indicators, medication adherence records, and subjective feelings of physical signs. The status analysis module performs multidimensional health status reconstruction processing on the individual dynamic data set to generate health status characteristics of the chronic disease patients; The solution generation module calls the patient management optimization model deployed on the cloud platform to perform personalized intervention analysis on the health status characteristics, and generates individualized health management solutions and high-risk warning indicators for the patients with chronic diseases. The plan correction module performs patient adaptation adjustment and correction processing on the individualized health management plan to generate a corrected health management plan. The patient adaptation adjustment and correction processing is based on the correlation between the subjective feeling text of the vital signs and the historical behavior pattern. The early warning and intervention module generates a set of early warning and intervention instructions based on the high-risk early warning identifier. The set of early warning and intervention instructions includes medication reminder plans, health monitoring suggestion plans, and health management suggestion plans.

2. The cloud-based chronic disease health data management system according to claim 1, characterized in that, The individual dynamic data set is subjected to multidimensional health status reconstruction processing to generate the health status characteristics of the chronic disease patients, including: The health status characteristics include the trend of indicator fluctuations, compliance scores, and a set of keywords related to abnormal sensations; The time-series data of the physiological indicators are divided into multiple monitoring cycle subsequences according to a preset time granularity, and each monitoring cycle subsequence corresponds to a continuous monitoring period. For each of the monitoring period subsequences, the following processing is performed: The medication behavior trajectory of the chronic disease patient is constructed by combining the medication adherence record, and the medication behavior trajectory includes the deviation sequence of actual medication time, dosage and preset plan; The medication behavior trajectory is correlated with the monitoring period subsequence to generate the health status reconstruction result for the current monitoring period. The health status reconstruction result includes the indicator change curve affected by medication behavior. Pattern recognition processing is performed on the health status reconstruction results from multiple consecutive monitoring periods to calculate the fluctuation trend of the indicators, compliance scores, and a set of keywords related to abnormal perceptions; among which, The fluctuation trend of the indicator refers to the direction and amplitude of change of key physiological indicators over a long period of time. The compliance score is calculated based on the deviation sequence of the actual medication time and dosage from the preset plan. The set of keywords for abnormal sensations is obtained by extracting them from the subjective sensation text of the physical signs through natural language processing.

3. The cloud-based chronic disease health data management system according to claim 2, characterized in that, The medication behavior trajectory is correlated with the monitoring period subsequence to generate a health status reconstruction result for the current monitoring period, including: Based on the preset pharmacokinetic model and the actual medication time and dosage, the theoretical blood drug concentration curve during the current monitoring period is calculated; The preset pharmacokinetic model is a one-compartment model or a two-compartment model, and the parameters include: Standard parameters, Ka, Ke, and Vd, are obtained from the drug's package insert or clinical pharmacology databases as typical population values. Individualized adjustments are made to Ke based on the patient's age, weight, and liver and kidney function. ； The oral medication was selected using a one-compartment model. ； Intravenous administration was performed using a two-compartment model; For new drugs not included in the database, the system uses a population pharmacokinetic model and Bayesian feedback to iteratively update individual parameters based on 2-3 blood drug concentration monitoring data of the patient. By comparing the changes of key indicators in the time series data of the physiological indicators with the theoretical blood drug concentration curve, a time-aligned comparison is made to establish an indicator-blood drug concentration response model. Based on the aforementioned index-blood drug concentration response model, the index change component directly driven by medication behavior and the index fluctuation component caused by other factors are separated. By integrating the indicator change component and the indicator fluctuation component, and combining them with the qualitative description extracted from the subjective feeling text of the vital signs, the health status reconstruction result containing the correlation between quantitative indicators and qualitative descriptions is generated.

4. The cloud-based chronic disease health data management system according to claim 3, characterized in that, The process involves calling a patient management optimization model deployed on a cloud platform to perform personalized intervention analysis on the health status characteristics, generating individualized health management plans and high-risk warning indicators for the chronic disease patients, including: The fluctuation trend of the indicator is input into the trend assessment layer of the patient management optimization model, and the health status evolution path and key intervention time window are determined through the health status trend prediction module. The compliance score is input into the behavior analysis layer of the patient management optimization model, and behavior pattern clustering calculation is performed to generate compliance improvement strategies and an assessment of the difficulty of strategy implementation. The abnormal sensation keyword set is input into the symptom analysis layer of the patient management optimization model, and potential health concerns and potential triggers are retrieved based on the knowledge graph matching algorithm; By integrating the key intervention time window, the compliance improvement strategy, and the potential health concerns, a personalized management needs index for the chronic disease patient is generated, and an individualized health management plan is determined based on the matching results of the personalized management needs index and the preset intervention strategy library. Based on the health status evolution path, the difficulty assessment of the strategy implementation, and the severity classification of the potential triggers, high-risk warnings requiring immediate intervention, medium-risk warnings requiring regular monitoring, and low-risk warnings requiring continued observation are identified.

5. The cloud-based chronic disease health data management system according to claim 4, characterized in that, The individualized health management plan is adjusted and modified to suit the patient's needs, resulting in a revised health management plan, including: Extract patient preference statements and behavioral constraint descriptions from the subjective feelings of the physical signs, and identify fixed patterns and variable links in the patient's lifestyle habits; The reminder time and frequency in the medication reminder scheme are calibrated based on the identified fixed pattern to generate a medication reminder correction scheme that conforms to the patient's circadian rhythm. Based on the behavioral constraint description, the feasibility of the recommendations in the health management proposal is screened, and adjustment recommendations that patients cannot implement or strongly resist are eliminated. By incorporating the aforementioned flexible steps, progressive adjustment alternatives are embedded into the health management recommendation plan to form a tiered behavior change plan; The revised health management plan is generated by integrating the medication reminder revision plan, the health management recommendation plan after feasibility screening, and the step-by-step behavior change plan.

6. The cloud-based chronic disease health data management system according to claim 5, characterized in that, The step of calibrating the reminder time and frequency in the medication reminder scheme according to the identified fixed pattern, and generating a medication reminder modification scheme that conforms to the patient's circadian rhythm, includes: Analyze the patient’s daily meal, sleep and work patterns from the subjective feelings of the physical signs or historical behavioral data. The matching degree between the preset medication time points in the medication reminder scheme and the patient's regular time period is calculated to identify conflicting time points and highly matching time points; The medication reminders at the conflicting time points are adjusted to the adjacent, highly matched time points with low patient activity, and the adjusted time intervals are ensured to meet the minimum interval requirements for medication administration. The adjusted daily medication time sequence is smoothed to ensure that the reminder intervals are evenly distributed, and the medication reminder correction scheme is generated. Adjust medication reminders from conflicting time points to adjacent, highly matched time points, including activity quantification based on patient historical behavioral data to calculate activity scores for each time point: ； in The average number of steps. For terminal interaction frequency, ; A value less than 0.3 is considered low activity. The minimum interval constraint is set according to the drug instructions. Antihypertensive drugs Hours, hypoglycemic drugs Hours, anticoagulant Within hours, the system automatically matches drugs based on their codes; Optimization objective: Maximize overall matching degree The constraints are Integer programming is used to solve the problem, and the coefficient of variation of the adjusted time series is checked for smoothness. Less than 0.2, that is ； Ensure the spacing is uniform.

7. The cloud-based chronic disease health data management system according to claim 5, characterized in that, A set of warning and intervention instructions is generated based on the high-risk warning identifier, including: For high-risk warnings requiring immediate intervention, an instant alarm command and emergency contact plan are generated. The instant alarm command is triggered simultaneously through strong terminal reminder and manual review on the platform side. To address the high-risk warnings that require regular monitoring, a structured health monitoring recommendation scheme is developed, which includes a list of follow-up examination items, recommended follow-up examination time ranges, and guidance on interpreting results. Based on the compliance improvement strategy and the step-by-step behavior change plan, a personalized health management recommendation plan is generated, which includes specific quantitative goals and achievement paths for diet, exercise, and monitoring. The real-time alarm instructions, emergency contact plan, structured health monitoring suggestion plan, and health management suggestion plan are prioritized and packaged according to the warning level and execution time to generate the warning and intervention instruction set.

8. The cloud-based chronic disease health data management system according to claim 7, characterized in that, The proposed structured health monitoring recommendation scheme includes: Based on the potential health concerns and potential triggers, corresponding clinical examination indicators and functional assessment items are mapped from the medical knowledge base; Based on the aforementioned key intervention time windows, a recommended implementation time interval is set for each of the aforementioned clinical examination indicators and functional assessment items; Configure a result interpretation template for each of the aforementioned clinical examination indicators and functional assessment items. The result interpretation template includes the normal value range, the meaning of abnormal values, and suggestions for the next steps. The mapped clinical examination indicators and functional assessment items, the corresponding recommended execution time intervals, and the result interpretation templates are linked and integrated to generate a structured follow-up plan document.

9. The cloud-based chronic disease health data management system according to claim 3, characterized in that, By comparing the changes of key indicators in the time-series data of the physiological indicators with the theoretical blood drug concentration curve over time, an indicator-blood drug concentration response relationship model is established, including: Obtain the peak and trough time points of the theoretical blood drug concentration curve; In the time series data of physiological indicators, the physiological indicator observation values ​​corresponding to the peak time point and the trough time point are identified. The physiological indicator observation values ​​within a set time window before and after the peak time point are marked as the first data set, and the physiological indicator observation values ​​within the set time window before and after the trough time point are marked as the second data set. The mean value of the physiological indicators observed in the first dataset is calculated as the high concentration correlation value, and the mean value of the physiological indicators observed in the second dataset is calculated as the low concentration correlation value. Based on the difference between the high concentration correlation value and the low concentration correlation value, and the difference between the peak value and the trough value of the theoretical blood drug concentration curve, the change in physiological indicators caused by a unit change in blood drug concentration is calculated as the core response coefficient. A time delay factor is introduced, which is obtained by fitting the offset of the peak value of physiological index change relative to the peak time point of blood drug concentration. By combining the core response coefficient and the time delay factor, a mathematical model is constructed to reflect the delayed response of physiological indicators to changes in theoretical blood drug concentration, which serves as the indicator-blood drug concentration response model.

10. The cloud-based chronic disease health data management system according to claim 4, characterized in that, The compliance score is input into the behavioral analysis layer of the patient management optimization model, and behavioral pattern clustering calculation is performed to generate compliance improvement strategies and an assessment of the difficulty of strategy implementation, including: A cluster analysis dataset was constructed by extracting adherence score sequences and corresponding demographic attributes and behavioral log data from historical data containing multiple patients. Z-score standardization was performed on the compliance score sequence to eliminate dimensional differences, and the DTW distance between patients was calculated. ； The summation iterates through all the regularized paths. Point pairs were used for hierarchical clustering using Ward's method, with the number of clusters... Determined by maximizing the profile coefficient. The value ranges from 3 to 8, and the identified behavioral patterns include: morning missed service, midday delayed service, nighttime regularity, and random fluctuation. For each category of behavioral pattern, high-frequency behavioral characteristics and environmental factors that lead to medication adherence behavior patterns are extracted from the demographic attributes and behavioral log data and used as behavioral pattern labels. The current adherence scores, demographic attributes, and behavioral log data of chronic disease patients are input into the behavior analysis layer, matched to the closest category of behavior patterns, and a targeted list of behavior change suggestions is generated based on the matched behavior pattern labels, as part of the adherence improvement strategy. Based on the overlap between the current chronic disease patients' behavioral log data and the high-frequency features in the matched behavioral pattern labels, as well as the average group compliance improvement difficulty corresponding to the behavioral pattern labels, the expected difficulty level of the current patients in implementing the compliance improvement strategy is quantified, which serves as an assessment of the difficulty of implementing the strategy.

Images