A personalized health management intervention recommendation system and method for chronic disease patients

Abstract

The application belongs to the technical field of medical data processing and intelligent decision-making assistance, and discloses a personalized health management intervention suggestion system and method for chronic disease patients. The system collects multi-source health data of the chronic disease patients, determines a dynamic time window; in the dynamic time window, the health data is preprocessed, synchronized, credibility evaluated and credibility gated state encoded to generate a patient state vector; candidate intervention suggestions are generated according to the patient state vector and chronic disease knowledge rules; a comprehensive adaptation score is calculated; finally, individualized parameters are updated according to doctor review feedback and patient execution feedback, and a next cycle intervention suggestion sorting result is generated. The reliability, safety, explainability and continuous individualized optimization capability of the chronic disease health management intervention suggestion generation are improved.

Description

Technical Field

[0001] This invention belongs to the field of medical data processing and intelligent decision support technology, specifically relating to a personalized health management intervention suggestion system and method for patients with chronic diseases. Background Technology

[0002] Patients with chronic diseases typically require long-term, multi-dimensional health management, including monitoring blood sugar, blood pressure, heart rate, exercise, diet, sleep, and medication adherence. Existing health management platforms can usually collect patients' test data, behavioral data, and follow-up data, and generate health reminders, lifestyle suggestions, or follow-up appointment prompts based on preset rules or models.

[0003] However, in actual chronic disease management scenarios, patient health data exhibits significant instability. On one hand, health data sources are complex, including hospital information systems, home testing devices, wearable devices, patient-generated records, and manual input from doctors. These different sources suffer from variations in sampling frequencies, unit formats, data delays, missing data, duplicates, anomalies, and cross-source inconsistencies. On the other hand, the health risks of chronic disease patients are not static; factors such as blood sugar fluctuations, blood pressure fluctuations, abnormal heart rate, hypoglycemia risk, fall risk, and kidney function abnormality risk change over time. While some existing technologies can generate personalized recommendations based on patient health data, and others can perform risk assessments or provide feedback adjustments, the following shortcomings remain: First, existing solutions often assume that the collected data can be directly used in recommendation calculations, lacking a comprehensive credibility assessment of different data sources, sampling completeness, cross-source consistency, physician confirmation status, and degree of abnormal deviation. This can lead to low-credibility data triggering inappropriate intervention recommendations.

[0004] Secondly, existing solutions typically aggregate health data at fixed intervals, such as daily, weekly, or monthly, making it difficult to dynamically adjust the data observation window based on the patient's current risk level and fluctuations in key indicators. This approach fails to balance timeliness during high-risk phases with stability during low-risk phases.

[0005] Third, existing solutions often involve manual review or simple filtering after recommendations are generated, lacking a risk hard constraint admission mechanism implemented before comprehensive scoring. This results in recommendations with high scores but with taboo risks still being included in the recommendation ranking.

[0006] Fourth, existing solutions typically focus on improving indicators or user preferences, failing to adequately incorporate data quality, physician historical review acceptance, patient implementation burden, and risk penalties into the scoring. This results in the ranking of recommendations failing to accurately reflect their suitability under current data and risk conditions.

[0007] Fifth, the feedback updates of existing programs mostly focus on whether patients have completed the recommendations or whether the indicators have improved, lacking a closed-loop optimization mechanism that transforms doctor review results, patient execution results, abnormal events, burden scores, and preference feedback into recommended parameters for the next cycle.

[0008] Therefore, it is necessary to propose a technical solution that can safely, reliably, interpretably, and iteratively generate personalized health management intervention recommendations under conditions of unstable quality of multi-source health data, dynamic changes in chronic disease risks, and continuous updates in doctor-patient feedback. Summary of the Invention

[0009] The purpose of this invention is to provide a personalized health management intervention suggestion system and method for patients with chronic diseases, in order to solve the problems of how to dynamically select the data aggregation time window, how to encode the credibility gating status of different health data, how to implement risk hard constraint access control before suggestion scoring, and how to continuously optimize the intervention suggestions for the next cycle by combining doctor review feedback and patient implementation feedback when there are missing, abnormal, cross-source inconsistencies and credibility differences in the multi-source health data of patients with chronic diseases.

[0010] To achieve the above objectives, the present invention adopts the following technical solution: On the one hand, the present invention provides a personalized health management intervention suggestion system for patients with chronic diseases.

[0011] It includes: a data governance and status coding unit, a candidate suggestion generation unit, a security access processing unit, an adaptation scoring output unit, and a feedback iterative optimization unit.

[0012] The data governance and state coding unit is used to collect multi-source health data of patients with chronic diseases. It determines a dynamic time window based on the patient's risk level, the degree of fluctuation of key indicators and data completeness, and performs preprocessing, synchronization, credibility assessment and credibility gating state coding on the multi-source health data within the dynamic time window to generate a patient state vector.

[0013] The candidate suggestion generation unit is used to generate a set of candidate intervention suggestions based on the patient state vector and chronic disease knowledge rules.

[0014] The safety access processing unit is used to perform elimination, downgrading, mandatory review, or approval of candidate intervention recommendations based on risk constraints, physician constraints, and contraindication rules before conducting a comprehensive evaluation of the candidate intervention recommendations, and to resolve conflicts for candidate intervention recommendations that have passed the processing.

[0015] The fit score output unit is used to calculate a comprehensive fit score for candidate intervention recommendations based on expected indicator improvement, patient implementation accessibility, historical intervention response, data quality, physician historical review acceptance, implementation burden, and risk penalty, and outputs the ranked intervention recommendations and corresponding explanatory information.

[0016] The feedback iteration optimization unit is used to collect doctor review feedback and patient execution feedback, and update individualized parameters based on the doctor review feedback and patient execution feedback to generate the ranking results of intervention recommendations for the next cycle.

[0017] Preferably, the multi-source health data includes data from medical institution systems, data from home testing devices, data from wearable devices, data actively entered by patients, and data entered by doctors.

[0018] Medical institution system data includes electronic medical records, diagnostic records, laboratory test results, prescription records, and follow-up records; home testing device data includes data generated by blood glucose meters, blood pressure monitors, body fat scales, pulse oximeters, and pulmonary function meters; wearable device data includes data generated by wristbands, smartwatches, and heart rate belts; patient-initiated data includes dietary records, exercise records, symptom records, medication records, and questionnaire data; and doctor-initiated data includes contraindications, intervention goals, follow-up cycles, manual risk levels, and review rules.

[0019] Preferably, the collected health data includes, but is not limited to, vital sign data such as blood sugar, blood pressure, and heart rate, as well as behavioral data such as medication, diet, and exercise.

[0020] Preferably, the data governance and state coding unit is used to calculate the comprehensive risk value for the current period. Preferably, the comprehensive risk value Based on blood glucose risk value Blood pressure risk value Heart rate abnormality risk value Fall risk value and doctors manually labeling risk values To be determined jointly.

[0021] All risk values ​​are normalized risk values, ranging from [0,1], with higher values ​​indicating higher risk. The system uses the following Sigmoid function to map risk factors to the [0,1] interval: Calculate as follows: in, Indicates the deviation of the mean blood glucose level. The numbers indicate blood glucose fluctuations: Hypo indicates the frequency of hypoglycemia, and Hyper indicates the frequency of hyperglycemia. , , , Let be the blood glucose risk weight parameter, and satisfy: Calculate as follows: in, Indicates the deviation of systolic blood pressure. Indicates the deviation of diastolic blood pressure. Indicates blood pressure variability. , , Let be the blood pressure risk weight parameter, and satisfy the following: Calculate as follows: in, Indicates the degree of deviation of resting heart rate. The values ​​indicate the degree of abnormality in heart rate variability; Tachy indicates the frequency of tachycardia, and Brady indicates the frequency of bradycardia. , , , Let be the heart rate risk weighting parameter, and satisfy: Calculate as follows: Among them, BalanceRisk represents the risk of balance ability, GaitRisk represents the risk of gait abnormalities, and FallHistory represents the risk of historical falls. , , Let be the fall risk weight parameter, and satisfy: Doctors manually label risk values The normalized risk value is input by the doctor and ranges from [0,1].

[0022] Current cycle comprehensive risk value Determine as follows: Preferably, the data governance and state coding unit calculates the fluctuation values ​​of key indicators: in, This refers to the fluctuation value of key indicators. The standard deviation of the key indicator within the candidate time window. This represents the average of the key indicators within the candidate time window.

[0023] Dynamic Time Window The rules for determining it are as follows: like ≥0.75 or ≥0.35. =1 day; If 0.45≤ <0.75, =3 days; like <0.45, =7 days; If there are no abnormal events for 14 consecutive days and the completeness of key data is ≥0.8, =14th, Preferably, the data governance and status coding unit sequentially performs format standardization, outlier identification, missing value completion, time window aggregation, and feature normalization on multi-source health data within a dynamic time window.

[0024] Standardized formats include: blood glucose units are standardized to mmol / L; blood pressure units are standardized to mmHg; exercise duration units are standardized to minute; sleep duration units are standardized to hour; and calories are standardized to kcal.

[0025] Outlier identification includes: if a step count is greater than 5 times the patient’s average over the past 30 days and the initial confidence level of the device is lower than a preset threshold, the step count is marked as a suspected outlier; if the time interval between two consecutive blood pressure measurements is less than 10 seconds and the difference in values ​​exceeds a preset range, it is marked as a duplicate outlier; if a data point is outside the physical acquisition range of the device, the data is removed.

[0026] Missing value completion includes: linear interpolation when adjacent valid data exist within the same time window; historical mean of the patient when similar valid data exist within the past 7 days for the same patient; median of the same disease and age group when historical data of the patient is insufficient; and reducing the reliability of the corresponding data within the time window when key indicators cannot be completed.

[0027] Preferably, the data governance and state coding unit calculates the credibility of each type of health data within the current time window. Credibility The value is a normalized value, ranging from [0,1]. The larger the value, the more reliable the health data.

[0028] Reliability of the i-th type of health data Calculate as follows: First, calculate the original linearly weighted values: The original value is truncated to obtain the credibility score: :when When ≥1, =1; when 0≤ <1 hour, = ;when When <0, =0; To assess the reliability of the equipment; For data integrity; For sampling frequency stability; To ensure consistency of similar data across sources; Mark the doctor for confirmation; The degree of abnormal deviation is represented by w1 to w6, which are preset weight parameters. And the following conditions must be met: in, , , , , and All values ​​are normalized values, ranging from [0,1].

[0029] Preferably, the data governance and status coding unit determines the level of trust. Generate gating factor Gating factor according to Determined according to the following rules: When When ≥0.8, =1; when 0.5≤ When <0.8, = When 0.3≤ When <0.5, ;when When <0.3, =0; in, This represents the gating weight when the i-th type of health data participates in state coding.

[0030] State characteristics after credibility gating Generate as follows: in, This represents the aggregated features of the i-th type of health data within a dynamic time window. Indicates to The eigenvalues ​​after normalization This represents the state characteristics after credibility gating.

[0031] Preferably, the patient state vector Including physiological indicator vectors after credibility gating Behavioral compliance vector after credibility gating Risk constraint vector Historical intervention response vector Data quality vector Doctor constraint vector And generate it in the following way: in, This represents the physiological index vector after credibility gating. This represents the behavioral compliance vector after credibility gating. Represents the risk constraint vector. Represents the historical intervention response vector. Represents the data quality vector. Let T represent the doctor constraint vector, and T represent the transpose.

[0032] Physiological index vector At least the following parameters should be included: mean fasting blood glucose, mean postprandial blood glucose, blood glucose variability coefficient, mean systolic blood pressure, mean diastolic blood pressure, number of abnormal blood pressure readings, mean heart rate, BMI, sleep duration, and post-exercise variability.

[0033] Behavioral compliance vector At least the completion rates for medication check-in, exercise, diet record, blood glucose or blood pressure monitoring, and follow-up questionnaires should be included.

[0034] Risk constraint vector At a minimum, it includes risks of hypoglycemia, hypertension, blood pressure fluctuations, abnormal heart rate, exercise contraindications, falls, abnormal kidney function, and physician-mandated contraindications.

[0035] Historical intervention response vector It should include at least the effectiveness of dietary intervention, exercise intervention, monitoring reminders, medication reminders, and follow-up reminders.

[0036] Data Quality Vector This includes at least the reliability of blood glucose data, blood pressure data, exercise data, diet data, sleep data, and medication record data.

[0037] Doctor constraint vector This should include at least the upper limit of exercise intensity set by the doctor, the review requirements for dietary recommendations set by the doctor, the lower limit of monitoring frequency set by the doctor, the rules for abnormal warning pushes set by the doctor, and the contraindications set by the doctor.

[0038] Preferably, the chronic disease knowledge rules include rule number, applicable chronic disease type, triggering conditions, required data type, required minimum data credibility, intervention type, intervention content, risk constraints, contraindication labels, rule priority, evidence level, and whether doctors are allowed to modify and review the rules.

[0039] The candidate suggestion generation unit generates candidate intervention suggestions when the following conditions are met simultaneously: the triggering conditions in the chronic disease knowledge rules are met; the minimum data credibility required by the rule is not higher than the credibility of the corresponding health data; the candidate intervention suggestion has not been disabled by the doctor; and the candidate intervention suggestion does not hit the absolute prohibition rule.

[0040] Preferably, a set of candidate intervention recommendations Represented as: Each candidate intervention recommendation This includes, but is not limited to, suggestion number, suggestion type, target metric, execution frequency, and execution cycle.

[0041] Candidate intervention types include, but are not limited to, dietary recommendations, exercise recommendations, weight management recommendations, indicator monitoring recommendations, medication adherence reminders, follow-up visit reminders, health education content delivery, abnormal risk warnings, manual follow-up recommendations, and data supplementation recommendations.

[0042] When the credibility of key data is lower than the generation threshold, the candidate suggestion generation unit does not generate automatic intervention suggestions based on the key data, but generates data supplementation suggestions or physician review tasks.

[0043] Preferably, the safety access processing unit processes the set of candidate intervention recommendations before multi-objective scoring. Perform hard constraint filtering for risk; the filtering function is: in, This represents the original set of candidate intervention recommendations. Represents the risk constraint vector. Let represent the doctor constraint vector, and ContraRules represent the set of taboo rules. This represents the set of filtered candidate intervention suggestions.

[0044] The system provides recommendations for each candidate intervention. Output processing actions : Among them, Reject indicates hitting an absolute taboo, and the candidate intervention suggestion is... The candidate intervention is not included in the scoring module; "Downgrade" indicates a moderate risk but not an absolute contraindication, thus lowering the recommended intervention level. After determining the intensity, frequency, or cycle, the data is entered into the scoring module; reviews indicate insufficient data or physician rule requirements, and candidate intervention recommendations are then submitted. Marked as mandatory doctor review; Pass indicates no contraindications or review rules were met, candidate intervention recommendations. Enter the scoring module.

[0045] Hard risk constraints include: like and ,but ; like If intervention_type(c_i) = exercise, then intensity_level(c_i) will be limited to 1, and set... ; like and For exercise recommendations on an empty stomach, then ; like and For strict dietary control recommendations, then ; like and For a high-protein diet recommendation, then ; like ,but ; If the doctor's constraint vector If a certain type of suggestion is required to be reviewed, then .

[0046] in, This indicates the risk value for abnormal heart rate. The value indicates the blood pressure risk; Hypo indicates the frequency of hypoglycemia. Indicates the risk value of abnormal kidney function. Indicating candidate intervention recommendations The data quality score of the data on which it is based.

[0047] Preferably, the security access processing unit is also used to perform conflict resolution on candidate recommendations that have passed the risk hard constraint filtering. Conflict types include objective conflicts, time conflicts, risk conflicts, and execution burden conflicts.

[0048] Goal conflict refers to different candidate recommendations leading to opposite management directions for the same health indicator; time conflict refers to overlapping execution times for multiple candidate recommendations; risk conflict refers to a conflict between high-scoring recommendations and risk contraindications; and execution burden conflict refers to the number of candidate recommendations or execution frequency exceeding the patient's achievable range within the same cycle.

[0049] The priority of conflict resolution is: Absolute contraindication rules > Doctor-preset rules > High-risk threshold rules > Data quality rules > Multi-objective scoring results > Patient preference rules.

[0050] When a conflict is detected, the system performs at least one of the following actions: deleting low-priority suggestions, reducing suggestion strength, adjusting suggestion execution time, merging similar suggestions, or marking conflicting suggestions as part of the doctor's review process.

[0051] Preferably, the adaptation scoring output unit calculates candidate intervention recommendations according to the following formula. Overall compatibility score : in, Indicating candidate intervention recommendations The expected improvement value of the indicator, This indicates that the patient's accessibility is achieved. This indicates the historical intervention response score. Indicating candidate intervention recommendations The data quality score of the data on which it is based. This indicates the acceptance rate of doctors' historical records. Indicates the burden of execution. This represents the risk penalty value, and α, β, γ, λ, μ, δ, and ε are the scoring weight parameters.

[0052] The scoring weight parameters satisfy: Candidate intervention recommendations Data quality score Calculate as follows: in, Indicating candidate intervention recommendations The credibility of the j-th type of key data on which it is relied. Indicating candidate intervention recommendations The number of key data types it depends on.

[0053] Acceptance of doctor's history review Calculate as follows: in, Indicating candidate intervention recommendations The number of times the corresponding recommendation type was accepted by doctors in the historical review records. Indicating candidate intervention recommendations The number of times the suggestion type has been reviewed by doctors in the historical review record.

[0054] when hour, Take the preset initial value . The default value of 0.6 indicates a moderate acceptance tendency, which can be configured by the administrator according to the department type during system deployment.

[0055] Risk penalty value Calculate as follows: in, This represents the risk value of type j. Indicating candidate intervention recommendations Does it include the j-th type of risk label, where m represents the number of risk types, and: Preferably, the adaptive scoring output unit is configured according to... The system generates a ranked list of suggestions from highest to lowest, retaining the rating component for each suggestion. The adapted rating output unit also generates a structured explanation object. The doctor's explanation text includes triggering rules, key indicators, data credibility, risk filtering results, conflict resolution results, scoring components, historical response basis, and matters requiring doctor confirmation.

[0056] The patient-side execution text includes details on what to do, when to do it, how long to do it, what to record, how to provide feedback in case of abnormalities, and whether it is necessary to contact a doctor.

[0057] Preferably, the language generation model is only used to optimize the textual expression of structured explanation objects, and does not directly generate intervention suggestions that have not been risk-filtered and ranked.

[0058] Preferably, the feedback iterative optimization unit collects doctor review feedback. Doctor review operations include confirming suggestions, modifying suggestion strength, modifying execution frequency, modifying execution cycle, adding contraindications, adding follow-up requirements, rejecting suggestions, marking rejection reasons, and setting mandatory review conditions.

[0059] Doctor review records include, but are not limited to, review record number, patient number, review action, and modified content.

[0060] The doctor's review results are converted into parameter feedback: ReviewResult_k=1, if the doctor confirms; ReviewResult_k=0.5, if the doctor modifies; ReviewResult_k=0, if the doctor rejects.

[0061] The formula for updating doctor acceptance is: in, Let ρ be the physician acceptance rate for the k-th suggestion in period t, and ρ be the historical decay coefficient, ranging from 0.6 to 0.9. The taboo conditions added by the physician are written into the physician constraint vector. And participate in the next cycle's hard constraint filtering of risk.

[0062] Preferably, the feedback iterative optimization unit maintains an individualized parameter set. : in, Indicates the weight of historical responses. Indicates the reachability weights to be applied. Indicates the risk penalty weight. Indicates patient preference weights, This indicates the weight of doctors' acceptance.

[0063] The actual intervention effect of the k-th candidate intervention suggestion Calculate as follows: in, This indicates the change in the target indicator after the implementation of the k-th candidate intervention recommendation. Indicates the execution completion rate. This represents the result of normalizing the number of abnormal events during the observation period. The normalization method is as follows: ,in This is the upper limit for reference (5 is recommended). Indicates the patient burden score, , , and Calculate weight parameters for the effect, and satisfy: Predictive intervention effect of candidate intervention in category k Determine as follows: in, This represents the expected improvement in the indicator value of the k-th candidate intervention recommendation. This indicates the historical intervention response score. This indicates the data quality score. Indicates the risk penalty value. , , and Let these be the weight parameters for the prediction effect, and satisfy the following: The feedback iterative optimization unit updates the individualized parameters as follows: in, , , and To update the step size parameter, and , , , ; This is the historical attenuation coefficient, and ; This represents the expected completion rate of the k-th type of candidate intervention recommendations; This indicates the patient's acceptance of the k-th candidate intervention recommendation; This indicates the doctor's review results.

[0064] After the update, each parameter is truncated to keep it within the range of [0,1].

[0065] Secondly, the present invention also provides a method for intelligent generation and optimization of personalized health management intervention recommendations for patients with chronic diseases, including: S1 collects multi-source health data from patients with chronic diseases.

[0066] S2 determines the dynamic time window based on the patient's risk level, the degree of fluctuation of key indicators, and the completeness of data.

[0067] S3 performs format unification, anomaly identification, missing data completion, time alignment, and feature normalization on multi-source health data within a dynamic time window.

[0068] S4, Calculate the credibility of each type of health data. And based on credibility Generate gating factor .

[0069] S5, based on the gating factor The patient state vector is generated by weighting physiological indicators and behavioral compliance characteristics to produce a confidence-gated vector. .

[0070] S6, based on the patient state vector Generate a set of candidate intervention suggestions based on chronic disease knowledge rules. .

[0071] S7, based on the risk constraint vector Doctor constraint vector And taboo rules, for the set of candidate intervention recommendations Perform risk hard constraint filtering to obtain the filtered candidate set. .

[0072] S8, The filtered candidate set The conflict resolution processes, including those related to objectives, time, risks, and execution burdens, yield a candidate set of results. .

[0073] S9 calculates a comprehensive fit score based on expected indicator improvement, execution achievability, historical response, data quality, physician acceptance, execution burden, and risk penalty.

[0074] S10 generates a sorted list of intervention recommendations based on the comprehensive fit score, and outputs the doctor's explanation text and the patient's execution text.

[0075] S11, collect patient feedback and doctor review feedback.

[0076] S12, Update individualized parameters based on patient feedback and physician review feedback. +1.

[0077] S13, based on the updated individualization parameters Generate the ranking results of intervention recommendations for the next cycle.

[0078] Thirdly, the present invention can also be implemented in the form of a computer program product, a computer-readable storage medium, or an electronic device. The electronic device includes one or more processors, a memory, a communication interface, and a data bus. The memory stores a computer program, which, when executed by the processor, implements the above-described method steps. The communication interface is used for data interaction with hospital information systems, doctor-end terminals, patient-end mobile terminals, home testing devices, and wearable devices. The memory stores patient health data, chronic disease knowledge rules, contraindication rules, patient state vectors, candidate suggestion sets, scoring results, doctor review records, patient execution feedback, and individualized parameters. The computer-readable storage medium can be a read-only memory, random access memory, a disk, an optical disk, flash memory, or other media capable of storing program instructions. When the program instructions stored in this medium are executed by the processor, the intelligent generation and optimization method for personalized health management intervention suggestions described in this invention is implemented.

[0079] Compared with the prior art, the present invention has at least the following advantages: 1. This invention uses a dynamic time window determination mechanism to dynamically select a data aggregation window of 1 day, 3 days, 7 days, or 14 days based on the patient's current risk level, the degree of fluctuation of key indicators, and the completeness of data. This enables high-risk patients or those with significant indicator fluctuations to receive more timely intervention suggestions, and enables low-risk patients with complete data to obtain more stable trend judgments.

[0080] 2. This invention incorporates device credibility, data integrity, sampling frequency stability, cross-source consistency, doctor confirmation marking, and abnormal deviation degree into the data credibility calculation through data credibility assessment and credibility gating mechanism. This can avoid low-credibility data directly triggering automatic intervention suggestions and improve the reliability of chronic disease health management suggestions.

[0081] 3. This invention sets up a risk hard constraint filtering mechanism before comprehensive scoring, and performs elimination, downgrading, mandatory review or approval processing on candidate suggestions, so that suggestions with absolute taboos or high risks will not enter the automatic recommendation results due to high model scores, thereby improving the safety of intervention suggestions.

[0082] 4. This invention incorporates data quality, physician historical review acceptance, risk penalties, and implementation burden into a multi-objective score, so that the recommended ranking not only reflects the expected improvement of indicators, but also reflects the current data conditions, physician judgment experience, patient feasibility, and potential risks.

[0083] 5. This invention transforms doctor review feedback and patient execution feedback into individualized parameters for the next cycle, enabling the system to continuously adjust the recommendation ranking based on the patient's actual completion rate, indicator changes, abnormal events, patient burden, and doctor review results, thereby improving long-term management adaptability. Detailed Implementation

[0084] To enable those skilled in the art to better understand the present application, the technical solutions in specific embodiments of the present application will be clearly and completely described below. Unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by those skilled in the art.

[0085] Example 1 This embodiment provides a system for intelligent generation and optimization of personalized health management intervention suggestions for patients with chronic diseases. The system can be deployed on a server, cloud platform, hospital health management platform, or patient management terminal, or it can be implemented collaboratively by a server and a mobile terminal.

[0086] The system includes a data governance and status coding unit, a candidate suggestion generation unit, a security access processing unit, an adaptation scoring output unit, and a feedback iterative optimization unit.

[0087] The data governance and status coding unit connects the hospital information system, home testing devices, wearable devices, patient applications, and doctor applications to receive multi-source health data. After the multi-source health data enters the system, the data governance and status coding unit first associates the patient's identity based on the patient_id, then sorts it by time based on the timestamp, and finally identifies the data type according to the data_type.

[0088] The data governance and state coding unit is based on the comprehensive risk value of the current period. and key indicator volatility Determine the dynamic time window If the patient is currently experiencing significant fluctuations in blood sugar or blood pressure, or if the doctor has manually marked the risk as high, the system will use a shorter time window, such as 1 or 3 days. If the patient's condition is stable and the key data is complete, the system will use a 7 or 14-day time window to reduce the impact of occasional noise on suggestion generation.

[0089] After determining the dynamic time window, the system performs format standardization, outlier identification, missing value completion, time window aggregation, and feature normalization on the multi-source health data within that time window. For example, blood glucose is standardized to mmol / L, blood pressure to mmHg, exercise duration to minutes, and sleep duration to hours. For suspected outlier data, the system identifies them based on device reliability, historical averages, sampling intervals, and device physical range.

[0090] The system then calculates the reliability of various types of health data. Taking blood pressure data as an example, if the blood pressure monitor has high reliability, stable sampling frequency, complete data for the past 7 days, and little difference from the doctor's outpatient measurement results without significant abnormal deviation, then the blood pressure data has high reliability. If the interval between two consecutive blood pressure measurements is less than 10 seconds and the values ​​are significantly different, or if the original reliability of the device is low, then the blood pressure data has low reliability.

[0091] Data governance and state coding unit based on trust Generate gating factor The gating factor is then applied to the normalized time window aggregated features to obtain the confidence-gated state features. Multiple state features are further concatenated into a patient state vector. The patient state vector includes not only physiological and behavioral compliance indicators, but also risk constraints, historical intervention responses, data quality, and physician constraints.

[0092] The candidate suggestion generation unit receives the patient state vector. The system accesses chronic disease knowledge rules. These rules include triggering conditions, required data types, minimum data credibility, intervention type, intervention content, risk constraints, contraindication labels, priority, and review level. When the rule triggering conditions are met and the required data credibility is not lower than the rule requirements, the system generates candidate intervention suggestions. If the credibility of key data is insufficient, the system does not generate automatic suggestions based on that data; instead, it generates suggestions for supplementary data collection or a physician review task.

[0093] The safety access processing unit processes candidate suggestions before comprehensive scoring. For suggestions that meet absolute contraindications, the system rejects them; for suggestions with moderate risk but adjustable risk, the system downgrades them; for suggestions with insufficient data or those requiring review by the physician, the system reviews them; and for suggestions that do not meet contraindications or review requirements, the system passes them. Subsequently, the safety access processing unit resolves conflicts in the processed candidate suggestions.

[0094] The adaptation scoring output unit comprehensively scores candidate suggestions after passing security access and conflict resolution. The scoring considers expected indicator improvement, patient accessibility, historical intervention response, data quality, physician historical review acceptance, implementation burden, and risk penalties. The system outputs a suggestion ranking list from highest to lowest based on the comprehensive adaptation score, and generates physician-side explanatory text and patient-side implementation text.

[0095] The feedback iteration optimization unit collects doctor review feedback and patient execution feedback. Doctor review feedback includes confirmation, modification, rejection, and addition of contraindications; patient execution feedback includes completion rate, number of executions, indicator changes, abnormal events, burden score, rejection marker, and device record consistency. Based on the above feedback, the system updates the historical response weight, execution attainability weight, risk penalty weight, patient preference weight, and doctor acceptance weight for the next cycle.

[0096] Example 2 This embodiment illustrates the specific execution process of the data governance and state coding unit. Assume patient P001 has type 2 diabetes and hypertension. In the current cycle, the system receives the following data: fasting blood glucose uploaded from the hospital's laboratory system, postprandial blood glucose uploaded from the patient's home blood glucose meter, blood pressure data uploaded from a home blood pressure monitor, step count and heart rate data uploaded from a smartwatch, dietary records and medication attendance data actively filled out by the patient, and exercise intensity limits and hypoglycemia risk warnings set by the doctor.

[0097] The system first calculates the overall risk value. : The system calculates the mean blood glucose deviation, blood glucose variability, frequency of hypoglycemia, and frequency of hyperglycemia. =0.72; Calculated based on systolic blood pressure deviation, diastolic blood pressure deviation, and blood pressure variability. =0.58; Calculated based on resting heart rate deviation, heart rate variability abnormality, frequency of tachycardia, and frequency of bradycardia. =0.20; Calculated based on balance ability risk, gait abnormality risk, and historical fall risk. =0.10; Doctors manually label risk values =0.60.

[0098] =max(0.72,0.58,0.20,0.10,0.60)=0.72.

[0099] The system calculates the fluctuation values ​​of key indicators. If the mean blood glucose level within the candidate time window is 8.2 mmol / L and the standard deviation is 2.1 mmol / L, then: =2.1 / 8.2=0.256.

[0100] Because 0.45≤ <0.75, the system determines the current dynamic time window. It is for 3 days.

[0101] The system then processes multi-source data from the past three days. If a step count is 40,000 steps, the patient's average step count over the past 30 days is 6,500 steps, and the initial device reliability is low, then this step count is marked as potentially abnormal. If a patient lacks sleep duration data for a particular day, but has valid sleep data for the days before and after, the system uses linear interpolation to complete the data. If key blood glucose data cannot be completed, the system lowers the reliability of the blood glucose data.

[0102] The system calculates the reliability of various data types. For example, for blood glucose data, let: =0.85; =0.75; =0.70; =0.80; =0.60; =0.20; =0.20; =0.20; =0.15; =0.20; =0.15; =0.10.

[0103] but: =0.655.

[0104] Because 0.5≤ <0.8, system command = =0.655, this blood glucose feature is reduced in weight and then used in the construction of the state vector.

[0105] If the reliability of a certain type of dietary record data is 0.25, then the system will... =0, this diet record is not used to automatically generate diet control recommendations and trigger supplementary diet records or doctor review tasks.

[0106] The system ultimately generates a patient state vector: in, This includes mean fasting blood glucose, mean postprandial blood glucose, glycemic variability coefficient, mean systolic blood pressure, mean diastolic blood pressure, and mean heart rate after confidence gating; This includes completion rates for medication check-ins, exercise, and diet records after credibility gating; This includes risks such as hypoglycemia, blood pressure fluctuations, and exercise contraindications. This includes the upper limit on exercise intensity set by doctors and the review requirements for dietary recommendations.

[0107] Example 3 This embodiment illustrates the process of candidate suggestion generation, safety access processing, and adaptation score output. The candidate suggestion generation unit reads the patient state vector. Rules for chronic disease knowledge. Rule R001 is: If the average postprandial blood glucose level is higher than a preset threshold, and the reliability of the postprandial blood glucose data is not lower than 0.6, then a postprandial dietary structure adjustment suggestion is generated; Rule R002 is: If the exercise completion rate is low and there are no exercise contraindications, then a moderate-intensity exercise suggestion is generated; Rule R003 is: If the risk of hypoglycemia is high, then an advice to prohibit fasting exercise is given.

[0108] If a patient's mean postprandial blood glucose level is high, the confidence level of the postprandial blood glucose data is 0.655, and the minimum confidence level required for rule R001 is 0.6, then the system generates candidate dietary adjustment suggestions. .

[0109] If a patient has a low exercise completion rate but a blood pressure fluctuation risk of 0.72, the system can still generate exercise candidate suggestions. However, this candidate recommendation needs to be assessed for risk within the security access processing unit.

[0110] The security access processing unit performs a HardFilter on the candidate suggestions. For motion-related candidate suggestions... ,like ≥0.7 and If =exercise, the system will limit the exercise intensity level to 1 and set... =Downgrade. This suggestion reduces the intensity before proceeding to the scoring module.

[0111] If another candidate suggestion For fasting exercise recommendations, and with Hypo ≥ 0.6, the system settings should be adjusted accordingly. =Reject, this suggestion will not be included in the rating module.

[0112] If candidate suggestions For strict dietary control recommendations, and with Hypo ≥ 0.6, the system settings are as follows: =Review, this suggestion will not be included in the automatic output list, but will be included in the doctor's mandatory review list.

[0113] For candidate suggestions that pass the safety access process, the system further performs conflict resolution. For example, if one suggestion requests an increase in exercise duration and another suggestion requests a decrease in exercise intensity due to blood pressure fluctuations, the system will process them according to the priority of absolute contraindication rules, doctor-preset rules, high-risk threshold rules, data quality rules, multi-objective scoring results, and patient preference rules. If the doctor has set an upper limit on exercise intensity, the doctor-preset rules take precedence, and the system retains low-intensity exercise suggestions while deleting or downgrading high-intensity exercise suggestions.

[0114] The adaptation scoring output unit calculates a comprehensive adaptation score for the remaining candidate suggestions. For example, candidate suggestions... The dietary adjustment recommendations are rated on the following scale: =0.78; =0.62; =0.70; =0.655; =0.82; =0.30; =0.20.

[0115] Assume: α=0.25; β=0.15; γ=0.15; λ=0.15; μ=0.10; δ=0.10; ε=0.10.

[0116] but: =0.52325.

[0117] The system follows Output a sorted list of recommendations from highest to lowest, and retain the score component, data quality source, risk filtering results, and conflict resolution results for each recommendation.

[0118] For doctors, the system outputs explanatory text, such as: This suggestion was triggered by a high average postprandial blood glucose level. The reliability of the postprandial blood glucose data is 0.655, and no absolute contraindication was hit. The comprehensive score is calculated by combining the expected indicator improvement, historical dietary intervention response, doctor acceptance, and implementation burden. It is recommended that doctors review the dietary adjustment content.

[0119] For the patient, the system outputs an execution text, such as: reduce the intake of high-sugar staple foods at dinner, execute for 3 consecutive days, and record post-meal blood glucose and dietary content daily; if symptoms of hypoglycemia or discomfort occur, stop execution and contact a doctor.

[0120] Example 4 This example illustrates how doctor review feedback and patient implementation feedback drive optimization in the next cycle. The doctor reviews the dietary recommendations output by the system. If the doctor confirms the recommendation, then... =1; If the doctor modifies the advice, for example, changing "reduce staple food by 50%" to "reduce staple food by 20% and increase vegetable intake," then =0.5; if the doctor rejects the suggestion, then =0.

[0121] The system updates the doctor acceptance rating based on the doctor's review results: like =0.8, =0.70, the doctor confirms the recommendation, then: =0.76.

[0122] After the patient follows the suggestions, the system collects the patient's feedback. Assuming a patient completion rate... The value is 0.85, indicating a change in the target indicator. The number of abnormal events is 0.60. The patient burden score is 0. If the value is 0.30, the system calculates the actual intervention effect: Let m1 = 0.4, m2 = 0.3, m3 = 0.2, m4 = 0.1, then: =0.465.

[0123] Before updating the historical response weights, the system calculates the predicted intervention effect of the k-th type of candidate intervention suggestion based on the rating components at the time of recommendation in the current period. : For example, if =0.78, =0.70, =0.655, =0.20, and =0.35, =0.25, =0.25, =0.15, then: =0.58175.

[0124] The system according to With predictive effect The difference is used to update the historical response weights; based on and The difference is used to update the execution accessibility weight; the risk penalty weight is updated based on the number of abnormal events; the patient preference weight is updated based on patient acceptance; and the doctor acceptance weight is updated based on the doctor's review results.

[0125] If a certain type of suggestion exhibits low completion rates, slow indicator improvement, or high anomaly events for several consecutive periods, the system reduces the ranking weight of that type of suggestion in subsequent periods. Conversely, if a certain type of suggestion exhibits high completion rates, significant indicator improvement, and no anomalies for several consecutive periods, the system increases the ranking weight of that type of suggestion in subsequent periods.

[0126] In the next cycle, the system reconstructs the patient state vector. +1, and combine with the updated individualized parameters +1 Regenerate the candidate suggestion set, perform hard-constraint risk filtering, resolve conflicts, and rank multi-objective scores to form new intervention suggestions. +1.

[0127] Finally, it should be noted that the described embodiments are merely some, not all, of the embodiments of the present invention. Those skilled in the art will understand that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the claims and their equivalents; that is, all other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

Claims

1. A personalized health management intervention recommendation system for patients with chronic diseases, characterized in that, include: The system includes a data governance and status coding unit, a candidate suggestion generation unit, a security access processing unit, an adaptation scoring output unit, and a feedback iterative optimization unit. The data governance and state coding unit is used to collect multi-source health data of patients with chronic diseases. It determines a dynamic time window based on the patient's risk level, the degree of fluctuation of key indicators and data completeness, and performs preprocessing, synchronization, credibility assessment and credibility gating state coding on the multi-source health data within the dynamic time window to generate a patient state vector. The candidate suggestion generation unit is used to generate a set of candidate intervention suggestions based on the patient state vector and chronic disease knowledge rules; The safety access processing unit is used to perform elimination, downgrading, mandatory review, or approval of candidate intervention recommendations based on risk constraints, physician constraints, and contraindication rules before conducting a comprehensive evaluation of the candidate intervention recommendations, and to resolve conflicts in the approved candidate intervention recommendations. The fit score output unit is used to calculate a comprehensive fit score for candidate intervention recommendations based on expected indicator improvement, patient implementation attainability, historical intervention response, data quality, physician historical review acceptance, implementation burden and risk penalty, and output the ranked intervention recommendations and corresponding explanatory information. The feedback iteration optimization unit is used to collect doctor review feedback and patient execution feedback, and update individualized parameters based on the doctor review feedback and patient execution feedback to generate the ranking results of intervention recommendations for the next cycle.

2. The system according to claim 1, characterized in that, The data governance and status coding unit determines the dynamic time window as follows: : in, This represents the overall risk value for the current cycle. This refers to the fluctuation value of key indicators; when or hour, It is the 1st; when hour, It is 3 days; when hour, It is 7 days; When there are no abnormal events for 14 consecutive days and the completeness of key data is not less than 0.8, It is the 14th.

3. The system according to claim 1, characterized in that, The data governance and state coding unit calculates the credibility of each type of health data. : in, For the sake of equipment reliability, For data integrity, For sampling frequency stability, To ensure cross-source consistency of similar data, For doctor confirmation marking, The degree of abnormal deviation is represented by w1 to w6, which are weighting parameters.

4. The system according to claim 3, characterized in that, Data governance and state coding unit based on trust Generate gating factor : when hour, ; when hour, ; when hour, ; when hour, ; The state features after confidence gating are generated according to the following formula: in, This represents the aggregated features of the original time window. These are the normalized eigenvalues.

5. The system according to claim 4, characterized in that, Patient state vector Including physiological indicator vectors after credibility gating Behavioral compliance vector after credibility gating Risk constraint vector Historical intervention response vector Data quality vector Doctor constraint vector And generate it in the following way: The risk constraint vector includes at least the risks of hypoglycemia, hypertension, blood pressure fluctuations, abnormal heart rate, exercise contraindications, falls, abnormal kidney function, and physician-mandated contraindications.

6. The system according to claim 1, characterized in that, The candidate suggestion generation unit generates candidate intervention suggestions when the triggering conditions of chronic disease knowledge rules are met, the required data credibility is not higher than the corresponding health data credibility, and the candidate intervention suggestions are not disabled by doctors. When the credibility of key data is lower than the generation threshold, the candidate suggestion generation unit does not generate automatic intervention suggestions based on the key data, but generates data supplementation suggestions or physician review tasks.

7. The system according to claim 1, characterized in that, The security access processing unit processes the candidate intervention suggestion set according to the following function. Perform security access control procedures: And output the processing action for each candidate intervention suggestion c_i: in, This indicates that an absolute taboo has been met and the candidate intervention recommendation has been eliminated. This indicates a reduction in the strength, frequency, or duration of candidate intervention recommendations; This indicates that candidate intervention recommendations will be marked for mandatory physician review; This indicates that the candidate intervention recommendations have been processed through security access procedures; Only for or The candidate intervention recommendations are used to calculate a comprehensive fit score.

8. The system according to claim 7, characterized in that, The security access processing unit is also used to perform conflict resolution on candidate intervention recommendations that have passed the security access processing. Conflict resolution includes target conflict resolution, time conflict resolution, risk conflict resolution, and execution burden conflict resolution. When conflicts exist, they should be handled according to the following priority: Absolute contraindication rules > Doctor-preset rules > High-risk threshold rules > Data quality rules > Multi-objective scoring results > Patient preference rules.

9. The system according to claim 1, characterized in that, The adaptation scoring output unit calculates candidate intervention recommendations according to the following formula. Overall compatibility score: in, This represents the expected improvement in the indicator. Implement accessibility for patients, For historical intervention response scores, To score the data quality, To improve the acceptance of doctors' medical history review, To fulfill the burden, This is the risk penalty value; The data quality score is: Risk penalty value: 。 10. A personalized health management intervention recommendation method for patients with chronic diseases, characterized in that, The method, applied to the system according to any one of claims 1-9, comprises: Collect multi-source health data from patients with chronic diseases and determine dynamic time windows based on patients' risk levels, fluctuations in key indicators, and data completeness; Within a dynamic time window, multi-source health data is preprocessed, synchronized, assessed for credibility, and gated for credibility state encoding to generate a patient state vector. A set of candidate intervention suggestions is generated based on the patient state vector and chronic disease knowledge rules; Before the comprehensive scoring, candidate intervention recommendations are eliminated, downgraded, subject to mandatory review, or approved based on risk constraints, physician constraints, and contraindication rules. Conflict resolution is also performed on candidate intervention recommendations that have been approved. A comprehensive fit score is calculated based on expected indicator improvement, patient accessibility, historical intervention response, data quality, physician historical review acceptance, implementation burden, and risk penalty, generating ranked intervention recommendations and corresponding explanatory information. Collect doctor review feedback and patient implementation feedback, and update individualized parameters based on the doctor review feedback and patient implementation feedback to generate the ranking results of intervention recommendations for the next cycle.

Images