Health management auxiliary decision-making method, system and computer readable storage medium

Abstract

A health management auxiliary decision-making method, system and computer readable storage medium, the method comprising the following steps: acquiring multi-modal data and preprocessing, and the preprocessed multi-modal data is organized into corresponding unified state objects; the unified state object is mapped into the corresponding health risk network, and the corresponding risk situation awareness data is output; based on the risk situation awareness data and the unified state object, prediction is carried out to obtain a health management scheme, the system is used for executing the above method, and the storage medium has a computer program stored thereon, and the computer program is executed by the processor to realize the above method. The health management auxiliary decision-making method, system and computer readable storage medium can make the originally scattered data be called and updated in the same state object, reduce the problem of multi-source data fragmentation, repeated processing and context loss, and enhance the traceability and safety boundary.

Description

Technical Field

[0001] This invention relates to the field of health information management technology, and in particular to a health management auxiliary decision-making method, system, and computer-readable storage medium. Background Technology

[0002] In the field of health management technology, the acquired multimodal data is directly used as input features and fed into the corresponding prediction model for processing, in order to fit the complex mapping relationship between monitoring data and future state.

[0003] However, the input data is scattered, making it difficult to maintain data quality throughout subsequent analysis. This results in recommendations lacking traceability and safety boundaries. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a health management auxiliary decision-making method, system and computer-readable storage medium that enables originally scattered data to be called and updated in the same state object, reducing the problems of multi-source data fragmentation, duplicate processing and context loss, and enhancing traceability and security boundaries.

[0005] This invention provides a health management decision support method, comprising the following steps:

[0006] Acquire multimodal data and preprocess it, then organize the preprocessed multimodal data into a corresponding unified state object;

[0007] The unified state object is mapped to the corresponding health risk network, and the corresponding risk situation awareness data is output.

[0008] Based on the risk situation awareness data and the unified state object, a prediction is made to obtain a health management plan.

[0009] In one embodiment, the step of acquiring multimodal data and preprocessing it, and then organizing the preprocessed multimodal data into a corresponding unified state object, further includes:

[0010] Acquire multimodal data and perform standardization processing;

[0011] Perform missing data detection and anomaly detection on standardized multimodal data;

[0012] The multimodal data after the detection is completed is scored for data quality.

[0013] Based on the missing data detection and anomaly detection results and the quality score, the multimodal data is organized into corresponding unified state objects.

[0014] In one embodiment, the step of performing data quality scoring on the multimodal data after detection further includes:

[0015] Q = a1×SourceScore+a2×CompletenessScore+a3×TimelinessScore + a4×ConsistencyScore + a5×RangeScore + a6×PermissionScore;

[0016] Among them, SourceScore represents the source credibility score, CompletenessScore represents the integrity score, TimelinessScore represents the timeliness score, ConsistencyScore represents the multi-source consistency score, RangeScore represents the reasonableness score of the numerical range, PermissionScore represents the data permission availability score, and a1 to a6 are preset weights.

[0017] In one embodiment, organizing the multimodal data into corresponding unified state objects based on missing data detection, anomaly detection results, and quality scores further includes:

[0018] Based on the missing and anomaly detection results and quality scores, the multimodal data is structured to form individual digital twin state objects.

[0019] Set corresponding read / write boundaries for each field of each individual digital twin state object;

[0020] Obtain the audit trail of each individual digital twin state object;

[0021] The unified state object is formed based on the individual digital twin state object, read / write boundaries, and audit trail.

[0022] In one embodiment, mapping the unified state object to a corresponding health risk network and outputting corresponding risk situation awareness data further includes:

[0023] Nodes and edges are constructed to form a health risk network;

[0024] The unified state object is mapped to the health risk network to obtain key risk-driven nodes, interventionable nodes, and non-interventionable nodes.

[0025] In one embodiment, the step of making predictions based on the risk situation awareness data and the unified state object to obtain a health management plan further includes:

[0026] The corresponding prediction model is invoked based on the task type and data type.

[0027] The risk situation awareness data and the unified state object are input into the prediction model to obtain the baseline trend and candidate solution trend.

[0028] The health management plan is obtained based on the baseline trend and candidate plan trend.

[0029] In one embodiment, obtaining the health management plan based on the baseline trend and candidate plan trend further includes:

[0030] Several candidate solutions are generated based on the baseline trend, candidate solution trend, and health management rule base.

[0031] Candidate solutions are screened based on constraints, evidence constraints, and security reviews to obtain the health management solution.

[0032] In one embodiment, the decision-making method further includes:

[0033] A follow-up task is generated based on the health management plan;

[0034] The unified state object is updated based on the follow-up results.

[0035] The present invention also provides a health management auxiliary decision-making system for implementing the method described in any one of the above, comprising:

[0036] The multimodal data access and quality control module is used to acquire multimodal data and perform preprocessing.

[0037] The digital twin state modeling module organizes the preprocessed multimodal data into corresponding unified state objects;

[0038] The multimodal core processing engine is used to map the unified state object to the corresponding health risk network and output the corresponding risk situation awareness data.

[0039] The health management plan generation module is used to make predictions based on the risk situation awareness data and the unified state object to obtain a health management plan.

[0040] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the method described in any of the preceding claims.

[0041] The health management auxiliary decision-making method, system, and computer-readable storage medium provided by this invention enable previously scattered data to be called and updated in the same state object, reducing the problems of multi-source data fragmentation, duplicate processing, and context loss, and enhancing traceability and security boundaries. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart illustrating the health management auxiliary decision-making method provided by the present invention.

[0044] Figure 2 This is a system block diagram of the health management auxiliary decision-making system provided by the present invention. Detailed Implementation

[0045] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. Based on the description of the present invention, all other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of the present invention.

[0046] In the description of this invention, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium. Those skilled in the art can understand the specific meaning of the above terms according to the specific circumstances.

[0047] The terms “upper,” “lower,” “left,” “right,” “front,” “back,” “top,” “bottom,” “inner,” and “outer,” etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use. They are only for the convenience of description and simplification, 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. Therefore, they should not be construed as limitations on the present invention.

[0048] The terms “first,” “second,” “third,” etc., are used merely to distinguish elements with similar properties, not to indicate or imply relative importance or a specific order.

[0049] The terms “include,” “comprising,” or any other variation thereof are intended to cover non-exclusive inclusion, which includes not only the elements listed but also other elements not expressly listed.

[0050] Example 1

[0051] Please see Figure 1 The health management auxiliary decision-making method provided by the present invention includes the following steps:

[0052] S1: Acquire multimodal data and preprocess it, then organize the preprocessed multimodal data into a corresponding unified state object.

[0053] It is understandable that the above steps may further include:

[0054] S101, acquire multimodal data and perform standardization processing.

[0055] It is known that the sources of multimodal data can include:

[0056] The data received by this module includes at least one or more of the following:

[0057] (1) Physical examination data, such as height, weight, BMI, waist circumference, blood pressure, fasting blood glucose, glycated hemoglobin, blood lipids, liver function, kidney function, uric acid, inflammatory markers, tumor markers, etc.;

[0058] (2) Clinical data, such as past medical history, medication history, allergy history, examination and test results, medical records, diagnostic records, follow-up records, etc.;

[0059] (3) Physiological time series data, such as heart rate, resting heart rate, steps, exercise duration, sleep duration, sleep stages, blood oxygen, body temperature, continuous blood glucose, etc. collected by wearable devices;

[0060] (4) Lifestyle data, such as diet, water intake, alcohol consumption, smoking, exercise habits, work intensity, sedentary time, and work-rest patterns;

[0061] (5) Questionnaire data, such as health risk questionnaires, sleep questionnaires, stress questionnaires, diet questionnaires, exercise ability questionnaires, compliance questionnaires, quality of life questionnaires, etc.

[0062] (6) Omics data, such as gut microbiota detection data, metagenomic data, metabolomics data, proteomics data, genetic risk data, etc.;

[0063] (7) Environmental exposure data, such as air quality, temperature and humidity, noise, geographical exposure, occupational exposure, seasonal factors, etc.;

[0064] (8) Preference constraint data, such as the target individual's dietary preferences, dietary restrictions, religious or cultural dietary restrictions, exercise preferences, available time, budget, willingness to implement, acceptable intervention intensity, willingness to follow up, etc.

[0065] After standardization, each standardized data item includes at least: field name; original value; standardized value; standard unit; original unit; collection time; data source; data source type; confidence level; data version; permission flag; data quality flag; anomaly flag; missing data flag; data hash value; write time; and write subject.

[0066] For example, for the data "fasting blood glucose 6.3 mmol / L, from a physical examination report, collected on June 1, 2026", the system converts it into the following structured data items:

[0067] {

[0068] "field_name": "fasting_glucose",

[0069] "raw_value": "6.3",

[0070] "standard_value": 6.3,

[0071] "raw_unit": "mmol / L",

[0072] "standard_unit": "mmol / L",

[0073] "timestamp": "2026-06-01",

[0074] "source": "physical_examination_report",

[0075] "source_type": "laboratory_test",

[0076] "confidence": 0.92,

[0077] "data_version": "2026-06-01-v1",

[0078] "permission_tag": "health_management_use",

[0079] "quality_flags": [],

[0080] "outlier_flag": false,

[0081] "missing_flag": false

[0082] }

[0083] For data from different sources and with different units, the system unifies the data according to preset unit conversion tables, medical terminology tables, indicator dictionaries, and field mapping rules. For example, weight is unified to kg, height to cm, blood glucose to preset standard units such as mmol / L or mg / dL, and time to ISO format. If the system cannot confirm the meaning of a unit or field, it does not directly infer a definite value but generates a "manual confirmation required" flag.

[0084] S102 performs missing data detection and anomaly detection on standardized multimodal data.

[0085] It is known that missing detection can include:

[0086] (1) Detection of missing key fields, such as age, gender, height, weight, collection time, and data source;

[0087] (2) Detection of missing essential fields in health management scenarios, such as missing waist circumference in weight management scenarios, missing glycated hemoglobin in blood glucose management scenarios, and missing sleep duration in sleep management scenarios;

[0088] (3) Detection of missing data in continuous time series, such as missing data for more than 3 consecutive days in the data of wearable devices in the past 30 days;

[0089] (4) Questionnaire timeliness detection, such as the questionnaire being completed after the preset validity period;

[0090] (5) Detection of missing annotations in omics data, such as missing sequencing platforms, database versions, quality control indicators, or species annotation levels;

[0091] Anomaly detection can include:

[0092] (1) Abnormal numerical range, such as height, weight, blood sugar, blood pressure, etc. exceeding the reasonable physiological range;

[0093] (2) Abnormal units, for example, the same indicator is used in a mixed manner with mg / dL and mmol / L but the units are not labeled;

[0094] (3) Time anomalies, such as the collection time being later than the current time, or contradictory timestamps in the same report;

[0095] (4) Multi-source conflict anomaly, such as the difference of the same indicator in records from different sources within the same date exceeds the preset threshold;

[0096] (5) Abnormal trend change, such as a non-physiological sudden increase or decrease in a certain indicator in continuous time series data.

[0097] S103, perform data quality scoring on the multimodal data after the detection is completed.

[0098] It is known that the quality score can be determined by the credibility of the source, completeness, timeliness, consistency, degree of anomaly, and accessibility.

[0099] The quality score Q for a single data point can be expressed as:

[0100] Q = a1×SourceScore+a2×CompletenessScore+a3×TimelinessScore + a4×ConsistencyScore + a5×RangeScore + a6×PermissionScore;

[0101] Among them, SourceScore represents the source credibility score, CompletenessScore represents the integrity score, TimelinessScore represents the timeliness score, ConsistencyScore represents the multi-source consistency score, RangeScore represents the reasonableness score of the numerical range, PermissionScore represents the data permission availability score, and a1 to a6 are preset weights, which can be preset by the system or pre-configured by the organization.

[0102] When the Q value is higher than the first threshold, the data can proceed to the subsequent analysis process; when Q is between the first and second thresholds, the data can be used as low-confidence information in the analysis, but the data limitation needs to be indicated in the report; when Q is lower than the second threshold, the system generates supplementary sampling suggestions or manual confirmation tasks, and does not use it as a key decision-making basis.

[0103] For example, the system can be configured to have Q ≥ 0.80 as passing quality control, 0.50 ≤ Q < 0.80 as limited availability, and Q < 0.50 as not directly usable. These thresholds can be configured according to institutional standards, business scenarios, and data types.

[0104] S104, based on the results of missing data detection and anomaly detection and the quality score, organizes the multimodal data into corresponding unified state objects.

[0105] It is understood that the above steps may further include:

[0106] The multimodal data is structured based on the results of missing and anomaly detection and quality scores to form individual digital twin state objects.

[0107] Set corresponding read / write boundaries for each field of each individual digital twin state object.

[0108] It is known that the purpose of controlled read / write boundaries for fields is to avoid data overwriting, context pollution, and untraceable modifications during multi-agent collaboration. The system sets read / write boundaries for each field of TwinState.

[0109] The data access agent only writes signals.raw_data and signals.parsed_data;

[0110] The data quality control agent only writes to data_quality and signals.standardized_data;

[0111] Phenotypic modeling agents only write to phenotypes;

[0112] The evidence retrieval agent only writes to evidence_context or safety_review.evidence;

[0113] The causal network agent only writes to risk_network;

[0114] The trend prediction agent only writes predictions;

[0115] The intervention simulation agent only writes to intervention_candidates;

[0116] The security review agent only writes to safety_review and review_status;

[0117] The report-generating agent is written to final_plan only after the security review has passed;

[0118] The human review node can be written to safety_review.human review and final_plan.reviewed_output.

[0119] Each field write generates a differentiated patch, including the write body, write time, input summary, output summary, model version, prompt word version, validation result, and change hash value. Concurrent writes are handled using version numbers or optimistic locking mechanisms. When field versions are inconsistent, overwrite writes are rejected, and a conflict resolution task is generated.

[0120] Obtain the audit trail of each individual digital twin state object.

[0121] It is known that audit_trace records key events throughout the entire system process, including: data upload events; data parsing events; quality control scoring events; supplementary data collection suggestion events; TwinState version update events; model invocation events; evidence retrieval events; risk network construction events; trend prediction events; intervention simulation events; security rule interception events; manual review events; report generation events; user viewing events; and follow-up feedback events.

[0122] Each record in the audit trail includes at least the following fields: event_id, event_type, operator, timestamp, input_summary, output_summary, state_version_before, state_version_after, model_name, model_version, prompt_version, rule_version, and result_hash. Through this mechanism, the system can trace which data, evidence, models, and review processes were used to generate any health management recommendation.

[0123] A unified state object is formed based on individual digital twin state objects, read / write boundaries, and audit trails.

[0124] It is known that a unified state object can include the following fields:

[0125] {

[0126] "person_id": "",

[0127] "data_version": "",

[0128] "profile": {},

[0129] "signals": {},

[0130] "data_quality": {},

[0131] "phenotypes": [],

[0132] "risk_network": {},

[0133] "predictions": [],

[0134] "intervention_candidates": [],

[0135] "safety_review": {},

[0136] "final_plan": {},

[0137] "review_status": "",

[0138] "audit_trace": []

[0139] }

[0140] in:

[0141] (1) person_id is used to identify the target individual and can be the de-identified user ID;

[0142] (2) data_version is used to record the current TwinState version;

[0143] (3) Profile is used to store demographic information and basic records, such as age, gender, height, weight, medical history, etc.;

[0144] (4) Signals are used to store multi-source health signals, including data from physical examinations, clinical data, wearable devices, questionnaires, lifestyle data, omics data, and environmental exposure data.

[0145] (5) data_quality is used to store quality scores, missing fields, anomaly markers, data source credibility, and re-collection tasks;

[0146] (6) phenotypes are used to store phenotype labels, such as weight management concern, insufficient sleep, insufficient exercise, blood sugar risk warning, blood lipid abnormality risk warning, etc.

[0147] (7) risk_network is used to store the nodes, edges, key drivers, and interventionable nodes of the health risk network;

[0148] (8) predictions are used to save baseline trend prediction results and candidate trend prediction results;

[0149] (9) `intervention_candidates` is used to store candidate health management plans and their ranking results;

[0150] (10) safety_review is used to save evidence retrieval results, rule interception results, manual review status, and security rewrite text;

[0151] (11) final_plan is used to store the final health management support plan that has passed the security review;

[0152] (12) review_status is used to record the current review status, such as pending, blocked, pending_human_review, approved, rejected, delivered;

[0153] (13) audit_trace is used to save records of each data modification, model call, rule trigger, manual review and report generation.

[0154] S2 maps the unified state object to the corresponding health risk network and outputs the corresponding risk situation awareness data.

[0155] It is understandable that the above steps may further include:

[0156] S201 involves constructing nodes and edges to form a health risk network.

[0157] It is known that the nodes of the health risk network include at least:

[0158] 1) Demographic nodes, such as age, sex, height, and weight;

[0159] 2) Clinical milestones, such as past medical history, medication history, and allergy history;

[0160] 3) Testing points, such as blood glucose, blood lipids, liver and kidney function, uric acid, and inflammatory markers;

[0161] 4) Physiological markers, such as heart rate, sleep, exercise, blood oxygen, and continuous glucose levels;

[0162] 5) Omics nodes, such as microbial diversity, specific genera, metabolic pathways, and genetic risks;

[0163] 6) Behavioral factors, such as diet, exercise habits, sleep patterns, alcohol consumption, and smoking;

[0164] 7) Environmental factors, such as air quality, temperature and humidity, and occupational exposure;

[0165] 8) Preference nodes, such as time budget, cost budget, taste preference, exercise preference;

[0166] 9) Outcome milestones, such as weight management goals, sleep improvement goals, and blood sugar risk management goals.

[0167] The boundaries of health risks include at least:

[0168] The edges of a health risk network should include at least the following: chronological edges; relevance edges; potential contribution edges; intervention impact edges; and evidence-supporting edges.

[0169] The temporal precedence edge is used to indicate that one health event or indicator change precedes another event or indicator change in time. For example, when sleep deprivation precedes a trend of weight gain or a trend of elevated fasting blood glucose over a continuous period of time, the system can generate a temporal precedence edge.

[0170] A relevance edge indicates that two nodes are statistically or empirically correlated in target individual data, similar population trajectories, or literature evidence. A relevance edge should include the correlation strength, data source, statistical method, and confidence level.

[0171] A potentially contributing edge is used to indicate that a factor may contribute to a certain risk state or health management goal. The system must not directly express this edge as a definite causal relationship, but should use expressions such as "potentially relevant," "potentially contributing," "hint," or "requires further confirmation" in conjunction with the confidence level.

[0172] The term "intervention impact" is used to indicate that a health management measure may affect a particular indicator or goal. For example, increasing moderate-intensity exercise may affect weight management, sleep quality, and cardiorespiratory fitness; improving sleep patterns may affect daytime fatigue and metabolic risk.

[0173] Supporting edges are used to connect risk nodes, intervention nodes, and evidence sources. Each supporting edge should include the evidence source, evidence level, applicable population, limiting conditions, and evidence summary.

[0174] Each node includes at least the following fields: node_id, node_type, name, value, unit, timestamp, source, confidence, abnormal_flag, modifiable_flag, and evidence_level.

[0175] S202 maps the unified state object to the health risk network to obtain key risk-driven nodes, interventionable nodes, and non-interventionable nodes.

[0176] S203, the health risk network can calculate key risk-driven nodes based on node anomaly degree, node manipulation, edge weight, evidence strength, centrality, and target relevance.

[0177] The key driver score, DriverScore, can be expressed as:

[0178] DriverScore = b1×Abnormality + b2×Modifiability + b3×EvidenceStrength + b4×NetworkCentrality + b5×GoalRelevance - b6 ×Uncertainty;

[0179] Here, Abnormality represents the degree of abnormality, Modifiability represents the modifiability, EvidenceStrength represents the strength of evidence, NetworkCentrality represents network centrality, GoalRelevance represents the relevance to health management goals, Uncertainty represents the uncertainty penalty, and b1 to b6 are preset or configurable weights.

[0180] For factors that cannot be intervened, such as age, gender, and medical history, these factors are only used as risk background or stratification variables; for factors that can be intervened, such as sleep duration, exercise volume, diet, and frequency of follow-up examinations, these factors can be used as targets for candidate health management programs.

[0181] S3 uses risk situation awareness data and unified state objects to make predictions in order to obtain health management solutions.

[0182] It is understandable that the above steps may further include:

[0183] S301, call the corresponding prediction model according to the task type and data type.

[0184] Understandably, various specialized prediction models can be invoked, including but not limited to: tree models; generalized linear models; time series models; deep time series models; similar population trajectory models; Bayesian models; rule-enhanced prediction models; and multi-model ensemble models. Model invocation can be implemented using model gateway units, which are used to invoke different models based on task type and record model invocation information. The model gateway can connect to: locally deployed large language models; multimodal models; vector embedding models; re-ranking models; specialized prediction models; classification models; risk scoring models; and rule engines.

[0185] The model gateway unit routes tasks based on their type. For example: data parsing tasks call document parsing models or multimodal models; evidence retrieval tasks call vector models and re-ranking models; trend prediction tasks call specialized prediction models; report generation tasks call large language models; security review tasks call rule engines, classification models, and large language models for verification; and risk network construction tasks call graph models, statistical models, or rule models.

[0186] Each model call records the model name, model version, prompt word version, input summary, output summary, output validation result, call time, call duration, and caller or call task number. Through the model gateway, the system supports private deployment, model replacement, model version rollback, and regulatory auditing.

[0187] S302, input risk situation awareness data and unified state objects into the prediction model to obtain baseline trends and candidate solution trends.

[0188] Understandably, among them,

[0189] 1) Baseline trend: This indicates the possible direction of change of the target indicator within a preset time range under the condition of no new health management intervention, that is, how the health status will change under the condition of no intervention;

[0190] 2) Candidate solution trend: This indicates the possible direction of change of the target indicator within a preset time range under the condition of implementing the candidate health management plan, that is, what corresponding changes will occur in the health status under the corresponding candidate health management plan.

[0191] Each of the above prediction results should include at least the following: prediction indicator; current value; prediction direction; prediction interval; prediction time range; confidence level; main influencing factors; data gap; model name; model version; input data version; and prediction generation time.

[0192] For example:

[0193] {

[0194] "metric": "fasting_glucose",

[0195] "current_value": 6.3,

[0196] "unit": "mmol / L",

[0197] "horizon": "3_months",

[0198] "scenario": "baseline",

[0199] "predicted_direction": "maintain_high_or_slightly_increase",

[0200] "prediction_interval": [6.1, 6.8],

[0201] "confidence": 0.68,

[0202] "key_factors": ["sleep_duration_low", "BMI_high", "physical_activity_low"],

[0203] "data_gaps": ["HbA1c_missing", "waist_circumference_missing"],

[0204] "model_name": "trajectory_model",

[0205] "model_version": "v1.2.0"

[0206] }

[0207] If the data is insufficient to support a reliable prediction, the system will not force the output of a numerical prediction. Instead, it will output a result indicating "Insufficient data", "For trend reference only", or "Data needs to be collected before prediction", and write the data gap into TwinStat.

[0208] S303, obtain health management plans based on baseline trends and candidate solution trends.

[0209] The above steps may further include:

[0210] S303a generates several candidate solutions based on baseline trends, candidate solution trends, and a health management rule base.

[0211] It is understandable that the health management rule base may include: diet management rule base; exercise management rule base; sleep management rule base; nutrition management rule base; follow-up recommendation rule base; follow-up management rule base; health education rule base; and behavior change rule base.

[0212] Each candidate intervention should include at least the following: intervention number; intervention name; intervention objective; implementation content; implementation frequency; duration; intensity level; preconditions; contraindications; expected benefits; risk level; compliance score; preference matching score; evidence score; time cost; cost; monitoring indicators; follow-up points; whether manual review is required; uncertainty statement; and alternative solutions.

[0213] Low-risk, gradual approaches among the candidate programs could include: increasing daily steps, maintaining a fixed sleep schedule, reducing sugary drinks, and submitting weight and sleep data after two weeks.

[0214] A comprehensive and intensive program may include dietary adjustments, increased exercise frequency, regular sleep patterns, and follow-up reminders; an extreme dieting program should be marked as not recommended or require manual review.

[0215] S303b, based on evidence constraints and safety reviews, screens candidate solutions to obtain health management solutions.

[0216] Constraints are divided into hard constraints and soft constraints.

[0217] Hard constraints include, but are not limited to: explicit contraindications; serious abnormal indicators; high-risk population markers; interventions that users explicitly cannot accept; restrictions manually set by doctors or health managers; programs that exceed the system's security boundaries; and content involving adjustments to diagnoses, prescriptions, or medications.

[0218] Candidate solutions that violate hard constraints cannot be output as directly recommended solutions; they can only be marked as "not recommended solutions" or "requires manual review".

[0219] Soft constraints include, but are not limited to: time budget; cost budget; dietary preferences; exercise preferences; willingness to perform; past compliance; limitations of home or work environment; and ease of follow-up.

[0220] Soft constraints do not necessarily prevent a solution from being blocked, but they will affect the overall score of the solution.

[0221] Evidence constraints and security reviews are used to trace evidence, assess suitability, intercept security rules, and perform manual review and gating before candidate solutions and report texts are delivered.

[0222] Evidence constraints can involve retrieving health management-related evidence from a local knowledge base. This local knowledge base may include: guidelines; expert consensus; institutional rule bases; product manuals; internal SOPs; health education materials; risk warning rules; and intervention rule descriptions.

[0223] Evidence retrieval can employ vector retrieval, full-text retrieval, keyword retrieval, graph retrieval, or a combination thereof. The system generates retrieval queries based on the target individual's phenotypic labels, risk network nodes, candidate intervention plans, and security review tasks, and returns evidence fragments, evidence sources, evidence levels, applicable conditions, limitations, and update times.

[0224] Determine whether the retrieved evidence is applicable to the target individual and the current candidate intervention. Suitability assessment should consider at least: age; gender; health risk level; medical history; medication history; allergy history; data quality; target indicators; intervention intensity; contraindications; internal institutional rules; user preferences; and implementation conditions.

[0225] If the evidence is only applicable to a specific population or under specific conditions, the system should clearly indicate the applicable limitations in the report. If the evidence does not match the target individual's conditions, the evidence score should be lowered or the candidate solution should be blocked.

[0226] Hard rule security blocking is used to block content that should not be automatically output by the system. It can be implemented using a combination of rule tables, keyword matching, medical entity recognition, classification models, and large language models for review.

[0227] Among these, the aforementioned hard rules include at least:

[0228] (1) Diagnostic statement interception rules

[0229] When candidate solutions or report texts contain diagnostic or therapeutic expressions such as "diagnosed as," "confirmed," "judged as a certain disease," or "treatment plan," the system should block direct output and rewrite it as a health management risk warning or a suggestion to consult a doctor.

[0230] (2) Rules for intercepting prescription-related statements

[0231] When the text contains prescription-like expressions such as "prescription," "taking a certain drug," or "using a certain drug for treatment," the system should block direct output.

[0232] (3) Drug-adjusted interception rules

[0233] When the text contains expressions related to drug adjustments such as "stop medication," "add medication," "reduce medication," "change medication," or "adjust dosage," the system should block direct output and prompt that a doctor's judgment is required.

[0234] (4) Rules for intercepting serious abnormal indicators

[0235] When the system identifies severely abnormal indicators or high-risk combinations of indicators, it does not directly generate general health management suggestions, but instead initiates a manual review process or prompts the user to seek professional medical evaluation as soon as possible.

[0236] (5) High-risk sports interception rules

[0237] When candidate programs include high-intensity exercise, extreme load training, or strenuous exercise without assessment prerequisites that are unsuitable for the target individual's current condition, the system should mark them as requiring manual review or not recommended.

[0238] (6) Extreme diet interception rules

[0239] When candidate options include very low-energy diets, very low-carb diets, long-term fasting, extreme dieting, or single-food substitutions, the system should block direct recommendations.

[0240] (7) Supplement combination interception rules

[0241] When a candidate regimen contains an unapproved combination of supplements, supplements that may conflict with the patient's medication history, or combinations with unclear dosages or unknown safety, the system should initiate a manual review process.

[0242] (8) Safety Statement Rules

[0243] All user-generated reports should include a statement that "This report is intended to assist in health management decisions and does not replace medical treatment."

[0244] When an unsafe expression is detected, but the risk can be mitigated by rewriting it, an alternative expression can be generated. For example:

[0245] Rewrite "You already have a certain disease" as "Relevant indicators suggest that there may be health risks. It is recommended to consult with a doctor or health manager for further confirmation."

[0246] Rewrite "stop taking a certain medication" as "Consult a doctor about any issues related to medication use; it is not recommended to adjust medication on your own."

[0247] The phrase "immediately begin high-intensity training" should be rewritten as "the amount of activity can be gradually increased after professional evaluation."

[0248] Securely rewritten text still needs to be reviewed according to rules, and if necessary, it will be placed in a manual review queue.

[0249] The manual review queue is used to process candidate solutions and report content that cannot be automatically approved. Manual review tasks must include at least: target individual summary; data quality report; key nodes in the risky network; candidate solutions; triggered security rules; retrieved evidence; system-generated security rewrite suggestions; reviewer comments; review conclusion; review time; and review version.

[0250] Manual review results include approval, return for modification, rejection of output, request for supplementary data, and recommendation for offline medical evaluation. The review results are written to the `safety_review` and `audit_trace` fields of `TwinState`.

[0251] Multiple candidate health management plans are generated by combining intervention rule bases including diet, exercise, sleep, nutrition, follow-up examinations, and health education. Each candidate plan includes implementation content, frequency, duration, expected benefits, risk level, compliance score, preference matching score, evidence score, monitoring indicators, and a marker indicating whether manual review is required.

[0252] The overall score of the candidate solutions can be calculated as follows:

[0253] Score = w1×Benefit + w2×Adherence + w3×Preference + w4×Evidence -w5×Risk - w6×DataGap;

[0254] Where Benefit is the expected return score, Adherence is the compliance score, Preference is the preference matching score, Evidence is the evidence score, Risk is the risk penalty item, DataGap is the data gap penalty item, and w1 to w6 are the preset or institutionally configured weights.

[0255] When a candidate protocol contains high-intensity exercise, extreme diets, supplement combinations, or is associated with severe abnormal indicators, the intervention projection phantom unit does not directly recommend the protocol, but instead marks it as requiring manual review or not recommended.

[0256] Based on the above health management plan, two types of reports can be generated:

[0257] 1) User Version Report

[0258] The user-version report is intended for target individuals or general health management users, and the language should be plain, cautious, and actionable. Content may include: a summary of current health status; data sources and data integrity guidelines; key health management concerns; key risk drivers; digital twin simulation results; recommended solutions; alternative solutions; practices not recommended; issues requiring confirmation from a doctor or health manager; follow-up plan; and a safety statement.

[0259] 2) Professionally audited report

[0260] The professional audit version report is intended for doctors, health managers, or institutional auditors. Content may include: data sources and quality scores; missing and outlier data; phenotypic labels; key nodes and edges of the risk network; trend prediction results; comparison of candidate solutions; evidence citations and suitability assessments; security review records; items reviewed manually; recommended follow-up indicators; and a summary of the audit trajectory.

[0261] S4 generates follow-up tasks based on the health management plan; updates the unified status object based on the follow-up results.

[0262] The system automatically generates follow-up tasks based on the final health management plan. Each follow-up task includes at least: follow-up time; follow-up method; data to be transmitted; questionnaires to be completed; indicators to be reviewed; task reminder method; overdue handling rules; responsible person; and task status.

[0263] For example, the system can be set to conduct follow-up visits for two weeks, one month, and three months. The two-week follow-up mainly collects data on implementation records, weight, sleep, and exercise; the one-month follow-up collects data on changes in periodic indicators and compliance; and the three-month follow-up collects data for review and the basis for adjusting the plan.

[0264] After the follow-up data is fed back, the actual changes are compared with the trend predictions, and TwinState is updated.

[0265] The updates include: adding health data; data quality scores; user compliance trends; intervention program implementation status; changes in target indicators; risk network edge weights; candidate program scoring parameters; trend prediction model inputs; and follow-up plans.

[0266] For example, if users' actual execution rate is low, the system will lower the compliance score for similar high-burden solutions; if a low-risk solution shows good improvement in metrics after execution, the system can raise the expected benefit score for that type of solution in similar scenarios. Through this mechanism, the system forms a closed loop of "data—modeling—deduction—review—execution—feedback—remodeling".

[0267] Example 2

[0268] Please see Figure 2 This embodiment provides a health management auxiliary decision-making system, a method for any of the above, comprising:

[0269] The multimodal data access and quality control module is used to acquire multimodal data and perform preprocessing.

[0270] The digital twin state modeling module organizes the preprocessed multimodal data into corresponding unified state objects.

[0271] Understandably, the digital twin state modeling module can be the TwinState digital twin state modeling module.

[0272] TwinState can use the following field structure:

[0273] {

[0274] "person_id": "demo_user_001",

[0275] "data_version": "yyyy-mm-dd-vn",

[0276] "profile": {"age": 0, "sex": "unknown", "height_cm": 0, "weight_kg": 0},

[0277] "signals": {"clinical_labs": [], "wearable": [], "questionnaire":[], "lifestyle": [], "omics": [], "environment": []},

[0278] "data_quality": {"score": 0, "missing_fields": [], "outlier_flags":[]},

[0279] "phenotypes": [],

[0280] "risk_network": {"nodes": [], "edges": [], "key_drivers": []},

[0281] "predictions": [],

[0282] "intervention_candidates": [],

[0283] "safety_review": {},

[0284] "final_plan": {},

[0285] "review_status": "pending_human_review",

[0286] "audit_trace": []

[0287] }

[0288] The multimodal core processing engine is used to map a unified state object to a corresponding health risk network and output the corresponding risk situation awareness data.

[0289] The health management plan generation module is used to make predictions based on risk situation awareness data and unified status objects to obtain health management plans.

[0290] Example 3

[0291] This embodiment provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described above.

[0292] As described above, the health management auxiliary decision-making method, system, and computer-readable storage medium provided by this invention enable previously scattered data to be accessed and updated within the same state object, reducing the problems of fragmented multi-source data, redundant processing, and context loss, while also enhancing traceability and security boundaries.

[0293] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the appended claims.

Claims

1. A health management decision support method, characterized in that, Includes the following steps: Acquire multimodal data and preprocess it, then organize the preprocessed multimodal data into a corresponding unified state object; The unified state object is mapped to the corresponding health risk network, and the corresponding risk situation awareness data is output. Based on the risk situation awareness data and the unified state object, a prediction is made to obtain a health management plan.

2. The health management auxiliary decision-making method as described in claim 1, characterized in that, The process of acquiring and preprocessing multimodal data, and organizing the preprocessed multimodal data into corresponding unified state objects, further includes: Acquire multimodal data and perform standardization processing; Perform missing data detection and anomaly detection on standardized multimodal data; The multimodal data after the detection is completed is scored for data quality. Based on the missing data detection and anomaly detection results and the quality score, the multimodal data is organized into corresponding unified state objects.

3. The health management auxiliary decision-making method as described in claim 1, characterized in that, The step of performing data quality scoring on the multimodal data after detection further includes: Q = a1×SourceScore+a2×CompletenessScore+a3×TimelinessScore + a4×ConsistencyScore + a5×RangeScore + a6×PermissionScore; Among them, SourceScore represents the source credibility score, CompletenessScore represents the integrity score, TimelinessScore represents the timeliness score, ConsistencyScore represents the multi-source consistency score, RangeScore represents the reasonableness score of the numerical range, PermissionScore represents the data permission availability score, and a1 to a6 are preset weights.

4. The health management auxiliary decision-making method as described in claim 2, characterized in that, The step of organizing the multimodal data into corresponding unified state objects based on missing and anomaly detection results and quality scores further includes: Based on the missing and anomaly detection results and quality scores, the multimodal data is structured to form individual digital twin state objects. Set corresponding read / write boundaries for each field of each individual digital twin state object; Obtain the audit trail of each individual digital twin state object; The unified state object is formed based on the individual digital twin state object, read / write boundaries, and audit trail.

5. The health management auxiliary decision-making method as described in claim 1, characterized in that, The step of mapping the unified state object to a corresponding health risk network and outputting the corresponding risk situation awareness data further includes: Nodes and edges are constructed to form a health risk network; The unified state object is mapped to the health risk network to obtain key risk-driven nodes, interventionable nodes, and non-interventionable nodes.

6. The health management auxiliary decision-making method as described in claim 5, characterized in that, The step of making predictions based on the risk situation awareness data and the unified state object to obtain a health management plan further includes: The corresponding prediction model is invoked based on the task type and data type. The risk situation awareness data and the unified state object are input into the prediction model to obtain the baseline trend and candidate solution trend. The health management plan is obtained based on the baseline trend and candidate plan trend.

7. The health management auxiliary decision-making method as described in claim 6, characterized in that, The step of obtaining the health management plan based on the baseline trend and candidate plan trend further includes: Several candidate solutions are generated based on the baseline trend, candidate solution trend, and health management rule base. Candidate solutions are screened based on constraints, evidence constraints, and security reviews to obtain the health management solution.

8. The health management auxiliary decision-making method as described in claim 1, characterized in that, The decision-making method also includes: A follow-up task is generated based on the health management plan; The unified state object is updated based on the follow-up results.

9. A health management auxiliary decision-making system, characterized in that, For implementing the method of any one of claims 1 to 8, comprising: The multimodal data access and quality control module is used to acquire multimodal data and perform preprocessing. The digital twin state modeling module organizes the preprocessed multimodal data into corresponding unified state objects; The multimodal core processing engine is used to map the unified state object to the corresponding health risk network and output the corresponding risk situation awareness data. The health management plan generation module is used to make predictions based on the risk situation awareness data and the unified state object to obtain a health management plan.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the method as described in any one of claims 1-8.

Images