Dialectical nursing health record management system and evaluation method suitable for prediabetes

Abstract

The application discloses a kind of pre-diabetes syndrome differentiation and nursing health record management system and evaluation method, it is related to diabetes technical field. Including: form pre-diabetes syndrome differentiation and nursing original health record collection;Get time sequence health record feature set with credibility mask;Form stratified health record feature sequence;Based on stratified health record feature sequence Monarch Mixer sparse mixed coding network is constructed, and stratified sparse mixed feature representation is obtained;Based on stratified sparse mixed feature representation, stratified feature decoupling processing is carried out, and stratified decoupling health record latent variable set is obtained;Form syndrome differentiation and nursing correlation score set;Get pre-diabetes syndrome differentiation and nursing health record evaluation result.The application reduces redundant interaction while retaining key cross-layer relationship, so that health record feature expression is more sparse, stable and has hierarchical semantics.

Description

Technical Field

[0001] This invention relates to the field of diabetes technology, and in particular to a health record management system and assessment method suitable for the diagnosis and treatment of prediabetes. Background Technology

[0002] With the increasing demand for health management among people with prediabetes, risk assessment and syndrome differentiation-based nursing care analysis based on health records are gradually becoming an important foundation for chronic disease nursing management.

[0003] Existing health record assessment methods typically rely on single physical examination indicators, manual experience judgment, or simple scale statistics, making it difficult to simultaneously handle multi-source heterogeneous data such as glucose metabolism indicators, information from the four diagnostic methods of traditional Chinese medicine, lifestyle behaviors, and care compliance. Due to differences in field naming, collection time, unit standards, missing data, and reliability among different data sources, existing methods are prone to problems such as inconsistent data merging, mixing of historical and recent states, and outliers and missing values ​​affecting the stability of the assessment.

[0004] Existing diagnostic and nursing care assessments often analyze syndrome manifestations, metabolic risks, and nursing intervention needs separately, lacking a unified model of the relationship between characteristics of multi-layered health records. They also struggle to distinguish the redundant coupling relationships between glucose metabolism risk, TCM syndrome manifestations, lifestyle interventions, and nursing compliance, resulting in a lack of stable data support for syndrome tendency scoring and nursing care priority ranking. Summary of the Invention

[0005] One objective of this invention is to propose a health record management system and assessment method suitable for syndrome differentiation and care in prediabetes. This invention reduces redundant interactions while retaining key cross-layer relationships, making the health record feature expression more sparse, stable and hierarchical semantic.

[0006] A method for managing and assessing health records for prediabetes based on syndrome differentiation and care, according to an embodiment of the present invention, includes: Multi-source health record data of the subjects to be evaluated were collected and fields were standardized and merged to form a set of original health records for the diagnosis and treatment of prediabetes. Based on the original health record set of prediabetes diagnosis and treatment and after preprocessing, a set of time-series health record features with confidence mask is obtained. Based on the feature set of time-series health records with credibility masks, the features are grouped according to the glucose metabolism risk layer, the TCM syndrome manifestation layer, the lifestyle behavior layer, and the care compliance layer to form a hierarchical health record feature sequence. Based on the feature sequences of hierarchical health records, a Monarch Mixer sparse hybrid coding network is constructed to obtain hierarchical sparse hybrid feature representations; Hierarchical feature decoupling is performed based on hierarchical sparse hybrid feature representation to obtain a hierarchical decoupled set of latent variables in health records. Based on the hierarchical decoupled set of latent variables in health records, a syndrome-nursing care correlation assessment matrix is ​​constructed to form a set of syndrome-nursing care correlation scores. Based on the syndrome differentiation and nursing care correlation score set combined with data credibility weight and time decay weight, the scores of each syndrome tendency and each nursing care priority are calibrated and integrated to generate a comprehensive health record assessment value, syndrome priority ranking and nursing intervention priority ranking, and obtain the health record assessment results of syndrome differentiation and nursing care for prediabetes.

[0007] Optionally, the collection and field unification of multi-source health record data of the object to be evaluated, and the merging of data, include: A master index for health record collection is established for the subjects to be evaluated, resulting in a set of master index records for the subjects; Based on the object's master index record set, multi-source health record data of the object to be evaluated is collected. During the collection, the data source, collection method, original field name, original field value, collection time, and record responsibility entity corresponding to each data item are retained to form a multi-source original collection record set. Based on a set of original data collected from multiple sources, a unified dictionary of fields is established to form a unified mapping set of fields. Based on the unified field mapping set, the multi-source original collection record set is classified by field. Fields that do not belong to the same semantic category are stored separately, and fields that belong to the same semantic category but have different sources are merged into the same category. Each merged field is attached with a field category identifier and a source category identifier to form a health record record set classified by field semantics. Based on the health record record set categorized by field semantics, records of the same object to be evaluated are aggregated according to object identifier. Records of the same object formed at different collection times are established in chronological order according to the collection time. Records of the same object from different data sources but collected at the same or similar time are archived in parallel according to the data source. Record status identifiers are set for duplicate records, supplementary records, revised records and conflicting records respectively, forming a time-series merged record set at the object dimension. Based on the object-dimensional time-series merged record set, each health record is checked for completeness and traceability. Records that pass the checks are retained as valid merged records, while those that fail the checks are retained as merged records pending verification, forming a merged health record set with record status identifiers, thus forming the original health record set for prediabetes diagnosis and care.

[0008] Optionally, the preprocessing of the original health record set based on the syndrome differentiation and care for prediabetes includes: Read the health record records corresponding to each subject to be assessed from the original health record set of prediabetes diagnosis and care, and form a record set with expanded field types; Expand the record set based on field type and perform unit unification processing on data fields from different sources to obtain a set of health record fields with unified units; Based on the unified set of health record fields of the unit, outlier identification and outlier truncation are performed on numerical fields, while retaining the source of outlier, outlier type and truncation mark, to form a set of health record fields after outlier processing. Based on the set of health record fields after outlier processing, missing fields are marked with missing values ​​to obtain a set of health record fields with missing value labels. Based on the set of health record fields with missing value identifiers, time window alignment is performed using the assessment time of the subject to be assessed as the benchmark to form a time window aligned time series health record field set. Based on the time-window aligned time-series health record field set, a data credibility weight is generated for each type of field, forming a time-series health record field set with data credibility weights; Based on the set of time-series health record fields with data credibility weights, a time decay weight is generated for each type of field to form a set of time-series health record fields with time decay weights. Based on the set of time-series health record fields with time decay weights, the data credibility weight, time decay weight, missing value identifier, outlier handling identifier, unit verification identifier, and source difference identifier are combined into a credibility mask. The credibility mask is then encapsulated synchronously with the corresponding health record fields to obtain a set of time-series health record features with credibility masks.

[0009] Optionally, the feature set based on the time-series health record with a credibility mask is grouped according to the glucose metabolism risk layer, the TCM syndrome manifestation layer, the lifestyle behavior layer, and the care compliance layer, including: Read the feature set of time-series health records with a credibility mask, parse each health record field, and form a groupable health record field set; Based on the groupable health record field set, the fields related to glucose metabolism screening and metabolic status are classified into the glucose metabolism risk layer, resulting in the glucose metabolism risk layer feature set. Based on the feature set of the glucose metabolism risk layer and the groupable health record field set, the fields related to the TCM syndrome differentiation are classified into the TCM syndrome manifestation layer, thus obtaining the feature set of the TCM syndrome manifestation layer. Based on the TCM syndrome manifestation feature set and the groupable health record field set, the fields related to daily health behaviors are classified into the lifestyle behavior layer, thus obtaining the lifestyle behavior layer feature set. Based on the lifestyle behavior layer feature set and the groupable health record field set, the fields related to nursing execution and follow-up cooperation are classified into the nursing compliance layer, thus obtaining the nursing compliance layer feature set; Based on the feature sets of the glucose metabolism risk layer, the feature sets of the lifestyle behavior layer, and the feature sets of the care compliance layer, the continuous indicators are converted into normalized numerical embeddings to form a continuous indicator embedding set. Based on the feature sets of the glucose metabolism risk layer, the feature sets of the lifestyle behavior layer, and the feature sets of the care compliance layer, the categorical indicators are converted into category embeddings to form a categorical indicator embedding set. Based on the feature set of TCM syndrome manifestation layer, the text information of TCM four diagnostic methods is converted into symptom word embeddings to form a TCM four diagnostic method symptom word embedding set. Based on continuous indicator embedding sets, categorical indicator embedding sets, and TCM four diagnostic symptoms word embedding sets, hierarchical position codes are added to the glucose metabolism risk layer, TCM syndrome manifestation layer, lifestyle behavior layer, and nursing compliance layer, respectively. Time position codes are added to the features arranged according to time windows in each layer to form a hierarchical health record feature sequence.

[0010] Optionally, the construction of the Monarch Mixer sparse hybrid coding network based on the hierarchical health record feature sequences includes: Read the feature sequence of the hierarchical health record, and establish hierarchical input sequences according to the glucose metabolism risk layer, TCM syndrome manifestation layer, lifestyle behavior layer and nursing compliance layer respectively. In each level input sequence, retain the feature order, field semantics, time position encoding, hierarchical position encoding, data credibility weight, time decay weight and credibility mask corresponding to that level to form a set of hierarchical input feature blocks. Based on the hierarchical input feature block set, the input sequence of each level is divided into blocks to form a set of feature blocks to be mixed within the layer; Based on the set of feature blocks to be mixed within the layer, the first diagonal mapping is performed within each layer to form the first layer local mapping feature set; Based on the local mapping feature set within the first layer, cross-block permutation mapping is performed to form a cross-block permutation feature set within the layer. Based on the intra-layer cross-block permutation feature set, perform the second block diagonal mapping to form the second intra-layer local mapping feature set; Based on the local mapping feature set within the second layer, gated residual normalization mapping is performed. The features after mixing within the layer are fused with the corresponding features before entering the Monarch mixing unit within the layer in a preservative manner. The participation degree of different features is controlled according to the confidence mask, so that the local combination relationship within the same layer can be expressed without losing the original field semantics, thus forming the Monarch mixing feature set within the layer. Based on the intra-layer Monarch mixed feature set, an inter-layer feature block set to be mixed is constructed according to the hierarchical correspondence between the glucose metabolism risk layer, the TCM syndrome manifestation layer, the lifestyle behavior layer, and the nursing compliance layer, thus forming an inter-layer feature block set to be mixed. Based on the set of inter-layer feature blocks to be mixed, cross-layer block diagonal mapping, hierarchical permutation mapping and confidence-weighted residual mapping are performed sequentially to form an inter-layer Monarch mixed feature set. Based on the inter-layer Monarch hybrid feature set, the local combination relationship within the layer and the global association relationship between the layers are uniformly encapsulated, and the hierarchical identifier, time position code, confidence mask and source backtracking identifier corresponding to each feature are retained to obtain the hierarchical sparse hybrid feature representation.

[0011] Optionally, the hierarchical feature decoupling process based on hierarchical sparse hybrid feature representation includes: Read the hierarchical sparse hybrid feature representation and perform hierarchical feature decoupling to form a set of hierarchical hybrid features to be decoupled; Based on the set of hierarchical hybrid features to be decoupled, the features corresponding to blood glucose status, anthropometric status, metabolic risk status and related historical health management status are projected into the latent space of glucose metabolism risk to form a set of latent variables of glucose metabolism risk. Based on the set of latent variables for glucose metabolism risk and the set of hierarchical mixed features to be decoupled, features related to inspection, auscultation, inquiry, palpation, symptom lexical, tongue appearance, pulse appearance, emotional expression and TCM syndrome tendency are projected into the TCM syndrome latent space to form a set of TCM syndrome latent variables. Based on the latent variable set of TCM syndromes and the hierarchical mixed feature set to be decoupled, features related to dietary structure, exercise habits, sedentary state, sleep state, emotional regulation, self-management behavior and lifestyle improvement space are projected into the latent space of lifestyle intervention to form the latent variable set of lifestyle intervention. Based on the latent variable set of lifestyle intervention and the hierarchical mixed feature set to be decoupled, features related to follow-up visits, implementation of nursing recommendations, implementation of dietary care, implementation of exercise care, implementation of sleep and mood care, participation in health education and the completeness of self-monitoring records are projected into the nursing compliance latent space to form the nursing compliance latent variable set. Based on the latent variable set of glucose metabolism risk, the latent variable set of traditional Chinese medicine syndrome, the latent variable set of lifestyle intervention, and the latent variable set of nursing compliance, orthogonal constraint processing is performed to form the latent variable set after orthogonal constraint. Based on the set of latent variables after orthogonal constraints, perform reconstruction consistency constraint processing to form a set of latent variables after reconstruction consistency constraints. Based on the set of latent variables after reconstructing the consistency constraints, a credibility mask constraint is performed to form a set of latent variables after credibility mask constraints. Based on the set of latent variables constrained by credibility mask, the latent variable sets of glucose metabolism risk, traditional Chinese medicine syndrome, lifestyle intervention, and nursing compliance are encapsulated according to object identifier, assessment time, and hierarchical identifier to obtain a hierarchical decoupled set of health record latent variables.

[0012] Optionally, the construction of the syndrome-care association assessment matrix based on the hierarchical decoupling of the set of latent variables in health records includes: Read the latent variable set of hierarchical decoupled health records, extract the latent variable sets of glucose metabolism risk, TCM syndrome, lifestyle intervention, and nursing compliance, and maintain the correspondence between the four types of latent variables according to object identifier, assessment time, hierarchical identifier, and confidence mask to form the input set of latent variables to be assessed. Based on the set of latent variable inputs to be evaluated, the syndrome-care association assessment matrix is ​​constructed to form a set of syndrome-side association structures. Based on the syndrome-related structure set, the latent space of TCM syndromes and the latent space of glucose metabolism risk are interactively mapped to obtain the syndrome interaction mapping feature set. Based on the syndrome interaction mapping feature set, syndrome tendency scores are formed for Qi and Yin deficiency, spleen deficiency with phlegm and dampness, liver stagnation with Qi stagnation, Qi stagnation with phlegm obstruction and damp-heat accumulation, forming a syndrome tendency score set. Based on the set of latent variable inputs to be evaluated, the nursing care side structure of the syndrome-nursing care correlation evaluation matrix is ​​constructed, forming a set of nursing care side correlation structures; Based on the set of care-side association structures, the implicit space of lifestyle intervention and the implicit space of care compliance are interactively mapped to obtain the set of care-side interaction mapping features. Based on the nursing interaction mapping feature set, nursing priority scores are generated for dietary nursing, exercise nursing and sleep and emotional nursing respectively, forming a nursing priority score set; Based on the syndrome tendency score set and the nursing care priority score set, the scores are matrix-encapsulated according to the object identifier, assessment time, syndrome category, nursing care category and credibility mask to form a dialectical nursing care correlation score set.

[0013] Optionally, the calibration and fusion of each syndrome tendency score and each care priority score based on the dialectical care association score set, combined with data reliability weights and time decay weights, includes: Read the set of correlation scores for syndrome differentiation and care, and expand the scores for each syndrome tendency and care priority according to object identifier, assessment time, syndrome category, care category, score source, field source, data credibility weight, time decay weight and credibility mask to form a set of correlation scores to be calibrated; Based on the set of correlation scores to be calibrated, the credibility of each syndrome tendency score and each nursing priority score is calibrated according to the data credibility weight, forming a set of correlation scores after credibility calibration. Based on the credibility-calibrated set of related scores, the scores of each syndrome tendency and each nursing priority are time-calibrated according to the time decay weight, and the time calibration identifier of each score is retained to form a time-calibrated set of related scores. Based on the time-calibrated associated score set, the syndrome tendency scores corresponding to Qi and Yin deficiency, spleen deficiency with phlegm and dampness, liver stagnation with Qi stagnation, Qi stagnation with phlegm obstruction, and damp-heat accumulation are fused to form a calibrated and fused syndrome assessment set. Based on the calibrated and fused syndrome assessment set and the time-calibrated correlation score set, the care priority scores corresponding to dietary care, exercise care and sleep and mood care are fused to form a calibrated and fused care assessment set. Based on the calibrated and fused syndrome assessment set and the calibrated and fused nursing care assessment set, a comprehensive health record assessment value is generated, and a comprehensive health record assessment value record is formed. Based on the comprehensive assessment value records of health records and the syndrome assessment set after calibration and fusion, the syndromes of Qi and Yin deficiency, spleen deficiency and phlegm dampness, liver stagnation and Qi stagnation, Qi stagnation and phlegm obstruction and damp-heat accumulation are ranked according to the degree of syndrome tendency after calibration and fusion. The syndrome primary and secondary ranking is obtained, and the ranking basis, credibility calibration mark and time calibration mark are retained to form the syndrome primary and secondary ranking record. Based on the primary and secondary syndrome ranking records and the calibrated and integrated nursing assessment set, dietary care, exercise care, and sleep and emotional care are ranked according to the calibrated and integrated nursing priority to obtain the nursing intervention priority ranking and form a nursing intervention priority ranking record. Based on the comprehensive assessment values ​​of health records, the ranking of syndromes, and the ranking of nursing intervention priorities, the results of the health record assessment for prediabetes syndrome differentiation and nursing care are obtained by structuring and encapsulating the data according to the subject identification and assessment time.

[0014] A health record management and assessment system for prediabetes based on syndrome differentiation and care, comprising: The data collection module collects and unifies the fields of the multi-source health record data of the subjects to be evaluated, and merges them to form a set of original health records for the diagnosis and treatment of prediabetes. The preprocessing module preprocesses the original health records set for prediabetes diagnosis and treatment to obtain a set of time-series health record features with a credibility mask. The feature grouping module, based on the time-series health record feature set with credibility mask, groups the features according to the glucose metabolism risk layer, the TCM syndrome manifestation layer, the lifestyle behavior layer, and the care compliance layer, forming a hierarchical health record feature sequence; The Monarch Mixer sparse hybrid coding module constructs a MonarchMixer sparse hybrid coding network based on the hierarchical health record feature sequences to obtain hierarchical sparse hybrid feature representations. The decoupling module performs hierarchical feature decoupling processing based on hierarchical sparse hybrid feature representation to obtain a hierarchical decoupled set of latent variables in health records. The syndrome differentiation module, based on the hierarchical decoupling of the latent variable set of health records, constructs a syndrome-nursing care correlation assessment matrix, forming a syndrome differentiation and nursing care correlation score set; The output module, based on the syndrome differentiation and nursing care correlation score set combined with data credibility weight and time decay weight, calibrates and integrates the scores of each syndrome tendency and each nursing care priority to generate a comprehensive health record assessment value, syndrome priority ranking, and nursing intervention priority ranking, thus obtaining the health record assessment results of syndrome differentiation and nursing care for prediabetes.

[0015] The beneficial effects of this invention are: This invention transforms multi-source health record data into a time-series health record feature set with a credibility mask by unifying fields, units, truncating outliers, identifying missing values, and aligning time windows. This ensures that data from different sources, times, and qualities have a consistent data structure before entering the model, enabling the differentiation of real and valid data, missing data, outlier truncated data, and low-credibility data. Furthermore, it reduces the interference of noisy and expired records on the evaluation results by using data credibility weights and time decay weights, thereby improving the stability and traceability of feature modeling.

[0016] This invention constructs a hierarchical health record feature sequence comprising a glucose metabolism risk layer, a traditional Chinese medicine syndrome manifestation layer, a lifestyle behavior layer, and a nursing compliance layer. It employs sparse hybrid encoding using intra-layer Monarch hybrid units and inter-layer Monarch hybrid units. Intra-layer Monarch hybrid units extract local combination relationships within the same level through a first-block diagonal mapping, cross-block permutation mapping, a second-block diagonal mapping, and gated residual normalization mapping. Inter-layer Monarch hybrid units extract global association relationships between different levels through cross-layer block diagonal mapping, hierarchical permutation mapping, and confidence-weighted residual mapping. This reduces redundant interactions while preserving key cross-layer relationships, making the health record feature expression more sparse, stable, and hierarchically semantic.

[0017] This invention projects hierarchical sparse hybrid feature representations onto the latent spaces of glucose metabolism risk, traditional Chinese medicine syndrome, lifestyle intervention, and nursing compliance, respectively. By suppressing redundant coupling between different latent spaces through orthogonal constraints, reconstruction consistency constraints, and credibility mask constraints, it can preserve the independent expression of metabolic risk, syndrome manifestation, lifestyle intervention needs, and nursing performance status, while maintaining their correspondence with the original health record fields. This provides a clearer and more stable feature basis for syndrome tendency scoring, nursing priority scoring, and comprehensive assessment. Attached Figure Description

[0018] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a flowchart of a health record management system and assessment method for prediabetes based on syndrome differentiation and care, proposed in this invention. Detailed Implementation

[0019] Example 1: Reference Figure 1 A method for managing and assessing health records for prediabetes based on syndrome differentiation and care, including: Multi-source health record data of the subjects to be evaluated were collected and fields were unified. The multi-source health record data were merged according to the subject identification, collection time and data source to form a set of original health records for the diagnosis and treatment of prediabetes. Based on the original health record set of prediabetes syndrome differentiation and care, the data fields from different sources were processed by unifying units, truncating outliers, identifying missing values, and aligning time windows. Data credibility weight and time decay weight were generated for each type of field to obtain a set of time-series health record features with credibility mask. Based on the feature set of time-series health records with credibility mask, the features are grouped according to the glucose metabolism risk layer, TCM syndrome manifestation layer, lifestyle behavior layer and nursing compliance layer. Continuous indicators are converted into normalized numerical embeddings, categorical indicators are converted into category embeddings, and TCM four diagnostic text information is converted into symptom word embeddings. Each layer of features is then attached with hierarchical position coding and temporal position coding to form a hierarchical health record feature sequence. Based on the feature sequences of hierarchical health records, a Monarch Mixer sparse hybrid coding network is constructed. The Monarch Mixer sparse hybrid coding network includes intra-layer Monarch mixing units and inter-layer Monarch mixing units. The intra-layer Monarch mixing unit is composed of a first block diagonal mapping, a cross-block permutation mapping, a second block diagonal mapping, and a gated residual normalization mapping connected in sequence. It is used to extract the local combination relationship between features within the same level. The inter-layer Monarch mixing unit is composed of a cross-layer block diagonal mapping, a hierarchical permutation mapping, and a confidence-weighted residual mapping connected in sequence. It is used to extract the global association relationship between the glucose metabolism risk layer, the TCM syndrome manifestation layer, the lifestyle behavior layer, and the nursing compliance layer, thus obtaining a hierarchical sparse hybrid feature representation. Based on the hierarchical sparse hybrid feature representation, hierarchical feature decoupling is performed. The hierarchical sparse hybrid feature representation is projected onto the latent space of glucose metabolism risk, the latent space of traditional Chinese medicine syndrome, the latent space of lifestyle intervention, and the latent space of nursing compliance, respectively. The redundant coupling between different latent spaces is suppressed by orthogonal constraints, reconstruction consistency constraints, and credibility mask constraints, so as to obtain the hierarchical decoupled health record latent variable set. Based on the hierarchical decoupled set of latent variables in health records, a syndrome-nursing care correlation assessment matrix is ​​constructed. The latent space of TCM syndromes is interactively mapped with the latent space of glucose metabolism risk to obtain syndrome tendency scores corresponding to Qi and Yin deficiency, spleen deficiency with phlegm and dampness, liver stagnation with Qi stagnation, Qi stagnation with phlegm obstruction, and damp-heat accumulation. The latent space of lifestyle interventions is interactively mapped with the latent space of nursing compliance to obtain nursing care priority scores corresponding to dietary care, exercise care, and sleep and emotional care, forming a set of syndrome differentiation and nursing care correlation scores. Based on the syndrome differentiation and nursing care correlation score set combined with data credibility weight and time decay weight, the scores of each syndrome tendency and each nursing care priority are calibrated and integrated to generate a comprehensive health record assessment value, syndrome priority ranking and nursing intervention priority ranking, and obtain the health record assessment results of syndrome differentiation and nursing care for prediabetes.

[0020] In this embodiment, multi-source health record data of the subjects to be evaluated are collected and fields are standardized. The multi-source health record data are merged according to the subject identifier, collection time, and data source to form a set of original health records for prediabetes diagnosis and care, including: A master index for health record collection is established for the subjects to be evaluated, resulting in a set of master index records for the subjects; In Example 1, the health record collection master index is used to record the object identifier, record number, collection batch, collection agency, collection personnel, collection terminal, collection time, and record version, so that the same object to be evaluated can form a traceable basic collection record in different collection scenarios.

[0021] Based on the object's master index record set, multi-source health record data of the object to be evaluated is collected. During the collection, the data source, collection method, original field name, original field value, collection time, and record responsibility entity corresponding to each data item are retained to form a multi-source original collection record set. Based on a set of original data collected from multiple sources, a unified dictionary of fields is established to form a unified mapping set of fields. In Example 1, the unified field dictionary maps fields with the same meaning but different names from different sources to a unified field name. It uniformly labels the field type, field length, value range, value enumeration and text description of the same field in different systems, and retains the correspondence between the original field name and the unified field name.

[0022] Based on the unified field mapping set, the multi-source original collection record set is classified by field. Fields that do not belong to the same semantic category are stored separately, and fields that belong to the same semantic category but have different sources are merged into the same category. Each merged field is attached with a field category identifier and a source category identifier to form a health record record set classified by field semantics. Based on the health record record set categorized by field semantics, records of the same object to be evaluated are aggregated according to object identifier. Records of the same object formed at different collection times are established in chronological order according to the collection time. Records of the same object from different data sources but collected at the same or similar time are archived in parallel according to the data source. Record status identifiers are set for duplicate records, supplementary records, revised records and conflicting records respectively, forming a time-series merged record set at the object dimension. Based on the object-dimensional time-series merged record set, each health record is checked for completeness and traceability. Records that pass the checks are retained as valid merged records, while those that fail the checks are retained as merged records pending verification, forming a merged health record set with record status identifiers, thus forming the original health record set for prediabetes diagnosis and care.

[0023] In Example 1, based on the health record merging set with record status identifiers, the valid merged records and the merged records to be verified are structurally encapsulated according to the object identifier, collection time and data source. This allows the basic information, physical measurement information, glucose metabolism screening indicators, diet and exercise information, sleep and mood information, TCM four diagnostic information, previous health management information and nursing follow-up records of the same subject to be assessed to be stored in the same file structure, forming the original health record set for prediabetes diagnosis and care.

[0024] In this embodiment, based on the original health record set for prediabetes diagnosis and treatment, data fields from different sources are processed by unifying units, truncating outliers, identifying missing values, and aligning time windows. Data credibility weights and time decay weights are generated for each type of field, resulting in a time-series health record feature set with a credibility mask, including: Read the health record records corresponding to each subject to be assessed from the original health record set of prediabetes diagnosis and care, and form a record set with expanded field types; In Example 1, health record records are expanded according to unified field name, field category identifier, source category identifier, collection time and record status identifier. Numeric fields, categorical fields, text fields and time fields are identified respectively, and the original value, original unit, unified field name and data source of each field are retained.

[0025] Expand the record set based on field type and perform unit unification processing on data fields from different sources to obtain a set of health record fields with unified units; In Example 1, the glucose metabolism screening indicators, anthropometric indicators, blood pressure related indicators, weight related indicators, waist circumference related indicators, exercise duration related indicators, sleep duration related indicators, and follow-up interval related indicators are converted into a preset unified unit expression form. For fields where the source of the unit cannot be confirmed, the original unit is retained and a unit verification mark is set.

[0026] Based on the unified set of health record fields of the unit, outlier identification and outlier truncation are performed on numerical fields, while retaining the source of outlier, outlier type and truncation mark, to form a set of health record fields after outlier processing. In Example 1, outlier identification is based on the physiologically reasonable range of the field, the allowable range of the acquisition device, the range of nursing follow-up records, and the historical record range of the same subject to be evaluated. Outlier truncation is used to adjust field values ​​that are obviously outside the reasonable range and are not suitable for direct participation in subsequent evaluation to the preset acceptable boundary.

[0027] Based on the set of health record fields after outlier processing, missing fields are marked with missing values ​​to obtain a set of health record fields with missing value labels. Based on the set of health record fields with missing value identifiers, time window alignment is performed using the assessment time of the subject to be assessed as the benchmark to form a time window aligned time series health record field set. Based on the time-window aligned time-series health record field set, a data credibility weight is generated for each type of field, forming a time-series health record field set with data credibility weights; Based on the set of time-series health record fields with data credibility weights, a time decay weight is generated for each type of field to form a set of time-series health record fields with time decay weights. Based on the set of time-series health record fields with time decay weights, the data credibility weight, time decay weight, missing value identifier, outlier handling identifier, unit verification identifier, and source difference identifier are combined into a credibility mask. The credibility mask is then encapsulated synchronously with the corresponding health record fields to obtain a set of time-series health record features with credibility masks.

[0028] Based on a time-series health record feature set with a credibility mask, features are grouped according to four layers: glucose metabolism risk, traditional Chinese medicine syndrome manifestation, lifestyle behavior, and care compliance. Continuous indicators are converted into normalized numerical embeddings, categorical indicators are converted into class embeddings, and textual information from the four diagnostic methods of traditional Chinese medicine is converted into symptom lexical embeddings. Each layer's features are further augmented with hierarchical positional encoding and temporal positional encoding to form a hierarchical health record feature sequence, including: Read the feature set of time-series health records with a credibility mask, parse each health record field, and form a groupable health record field set; In Example 1, each health record field is parsed according to the unified field name, field category identifier, collection time, data source, data credibility weight, time decay weight, and missing value identifier, thus distinguishing between continuous indicators that can be directly used for numerical representation, categorical indicators used for category representation, and TCM four diagnostic text information that needs to be represented by text lexicalization.

[0029] Based on the groupable health record field set, the fields related to glucose metabolism screening and metabolic status are classified into the glucose metabolism risk layer, resulting in the glucose metabolism risk layer feature set. In Example 1, the glucose metabolism risk layer includes fields related to fasting blood glucose, postprandial blood glucose, glycated hemoglobin, body mass index, waist circumference, blood pressure, and previous health management related to glucose metabolism risk assessment, and retains the confidence mask and time window information corresponding to each field.

[0030] Based on the feature set of the glucose metabolism risk layer and the groupable health record field set, the fields related to the TCM syndrome differentiation are classified into the TCM syndrome manifestation layer, thus obtaining the feature set of the TCM syndrome manifestation layer. In Example 1, the TCM syndrome manifestation layer includes information from inspection, auscultation and olfaction, inquiry, palpation, descriptions related to thirst and dry mouth, descriptions related to fatigue and weakness, descriptions related to chest and rib discomfort, descriptions related to phlegm and dampness, descriptions related to dietary preferences, descriptions related to sleep and emotions, descriptions of tongue appearance, and descriptions of pulse appearance, while retaining the original text descriptions, standardized descriptions, and credibility masks.

[0031] Based on the TCM syndrome manifestation feature set and the groupable health record field set, the fields related to daily health behaviors are classified into the lifestyle behavior layer, thus obtaining the lifestyle behavior layer feature set. In Example 1, the lifestyle behavior layer includes dietary structure, staple food intake habits, sugar intake, exercise type, exercise frequency, exercise duration, sedentary status, sleep duration, sleep quality, emotional stress state, and self-management behavior records, and retains their time window information and source category identifiers.

[0032] Based on the lifestyle behavior layer feature set and the groupable health record field set, the fields related to nursing execution and follow-up cooperation are classified into the nursing compliance layer, thus obtaining the nursing compliance layer feature set; Based on the feature sets of the glucose metabolism risk layer, the feature sets of the lifestyle behavior layer, and the feature sets of the care compliance layer, the continuous indicators are converted into normalized numerical embeddings to form a continuous indicator embedding set. Based on the feature sets of the glucose metabolism risk layer, the feature sets of the lifestyle behavior layer, and the feature sets of the care compliance layer, the categorical indicators are converted into category embeddings to form a categorical indicator embedding set. Based on the feature set of TCM syndrome manifestation layer, the text information of TCM four diagnostic methods is converted into symptom word embeddings to form a TCM four diagnostic method symptom word embedding set. Based on continuous indicator embedding sets, categorical indicator embedding sets, and TCM four diagnostic symptoms word embedding sets, hierarchical position codes are added to the glucose metabolism risk layer, TCM syndrome manifestation layer, lifestyle behavior layer, and nursing compliance layer, respectively. Time position codes are added to the features arranged according to time windows in each layer to form a hierarchical health record feature sequence.

[0033] In this embodiment, a Monarch Mixer sparse hybrid coding network is constructed based on the feature sequences of hierarchical health records, including: Read the feature sequence of the hierarchical health record, and establish hierarchical input sequences according to the glucose metabolism risk layer, TCM syndrome manifestation layer, lifestyle behavior layer and nursing compliance layer respectively. In each level input sequence, retain the feature order, field semantics, time position encoding, hierarchical position encoding, data credibility weight, time decay weight and credibility mask corresponding to that level to form a set of hierarchical input feature blocks. Based on the hierarchical input feature block set, the input sequence of each level is divided into blocks to form a set of feature blocks to be mixed within the layer; In Example 1, features with similar semantics or similar time windows within the same level are grouped into adjacent feature blocks. Feature blocks that do not have reliable values ​​or are in a state of pending verification are marked with a mask, and the traceable correspondence between feature blocks and original health record fields is maintained.

[0034] Based on the set of feature blocks to be mixed within the layer, the first diagonal mapping is performed within each layer to form the first layer local mapping feature set; In Example 1, the first diagonal mapping is used to perform local mapping on features with similar semantics, collection time, or source category within the same feature block, so that the combination of indicators within the sugar metabolism risk layer, the combination of symptoms within the TCM syndrome manifestation layer, the combination of behaviors within the lifestyle behavior layer, and the combination of execution status within the nursing compliance layer can obtain initial local representations respectively.

[0035] Based on the local mapping feature set within the first layer, cross-block permutation mapping is performed to form a cross-block permutation feature set within the layer. In Example 1, cross-block permutation mapping is used to change the arrangement correspondence between different feature blocks within the same level, so that fields originally located in different feature blocks can generate controlled information interaction in subsequent mappings, while keeping the hierarchical boundaries between the glucose metabolism risk layer, the TCM syndrome manifestation layer, the lifestyle behavior layer, and the care compliance layer intact.

[0036] Based on the intra-layer cross-block permutation feature set, perform the second block diagonal mapping to form the second intra-layer local mapping feature set; In Example 1, the second diagonal mapping is used to perform local mixing again within the feature block after cross-block permutation, so that features that were originally far apart in the same level can establish a local combination relationship through the permuted intra-block mapping, and restrict the participation of features with low confidence masks or missing value identifiers.

[0037] Based on the local mapping feature set within the second layer, gated residual normalization mapping is performed. The features after mixing within the layer are fused with the corresponding features before entering the Monarch mixing unit within the layer in a preservative manner. The participation degree of different features is controlled according to the confidence mask, so that the local combination relationship within the same layer can be expressed without losing the original field semantics, thus forming the Monarch mixing feature set within the layer. Based on the intra-layer Monarch mixed feature set, an inter-layer feature block set to be mixed is constructed according to the hierarchical correspondence between the glucose metabolism risk layer, the TCM syndrome manifestation layer, the lifestyle behavior layer, and the nursing compliance layer, thus forming an inter-layer feature block set to be mixed. Based on the set of inter-layer feature blocks to be mixed, cross-layer block diagonal mapping, hierarchical permutation mapping and confidence-weighted residual mapping are performed sequentially to form an inter-layer Monarch mixed feature set. In Example 1, cross-layer block diagonal mapping is used to establish restricted block associations between different layers, hierarchical permutation mapping is used to change the interaction order between the glucose metabolism risk layer, TCM syndrome manifestation layer, lifestyle behavior layer and care compliance layer, and credibility weighted residual mapping is used to retain credible hierarchical information and weaken low-credible hierarchical information according to data credibility weight and time decay weight.

[0038] Based on the inter-layer Monarch hybrid feature set, the local combination relationship within the layer and the global association relationship between the layers are uniformly encapsulated, and the hierarchical identifier, time position code, confidence mask and source backtracking identifier corresponding to each feature are retained to obtain the hierarchical sparse hybrid feature representation.

[0039] In this embodiment, hierarchical feature decoupling processing based on hierarchical sparse hybrid feature representation includes: Read the hierarchical sparse hybrid feature representation and perform hierarchical feature decoupling to form a set of hierarchical hybrid features to be decoupled; In Example 1, the mixed features are organized according to the feature level, feature semantic category, time position encoding, credibility mask and source backtracking identifier. Features related to the sugar metabolism risk layer, features related to the TCM syndrome manifestation layer, features related to the lifestyle behavior layer and features related to the nursing compliance layer are marked as inputs to be decoupled, while retaining the cross-layer association information formed in the inter-layer Monarch hybrid unit.

[0040] Based on the set of hierarchical hybrid features to be decoupled, the features corresponding to blood glucose status, anthropometric status, metabolic risk status and related historical health management status are projected into the latent space of glucose metabolism risk to form a set of latent variables of glucose metabolism risk. In Example 1, the latent space of glucose metabolism risk is used to centrally express the semantics of metabolic risk related to the assessment of prediabetes health records, and to preserve the correspondence between it and the original glucose metabolism risk layer fields, data confidence weights and time decay weights.

[0041] Based on the set of latent variables for glucose metabolism risk and the set of hierarchical mixed features to be decoupled, features related to inspection, auscultation, inquiry, palpation, symptom lexical, tongue appearance, pulse appearance, emotional expression and TCM syndrome tendency are projected into the TCM syndrome latent space to form a set of TCM syndrome latent variables. In Example 1, the TCM syndrome latent space is used to centrally express the semantics of TCM syndrome manifestations and retain the correspondence between it and the TCM syndrome manifestation layer fields, symptom lexical embeddings and credibility masks.

[0042] Based on the latent variable set of TCM syndromes and the hierarchical mixed feature set to be decoupled, features related to dietary structure, exercise habits, sedentary state, sleep state, emotional regulation, self-management behavior and lifestyle improvement space are projected into the latent space of lifestyle intervention to form the latent variable set of lifestyle intervention. In Example 1, the lifestyle intervention latent space is used to centrally express the behavioral semantics of the subject to be assessed in terms of lifestyle that can be utilized by nursing interventions, and to preserve the correspondence between it and the lifestyle behavior layer fields and time window information.

[0043] Based on the latent variable set of lifestyle intervention and the hierarchical mixed feature set to be decoupled, features related to follow-up visits, implementation of nursing recommendations, implementation of dietary care, implementation of exercise care, implementation of sleep and mood care, participation in health education and the completeness of self-monitoring records are projected into the nursing compliance latent space to form the nursing compliance latent variable set. Based on the latent variable set of glucose metabolism risk, the latent variable set of traditional Chinese medicine syndrome, the latent variable set of lifestyle intervention, and the latent variable set of nursing compliance, orthogonal constraint processing is performed to form the latent variable set after orthogonal constraint. In Example 1, orthogonal constraint processing is used to enable different latent spaces to focus on different health record semantics, reduce the repetitive expression between the latent space of glucose metabolism risk and the latent space of traditional Chinese medicine syndrome, between the latent space of lifestyle intervention and the latent space of nursing compliance, and between any two latent spaces, and suppress redundant coupling while preserving necessary cross-layer associations.

[0044] Based on the set of latent variables after orthogonal constraints, perform reconstruction consistency constraint processing to form a set of latent variables after reconstruction consistency constraints. In Example 1, the reconstruction consistency constraint processing is used to enable the latent variables in each latent space to be traced back to the main field semantics, hierarchical position, temporal position and source identifier in the hierarchical sparse hybrid feature representation, so as to prevent the loss of the necessary correspondence between glucose metabolism risk features, TCM syndrome manifestation features, lifestyle behavior features and nursing compliance features during the decoupling process.

[0045] Based on the set of latent variables after reconstructing the consistency constraints, a credibility mask constraint is performed to form a set of latent variables after credibility mask constraints. In Example 1, the credibility mask constraint processing is used to reduce the impact of missing fields, abnormal truncated fields, unit unverified fields, source conflict fields and low credibility fields on the latent space expression during the latent variable generation process, and to maintain the expression stability of high credibility fields and recently valid fields in the latent space.

[0046] Based on the set of latent variables constrained by credibility mask, the latent variable sets of glucose metabolism risk, traditional Chinese medicine syndrome, lifestyle intervention, and nursing compliance are encapsulated according to object identifier, assessment time, and hierarchical identifier to obtain a hierarchical decoupled set of health record latent variables.

[0047] In this embodiment, a syndrome-care association assessment matrix is ​​constructed based on the hierarchical decoupling of the latent variable set in health records, including: Read the latent variable set of hierarchical decoupled health records, extract the latent variable sets of glucose metabolism risk, TCM syndrome, lifestyle intervention, and nursing compliance, and maintain the correspondence between the four types of latent variables according to object identifier, assessment time, hierarchical identifier, and confidence mask to form the input set of latent variables to be assessed. Based on the set of latent variable inputs to be evaluated, the syndrome-care association assessment matrix is ​​constructed to form a set of syndrome-side association structures. In Example 1, the syndrome-side structure includes four syndrome tendency positions: Qi and Yin deficiency, spleen deficiency with phlegm and dampness, liver stagnation and Qi stagnation, Qi stagnation and phlegm obstruction, and damp-heat accumulation. Each syndrome tendency position is used to receive symptom manifestation semantics from the latent space of TCM syndromes and metabolic risk semantics from the latent space of glucose metabolism risk, and retains the corresponding symptom term source, glucose metabolism field source, and confidence mask.

[0048] Based on the syndrome-related structure set, the latent space of TCM syndromes and the latent space of glucose metabolism risk are interactively mapped to obtain the syndrome interaction mapping feature set. In Example 1, the interactive mapping caused latent variables related to fatigue, dry mouth and thirst, shortness of breath and reluctance to speak, tongue and pulse manifestations, and glucose metabolism risk to jointly act on the syndrome tendency position corresponding to Qi and Yin deficiency; caused latent variables related to body shape, phlegm and dampness manifestations, abdominal heaviness, dietary preferences, tongue coating manifestations, and glucose metabolism risk to jointly act on the syndrome tendency position corresponding to spleen deficiency and phlegm and dampness; caused latent variables related to emotional state, chest and rib discomfort, sleep state, stress state, and glucose metabolism risk to jointly act on the syndrome tendency position corresponding to liver stagnation and Qi stagnation; and caused latent variables related to thirst, sticky stools, damp-heat manifestations, records of spicy, sweet, and rich foods, and glucose metabolism risk to jointly act on the syndrome tendency position corresponding to internal damp-heat.

[0049] Based on the syndrome interaction mapping feature set, syndrome tendency scores are formed for Qi and Yin deficiency, spleen deficiency with phlegm and dampness, liver stagnation with Qi stagnation, Qi stagnation with phlegm obstruction and damp-heat accumulation, forming a syndrome tendency score set. Based on the set of latent variable inputs to be evaluated, the nursing care side structure of the syndrome-nursing care correlation evaluation matrix is ​​constructed, forming a set of nursing care side correlation structures; In Example 1, the care-providing structure includes three care-providing priority positions: dietary care, exercise care, and sleep and mood care. Each care-providing priority position is used to receive behavioral improvement semantics from the lifestyle intervention latent space and execution cooperation semantics from the nursing compliance latent space, and retains the corresponding lifestyle field source, nursing follow-up field source, and credibility mask.

[0050] Based on the set of care-side association structures, the implicit space of lifestyle intervention and the implicit space of care compliance are interactively mapped to obtain the set of care-side interaction mapping features. In Example 1, the interactive mapping causes latent variables related to dietary structure, staple food control, sugar intake, completeness of dietary records, and implementation of dietary care to jointly affect the care priority position corresponding to dietary care; causes latent variables related to exercise frequency, exercise duration, exercise type, sedentary status, and implementation of exercise care to jointly affect the care priority position corresponding to exercise care; and causes latent variables related to sleep duration, sleep quality, emotional stress, mood regulation records, and implementation of sleep-emotional care to jointly affect the care priority position corresponding to sleep-emotional care.

[0051] Based on the nursing interaction mapping feature set, nursing priority scores are generated for dietary nursing, exercise nursing and sleep and emotional nursing respectively, forming a nursing priority score set; In Example 1, the nursing priority score is used to indicate the priority of the corresponding nursing intervention direction in the current state reflected by the health record data, and each nursing priority score is associated with and stored with its corresponding lifestyle behavior characteristics, nursing compliance characteristics, data credibility weight and time decay weight.

[0052] Based on the syndrome tendency score set and the nursing care priority score set, the scores are matrix-encapsulated according to the object identifier, assessment time, syndrome category, nursing care category and credibility mask to form a dialectical nursing care correlation score set.

[0053] In this embodiment, based on the syndrome differentiation and nursing care correlation score set combined with data reliability weight and time decay weight, the scores of each syndrome tendency and each nursing care priority are calibrated and fused to generate a comprehensive health record assessment value, syndrome priority ranking, and nursing intervention priority ranking, including: Read the set of correlation scores for syndrome differentiation and care, and expand the scores for each syndrome tendency and care priority according to object identifier, assessment time, syndrome category, care category, score source, field source, data credibility weight, time decay weight and credibility mask to form a set of correlation scores to be calibrated; In Example 1, the scores for each syndrome tendency and each nursing priority are expanded so that each syndrome tendency score can be mapped to the latent variables of TCM syndrome and glucose metabolism risk on which it is based, and each nursing priority score can be mapped to the latent variables of lifestyle intervention and nursing compliance on which it is based.

[0054] Based on the set of correlation scores to be calibrated, the credibility of each syndrome tendency score and each nursing priority score is calibrated according to the data credibility weight, forming a set of correlation scores after credibility calibration. In Example 1, scores with clear sources, complete fields, confirmed units, normal missing value indicators, stable outlier handling status, and consistent records from multiple sources are adopted at a high level. Scores with unknown sources, missing fields, units to be verified, abnormal truncation, record conflicts, or unstable credibility mask display are adopted at a lower level, and the credibility calibration indicator of each score is retained.

[0055] Based on the credibility-calibrated set of related scores, the scores of each syndrome tendency and each nursing priority are time-calibrated according to the time decay weight, and the time calibration identifier of each score is retained to form a time-calibrated set of related scores. In Example 1, scores collected close to the assessment time, within the recent follow-up time window, or belonging to the current health management stage are adopted with a high degree of timeliness, while scores collected far from the assessment time, belonging to historical health management records, or with low update frequency are adopted with a lower degree of timeliness.

[0056] Based on the time-calibrated associated score set, the syndrome tendency scores corresponding to Qi and Yin deficiency, spleen deficiency with phlegm and dampness, liver stagnation with Qi stagnation, Qi stagnation with phlegm obstruction, and damp-heat accumulation are fused to form a calibrated and fused syndrome assessment set. In Example 1, during fusion, the relative high-low relationship, credibility calibration mark, time calibration mark and source tracing relationship between the scores of each syndrome tendency are preserved, and parallel or pending verification prompts are set for syndrome categories with similar scores, conflicting sources or insufficient credibility.

[0057] Based on the calibrated and fused syndrome assessment set and the time-calibrated correlation score set, the care priority scores corresponding to dietary care, exercise care and sleep and mood care are fused to form a calibrated and fused care assessment set. In Example 1, during the fusion process, the relative high-low relationship between the nursing priority scores, the credibility calibration mark, the time calibration mark, and the nursing execution source are preserved. For nursing categories with similar scores, missing execution records, or insufficient nursing follow-up records, parallel or pending verification prompts are set.

[0058] Based on the calibrated and fused syndrome assessment set and the calibrated and fused nursing care assessment set, a comprehensive health record assessment value is generated, and a comprehensive health record assessment value record is formed. In Example 1, the comprehensive health record assessment value is used to comprehensively represent the overall assessment status of the subject in the prediabetes syndrome differentiation and care health record. The comprehensive health record assessment value is formed by the syndrome tendency score, care priority score, data credibility weight, time decay weight, missing value identifier and outlier handling identifier, and retains its corresponding data source and time window information.

[0059] Based on the comprehensive assessment value records of health records and the syndrome assessment set after calibration and fusion, the syndromes of Qi and Yin deficiency, spleen deficiency and phlegm dampness, liver stagnation and Qi stagnation, Qi stagnation and phlegm obstruction and damp-heat accumulation are ranked according to the degree of syndrome tendency after calibration and fusion. The syndrome primary and secondary ranking is obtained, and the ranking basis, credibility calibration mark and time calibration mark are retained to form the syndrome primary and secondary ranking record. Based on the primary and secondary syndrome ranking records and the calibrated and integrated nursing assessment set, dietary care, exercise care, and sleep and emotional care are ranked according to the calibrated and integrated nursing priority to obtain the nursing intervention priority ranking and form a nursing intervention priority ranking record. In Example 1, the nursing intervention priority ranking is used to indicate the priority relationship of different nursing intervention directions in the current health record assessment, and retains the ranking basis, nursing follow-up source, lifestyle behavior source, credibility calibration mark and time calibration mark.

[0060] Based on the comprehensive assessment values ​​of health records, the ranking of syndromes, and the ranking of nursing intervention priorities, the results of the health record assessment for prediabetes syndrome differentiation and nursing care are obtained by structuring and encapsulating the data according to the subject identification and assessment time.

[0061] Example 2: This example uses a dynamic management project of health records for people with prediabetes, carried out by a community chronic disease management center in a prefecture-level city, as the application scenario. The project serves a population of approximately 120,000, of which 3,260 are high-risk individuals with prediabetes and are included in the chronic disease management system. From March to December 2024, the center conducted a comprehensive assessment and intervention management based on multi-source health records. Example 2 selected 512 individuals with prediabetes who had participated in follow-up for more than 6 months and had relatively complete data records as research subjects. Among them, 268 were male and 244 were female, with ages ranging from 35 to 68 years and an average age of 52.4 years.

[0062] Health record data sources include community physical examination systems, family doctor contract follow-up systems, wearable device data platforms, and traditional Chinese medicine constitution identification and consultation systems. Each participant is assigned a unified health record identifier upon initial inclusion, and their physical measurement data, fasting blood glucose, postprandial blood glucose, glycated hemoglobin, diet records, exercise records, sleep records, and traditional Chinese medicine four diagnostic records are continuously updated during subsequent management.

[0063] In practical application, community doctors first collected data from 512 individuals using multi-terminal devices. In Example 2, one individual, Mr. Li, a 52-year-old male, underwent his initial health assessment on March 15, 2024. His fasting blood glucose was recorded as 6.3 mmol / L, his 2-hour postprandial blood glucose as 8.7 mmol / L, his weight as 78.2 kg, his height as 168 cm, and his waist circumference as 96 cm. His medical history also showed symptoms of "dry mouth, weakness, fatigue, and restless sleep," a "red tongue with little saliva," and a "thready and rapid" pulse. His lifestyle records indicated that he averaged less than 4000 steps per day over the past three months, preferred high-carbohydrate staple foods, had irregular dinner times, and sometimes stayed up late.

[0064] The system standardizes fields for data from different systems. In Example 2, the "FBG" field in the physical examination system is mapped uniformly to the "fasting blood glucose" field in the follow-up system. Step counts uploaded from different devices are standardized to "average daily steps," and blood glucose is standardized to mmol / L and weight to kg. For individual abnormal records, such as a device uploading "99,999 steps," the system identifies it as an outlier and truncates it, marking the data as having low reliability. For missing data, such as sleep records not being uploaded for certain dates, it is marked as missing.

[0065] The system then aligns the data according to time windows, assigning Li's data from the past 30 days, the past 90 days, and the past year to different time windows, and assigning a time decay weight to each data point. In Example 2, the blood glucose data from the past 7 days had the highest weight, while the weight of data from three months ago decreased significantly.

[0066] During the feature modeling phase, the system divided Li's health record into four layers: glucose metabolism risk, traditional Chinese medicine syndrome manifestation, lifestyle behavior, and nursing compliance. The glucose metabolism risk layer includes indicators such as blood glucose, weight, and waist circumference; the traditional Chinese medicine syndrome manifestation layer includes textual features such as dry mouth, fatigue, tongue appearance, and pulse; the lifestyle behavior layer includes steps, dietary structure, and sleep duration; and the nursing compliance layer includes whether he / she attended follow-up appointments on time and followed dietary recommendations.

[0067] After processing with the Monarch Mixer sparse hybrid coding network, the system extracted a strong correlation between Li's glucose metabolism risk layer and TCM syndrome manifestation layer, that is, the correlation between high blood sugar and Qi and Yin deficiency syndrome is enhanced. At the same time, the lifestyle behavior layer and the care compliance layer show coupling characteristics of insufficient exercise and poor compliance.

[0068] In the hierarchical decoupling phase, the system generates latent variables for Li's glucose metabolism risk, traditional Chinese medicine syndrome, lifestyle intervention, and nursing compliance, respectively, and avoids confounding between different factors through orthogonal constraints. In Example 2, elevated blood glucose and emotional stress are assigned to different latent spaces instead of being mixed into a single score.

[0069] During the assessment phase, the system output the following syndrome tendency scores for Mr. Li: Qi and Yin deficiency 0.72, Spleen deficiency with phlegm and dampness 0.48, Liver Qi stagnation 0.36, and Damp-heat retention 0.29; and the following nursing priority scores: Dietary care 0.81, Exercise care 0.87, and Sleep and emotional care 0.65. Based on these scores, the system ranked the nursing priority as Exercise care > Dietary care > Sleep and emotional care.

[0070] To verify the effectiveness of this invention, a comparison was made between the traditional method and the method of this invention in the same batch of 512 subjects. The traditional method uses a single-index scoring model, that is, it calculates a weighted average based on fasting blood glucose, body mass index, and questionnaire score, without performing multi-source data fusion and feature decoupling.

[0071] Regarding the training samples, the method of this invention uses approximately 180,000 time-series health records from 512 individuals as training input, including approximately 42,000 blood glucose records, approximately 68,000 exercise records, approximately 35,000 diet records, and approximately 35,000 TCM consultation texts; the traditional method only uses each person's most recent physical examination data and questionnaire scores, totaling 512 samples.

[0072] Regarding the accuracy of the assessment, the results of manual diagnosis by community doctors were used as a reference standard, and a comparative analysis was conducted on 512 individuals. The consistency rate of the method of this invention in syndrome classification was 86.3%, while that of the traditional method was 68.5%; in terms of the consistency rate in determining the priority of nursing care, the method of this invention was 83.7%, while that of the traditional method was 61.2%.

[0073] In terms of stability, when the same object was evaluated three times at different time points, the average fluctuation range of the method of the present invention was ±6.2%, while the fluctuation range of the traditional method was ±15.8%, indicating that the present invention is more robust to data noise and time changes.

[0074] Regarding the actual intervention effect, 512 people were followed up for 6 months. Among the 256 people who received intervention guided by the method of this invention, their fasting blood glucose decreased by an average of 0.48 mmol / L, their weight decreased by an average of 2.6 kg, and their daily steps increased to more than 7,200 steps. Among the 256 people who received intervention using the traditional method, their fasting blood glucose decreased by an average of 0.21 mmol / L, their weight decreased by 1.1 kg, and their daily steps increased to 5,100 steps.

[0075] In terms of time cost, the method of this invention automatically completes data processing and evaluation, with an average processing time of 1.8 seconds per person, while the traditional manual + rule-based method takes an average of about 3.5 minutes.

[0076] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for managing and assessing health records for prediabetes based on syndrome differentiation and care, characterized in that, include: Multi-source health record data of the subjects to be evaluated were collected and fields were standardized and merged to form a set of original health records for the diagnosis and treatment of prediabetes. Based on the original health record set of prediabetes diagnosis and treatment and after preprocessing, a set of time-series health record features with confidence mask is obtained. Based on the feature set of time-series health records with credibility masks, the features are grouped according to the glucose metabolism risk layer, the TCM syndrome manifestation layer, the lifestyle behavior layer, and the care compliance layer to form a hierarchical health record feature sequence. Based on the feature sequences of hierarchical health records, a Monarch Mixer sparse hybrid coding network is constructed to obtain hierarchical sparse hybrid feature representations; Hierarchical feature decoupling is performed based on hierarchical sparse hybrid feature representation to obtain a hierarchical decoupled set of latent variables in health records. Based on the hierarchical decoupled set of latent variables in health records, a syndrome-nursing care correlation assessment matrix is ​​constructed to form a set of syndrome-nursing care correlation scores. Based on the syndrome differentiation and nursing care correlation score set combined with data credibility weight and time decay weight, the scores of each syndrome tendency and each nursing care priority are calibrated and integrated to generate a comprehensive health record assessment value, syndrome priority ranking and nursing intervention priority ranking, and obtain the health record assessment results of syndrome differentiation and nursing care for prediabetes.

2. The method for managing and assessing health records for prediabetes based on syndrome differentiation and care, as described in claim 1, is characterized in that... The process of collecting and unifying the fields of the multi-source health record data of the subjects to be evaluated, and then merging them, includes: A master index for health record collection is established for the subjects to be evaluated, resulting in a set of master index records for the subjects; Based on the object's master index record set, multi-source health record data of the object to be evaluated is collected. During the collection, the data source, collection method, original field name, original field value, collection time, and record responsibility entity corresponding to each data item are retained to form a multi-source original collection record set. Based on a set of original data collected from multiple sources, a unified dictionary of fields is established to form a unified mapping set of fields. Based on the unified field mapping set, the multi-source original collection record set is classified by field. Fields that do not belong to the same semantic category are stored separately, and fields that belong to the same semantic category but have different sources are merged into the same category. Each merged field is attached with a field category identifier and a source category identifier to form a health record record set classified by field semantics. Based on the health record record set categorized by field semantics, records of the same object to be evaluated are aggregated according to object identifier. Records of the same object formed at different collection times are established in chronological order according to the collection time. Records of the same object from different data sources but collected at the same or similar time are archived in parallel according to the data source. Record status identifiers are set for duplicate records, supplementary records, revised records and conflicting records respectively, forming a time-series merged record set at the object dimension. Based on the object-dimensional time-series merged record set, each health record is checked for completeness and traceability. Records that pass the checks are retained as valid merged records, while those that fail the checks are retained as merged records pending verification, forming a merged health record set with record status identifiers, thus forming the original health record set for prediabetes diagnosis and care.

3. The method for managing and assessing health records for prediabetes based on syndrome differentiation and care, as described in claim 1, is characterized in that... The process of preprocessing the original health records based on the syndrome differentiation and care for prediabetes includes: Read the health record records corresponding to each subject to be assessed from the original health record set of prediabetes diagnosis and care, and form a record set with expanded field types; Expand the record set based on field type and perform unit unification processing on data fields from different sources to obtain a set of health record fields with unified units; Based on the unified set of health record fields of the unit, outlier identification and outlier truncation are performed on numerical fields, while retaining the source of outlier, outlier type and truncation mark, to form a set of health record fields after outlier processing. Based on the set of health record fields after outlier processing, missing fields are marked with missing values ​​to obtain a set of health record fields with missing value labels. Based on the set of health record fields with missing value identifiers, time window alignment is performed using the assessment time of the subject to be assessed as the benchmark to form a time window aligned time series health record field set. Based on the time-window aligned time-series health record field set, a data credibility weight is generated for each type of field, forming a time-series health record field set with data credibility weights; Based on the set of time-series health record fields with data credibility weights, a time decay weight is generated for each type of field to form a set of time-series health record fields with time decay weights. Based on the set of time-series health record fields with time decay weights, the data credibility weight, time decay weight, missing value identifier, outlier handling identifier, unit verification identifier, and source difference identifier are combined into a credibility mask. The credibility mask is then encapsulated synchronously with the corresponding health record fields to obtain a set of time-series health record features with credibility masks.

4. The method for managing and assessing health records for prediabetes based on syndrome differentiation and care, as described in claim 1, is characterized in that... The time-series health record feature set based on credibility masks is grouped according to the following layers: glucose metabolism risk layer, traditional Chinese medicine syndrome manifestation layer, lifestyle behavior layer, and care compliance layer, including: Read the feature set of time-series health records with a credibility mask, parse each health record field, and form a groupable health record field set; Based on the groupable health record field set, the fields related to glucose metabolism screening and metabolic status are classified into the glucose metabolism risk layer, resulting in the glucose metabolism risk layer feature set. Based on the feature set of the glucose metabolism risk layer and the groupable health record field set, the fields related to the TCM syndrome differentiation are classified into the TCM syndrome manifestation layer, thus obtaining the feature set of the TCM syndrome manifestation layer. Based on the TCM syndrome manifestation feature set and the groupable health record field set, the fields related to daily health behaviors are classified into the lifestyle behavior layer, thus obtaining the lifestyle behavior layer feature set. Based on the lifestyle behavior layer feature set and the groupable health record field set, the fields related to nursing execution and follow-up cooperation are classified into the nursing compliance layer, thus obtaining the nursing compliance layer feature set; Based on the feature sets of the glucose metabolism risk layer, the feature sets of the lifestyle behavior layer, and the feature sets of the care compliance layer, the continuous indicators are converted into normalized numerical embeddings to form a continuous indicator embedding set. Based on the feature sets of the glucose metabolism risk layer, the feature sets of the lifestyle behavior layer, and the feature sets of the care compliance layer, the categorical indicators are converted into category embeddings to form a categorical indicator embedding set. Based on the feature set of TCM syndrome manifestation layer, the text information of TCM four diagnostic methods is converted into symptom word embeddings to form a TCM four diagnostic method symptom word embedding set. Based on continuous indicator embedding sets, categorical indicator embedding sets, and TCM four diagnostic symptoms word embedding sets, hierarchical position codes are added to the glucose metabolism risk layer, TCM syndrome manifestation layer, lifestyle behavior layer, and nursing compliance layer, respectively. Time position codes are added to the features arranged according to time windows in each layer to form a hierarchical health record feature sequence.

5. The method for managing and assessing health records for prediabetes based on syndrome differentiation and care, as described in claim 1, is characterized in that... The construction of the Monarch Mixer sparse hybrid coding network based on the feature sequences of hierarchical health records includes: Read the feature sequence of the hierarchical health record, and establish hierarchical input sequences according to the glucose metabolism risk layer, TCM syndrome manifestation layer, lifestyle behavior layer and nursing compliance layer respectively. In each level input sequence, retain the feature order, field semantics, time position encoding, hierarchical position encoding, data credibility weight, time decay weight and credibility mask corresponding to that level to form a set of hierarchical input feature blocks. Based on the hierarchical input feature block set, the input sequence of each level is divided into blocks to form a set of feature blocks to be mixed within the layer; Based on the set of feature blocks to be mixed within the layer, the first diagonal mapping is performed within each layer to form the first layer local mapping feature set; Based on the local mapping feature set within the first layer, cross-block permutation mapping is performed to form a cross-block permutation feature set within the layer. Based on the intra-layer cross-block permutation feature set, perform the second block diagonal mapping to form the second intra-layer local mapping feature set; Based on the local mapping feature set within the second layer, gated residual normalization mapping is performed. The features after mixing within the layer are fused with the corresponding features before entering the Monarch mixing unit within the layer in a preservative manner. The participation degree of different features is controlled according to the confidence mask, so that the local combination relationship within the same layer can be expressed without losing the original field semantics, thus forming the Monarch mixing feature set within the layer. Based on the intra-layer Monarch mixed feature set, an inter-layer feature block set to be mixed is constructed according to the hierarchical correspondence between the glucose metabolism risk layer, the TCM syndrome manifestation layer, the lifestyle behavior layer, and the nursing compliance layer, thus forming an inter-layer feature block set to be mixed. Based on the set of inter-layer feature blocks to be mixed, cross-layer block diagonal mapping, hierarchical permutation mapping and confidence-weighted residual mapping are performed sequentially to form an inter-layer Monarch mixed feature set. Based on the inter-layer Monarch hybrid feature set, the local combination relationship within the layer and the global association relationship between the layers are uniformly encapsulated, and the hierarchical identifier, time position code, confidence mask and source backtracking identifier corresponding to each feature are retained to obtain the hierarchical sparse hybrid feature representation.

6. The method for managing and assessing health records for prediabetes based on syndrome differentiation and care, as described in claim 1, is characterized in that... The hierarchical feature decoupling process based on hierarchical sparse hybrid feature representation includes: Read the hierarchical sparse hybrid feature representation and perform hierarchical feature decoupling to form a set of hierarchical hybrid features to be decoupled; Based on the set of hierarchical hybrid features to be decoupled, the features corresponding to blood glucose status, anthropometric status, metabolic risk status and related historical health management status are projected into the latent space of glucose metabolism risk to form a set of latent variables of glucose metabolism risk. Based on the set of latent variables for glucose metabolism risk and the set of hierarchical mixed features to be decoupled, features related to inspection, auscultation, inquiry, palpation, symptom lexical, tongue appearance, pulse appearance, emotional expression and TCM syndrome tendency are projected into the TCM syndrome latent space to form a set of TCM syndrome latent variables. Based on the latent variable set of TCM syndromes and the hierarchical mixed feature set to be decoupled, features related to dietary structure, exercise habits, sedentary state, sleep state, emotional regulation, self-management behavior and lifestyle improvement space are projected into the latent space of lifestyle intervention to form the latent variable set of lifestyle intervention. Based on the latent variable set of lifestyle intervention and the hierarchical mixed feature set to be decoupled, features related to follow-up visits, implementation of nursing recommendations, implementation of dietary care, implementation of exercise care, implementation of sleep and mood care, participation in health education and the completeness of self-monitoring records are projected into the nursing compliance latent space to form the nursing compliance latent variable set. Based on the latent variable set of glucose metabolism risk, the latent variable set of traditional Chinese medicine syndrome, the latent variable set of lifestyle intervention, and the latent variable set of nursing compliance, orthogonal constraint processing is performed to form the latent variable set after orthogonal constraint. Based on the set of latent variables after orthogonal constraints, perform reconstruction consistency constraint processing to form a set of latent variables after reconstruction consistency constraints. Based on the set of latent variables after reconstructing the consistency constraints, a credibility mask constraint is performed to form a set of latent variables after credibility mask constraints. Based on the set of latent variables constrained by credibility mask, the latent variable sets of glucose metabolism risk, traditional Chinese medicine syndrome, lifestyle intervention, and nursing compliance are encapsulated according to object identifier, assessment time, and hierarchical identifier to obtain a hierarchical decoupled set of health record latent variables.

7. The method for managing and assessing health records for prediabetes based on syndrome differentiation and care, as described in claim 1, is characterized in that... The method for constructing a syndrome-care correlation assessment matrix based on a hierarchically decoupled set of latent variables in health records includes: Read the latent variable set of hierarchical decoupled health records, extract the latent variable sets of glucose metabolism risk, TCM syndrome, lifestyle intervention, and nursing compliance, and maintain the correspondence between the four types of latent variables according to object identifier, assessment time, hierarchical identifier, and confidence mask to form the input set of latent variables to be assessed. Based on the set of latent variable inputs to be evaluated, the syndrome-care association assessment matrix is ​​constructed to form a set of syndrome-side association structures. Based on the syndrome-related structure set, the latent space of TCM syndromes and the latent space of glucose metabolism risk are interactively mapped to obtain the syndrome interaction mapping feature set. Based on the syndrome interaction mapping feature set, syndrome tendency scores are formed for Qi and Yin deficiency, spleen deficiency with phlegm and dampness, liver stagnation with Qi stagnation, Qi stagnation with phlegm obstruction and damp-heat accumulation, forming a syndrome tendency score set. Based on the set of latent variable inputs to be evaluated, the nursing care side structure of the syndrome-nursing care correlation evaluation matrix is ​​constructed, forming a set of nursing care side correlation structures; Based on the set of care-side association structures, the implicit space of lifestyle intervention and the implicit space of care compliance are interactively mapped to obtain the set of care-side interaction mapping features. Based on the nursing interaction mapping feature set, nursing priority scores are generated for dietary nursing, exercise nursing and sleep and emotional nursing respectively, forming a nursing priority score set; Based on the syndrome tendency score set and the nursing care priority score set, the scores are matrix-encapsulated according to the object identifier, assessment time, syndrome category, nursing care category and credibility mask to form a dialectical nursing care correlation score set.

8. The method for managing and assessing health records for prediabetes based on syndrome differentiation and care, as described in claim 1, is characterized in that... The method of calibrating and fusing the scores for each syndrome tendency and each care priority based on the dialectical care correlation score set, combined with data reliability weights and time decay weights, includes: Read the set of correlation scores for syndrome differentiation and care, and expand the scores for each syndrome tendency and care priority according to object identifier, assessment time, syndrome category, care category, score source, field source, data credibility weight, time decay weight and credibility mask to form a set of correlation scores to be calibrated; Based on the set of correlation scores to be calibrated, the credibility of each syndrome tendency score and each nursing priority score is calibrated according to the data credibility weight, forming a set of correlation scores after credibility calibration. Based on the credibility-calibrated set of related scores, the scores of each syndrome tendency and each nursing priority are time-calibrated according to the time decay weight, and the time calibration identifier of each score is retained to form a time-calibrated set of related scores. Based on the time-calibrated associated score set, the syndrome tendency scores corresponding to Qi and Yin deficiency, spleen deficiency with phlegm and dampness, liver stagnation with Qi stagnation, Qi stagnation with phlegm obstruction, and damp-heat accumulation are fused to form a calibrated and fused syndrome assessment set. Based on the calibrated and fused syndrome assessment set and the time-calibrated correlation score set, the care priority scores corresponding to dietary care, exercise care and sleep and mood care are fused to form a calibrated and fused care assessment set. Based on the calibrated and fused syndrome assessment set and the calibrated and fused nursing care assessment set, a comprehensive health record assessment value is generated, and a comprehensive health record assessment value record is formed. Based on the comprehensive assessment value records of health records and the syndrome assessment set after calibration and fusion, the syndromes of Qi and Yin deficiency, spleen deficiency and phlegm dampness, liver stagnation and Qi stagnation, Qi stagnation and phlegm obstruction and damp-heat accumulation are ranked according to the degree of syndrome tendency after calibration and fusion. The syndrome primary and secondary ranking is obtained, and the ranking basis, credibility calibration mark and time calibration mark are retained to form the syndrome primary and secondary ranking record. Based on the primary and secondary syndrome ranking records and the calibrated and integrated nursing assessment set, dietary care, exercise care, and sleep and emotional care are ranked according to the calibrated and integrated nursing priority to obtain the nursing intervention priority ranking and form a nursing intervention priority ranking record. Based on the comprehensive assessment values ​​of health records, the ranking of syndromes, and the ranking of nursing intervention priorities, the results of the health record assessment for prediabetes syndrome differentiation and nursing care are obtained by structuring and encapsulating the data according to the subject identification and assessment time.

9. A health record management and assessment system for prediabetes based on syndrome differentiation and care, characterized in that, include: The data collection module collects and unifies the fields of the multi-source health record data of the subjects to be evaluated, and merges them to form a set of original health records for the diagnosis and treatment of prediabetes. The preprocessing module preprocesses the original health records set for prediabetes diagnosis and treatment to obtain a set of time-series health record features with a credibility mask. The feature grouping module, based on the time-series health record feature set with credibility mask, groups the features according to the glucose metabolism risk layer, the TCM syndrome manifestation layer, the lifestyle behavior layer, and the care compliance layer, forming a hierarchical health record feature sequence; The Monarch Mixer sparse hybrid coding module constructs a Monarch Mixer sparse hybrid coding network based on hierarchical health record feature sequences to obtain hierarchical sparse hybrid feature representations. The decoupling module performs hierarchical feature decoupling processing based on hierarchical sparse hybrid feature representation to obtain a hierarchical decoupled set of latent variables in health records. The syndrome differentiation module, based on the hierarchical decoupling of the latent variable set of health records, constructs a syndrome-nursing care correlation assessment matrix, forming a syndrome differentiation and nursing care correlation score set; The output module, based on the syndrome differentiation and nursing care correlation score set combined with data credibility weight and time decay weight, calibrates and integrates the scores of each syndrome tendency and each nursing care priority to generate a comprehensive health record assessment value, syndrome priority ranking, and nursing intervention priority ranking, thus obtaining the health record assessment results of syndrome differentiation and nursing care for prediabetes.

Images