Health trend view generation method and system fusing multi-source heterogeneous feature mining

Abstract

The present application relates to the field of medical data processing, and provides a health trend view generation method and system fusing multi-source heterogeneous feature mining, which comprises the following steps: obtaining original follow-up data, sequentially performing original record analysis, cycle division and time correspondence establishment on the original follow-up data, and encapsulating and generating cycle follow-up data packets; extracting text health features, index change features and path execution results according to the cycle follow-up data packets, fusing and generating standard disease course feature records to construct a standard disease course feature table; extracting path node association feature records according to the standard disease course feature table, performing source direction determination and evidence summary encapsulation, and obtaining a disease course evidence unit table; generating a doctor-end health trend view and a patient-end health trend view according to the disease course evidence unit table, and generating a double-track health trend view based on the two views. The present application fuses multi-source heterogeneous follow-up features and standardized disease course paths, and constructs a health trend view generation mechanism for the doctor and patient ends.

Description

Technical Field

[0001] This invention relates to the field of medical data processing, and in particular to a method and system for generating health trend views by integrating multi-source heterogeneous feature mining. Background Technology

[0002] With the increasing number of follow-up data sources in the whole course of disease management, health status monitoring, anomaly identification, and trend display have become important aspects of continuous management. Existing follow-up data is mostly scattered across text records, indicator records, and path execution records, lacking periodic correlation and dual-view representation, which can easily lead to delays in physician verification and inconsistencies in patient understanding. How to integrate and process multi-source follow-up data to generate a health trend view for both doctors and patients has become an urgent technical problem to be solved.

[0003] However, current technologies still face many challenges. In the context of cyclical follow-up for full-course disease management, patients' text records, indicator records, and pathway execution records are often stored in a scattered manner. Existing processing methods mostly rely on presenting data separately according to its source, making it difficult to form a continuous disease progression thread around the same patient, the same follow-up period, and related pathway nodes. When doctors assess changes in disease progression, they usually need to repeatedly review and compare different records. Abnormal symptoms, indicator fluctuations, or pathway execution deviations may not be incorporated into the same judgment process in a timely manner due to the scattered evidence. Especially when textual descriptions, indicator changes, or pathway execution results are inconsistent, without a differentiated view organization for both doctors and patients, the single trend conclusion obtained by the patient may not match the comprehensive judgment of the doctor based on multi-source heterogeneous evidence, leading to misunderstandings between doctors and patients and delays in follow-up treatment. Summary of the Invention

[0004] To achieve the above objectives, this invention provides a method for generating health trend views by integrating multi-source heterogeneous feature mining, the specific technical solution of which is as follows:

[0005] The original follow-up data is obtained, and the original record parsing, period division and time correspondence establishment are performed on the original follow-up data in sequence, and the data is packaged to generate periodic follow-up data packets.

[0006] Based on the periodic follow-up data package, extract text health features, indicator change features and path execution results, integrate the text health features, indicator change features and path execution results to generate standard disease course feature records to construct a standard disease course feature table;

[0007] Based on the standard disease course feature table, path node associated feature records are extracted, and the source direction of the path node associated feature records is determined and evidence summary is encapsulated to obtain the disease course evidence unit table.

[0008] Generate a doctor-side health trend view and a patient-side health trend view based on the disease course evidence unit table, and generate a dual-track health trend view based on the doctor-side health trend view and the patient-side health trend view.

[0009] Further, the process of acquiring raw follow-up data, sequentially performing raw record parsing, period division, and time correspondence establishment on the raw follow-up data, and encapsulating it to generate a periodic follow-up data packet includes:

[0010] The patient identifier, collection time, source type, and original content of each original record in the original follow-up data are analyzed and combined to generate source registration data;

[0011] The period number of each source registration record in the source registration data is calculated based on the initial filing time and the follow-up period step, and the source registration data is classified according to the period number and source type to generate period-aligned data;

[0012] Determine the minimum time distance between the record to be matched and the target indicator record in the periodic alignment data. Based on the minimum time distance and the preset corresponding time upper limit, establish the time correspondence between the record to be matched and the target indicator record, and encapsulate and generate a periodic follow-up data package.

[0013] Further, the determination of the minimum time distance between the record to be matched and the target indicator record in the periodic alignment data, based on the minimum time distance and a preset corresponding time upper limit, establishes a time correspondence between the record to be matched and the target indicator record, and encapsulates and generates a periodic follow-up data package, including:

[0014] Within the follow-up time interval corresponding to the same patient and the same cycle number, the text source record or the path source record is identified as the record to be matched;

[0015] Within the follow-up time interval, the time distance between the collection time of each indicator source record and the collection time of the record to be matched is calculated, and the indicator source record with the smallest time distance is selected as the target indicator record.

[0016] When the minimum time distance between the record to be matched and the target indicator record is less than or equal to the preset corresponding time limit, a time correspondence relationship between the record to be matched and the target indicator record is established;

[0017] The text source records, indicator source records, path source records, and established time correspondences are encapsulated into periodic follow-up data packages.

[0018] Furthermore, based on the periodic follow-up data package, text health features, indicator change features, and path execution results are extracted. These features are then fused to generate standard disease progression feature records to construct a standard disease progression feature table, including:

[0019] The periodic follow-up data package includes a set of text source records, a set of indicator source records, and a set of path source records;

[0020] Normalization and semantic segmentation are performed on the set of text source records in the periodic follow-up data package to obtain semantic segments; health standard terms in the semantic segments are identified, and the semantic direction of the text is determined by combining semantic modifiers, and the text health features are encapsulated and generated.

[0021] Based on the health features of the text, the indicator observation window is determined in the periodic follow-up data package. The indicator records matched in the indicator source record set are read within the indicator observation window. The indicator change amount and direction are calculated based on the indicator records, and the indicator change features are encapsulated and generated.

[0022] The path execution result is determined based on the path source record set in the periodic follow-up data package. It is then associated with text health features and indicator change features to generate standard disease course feature records and determine disease course time markers. Finally, a standard disease course feature table is generated.

[0023] Further, the text source record set in the periodic follow-up data packet is normalized and segmented into semantic segments to obtain semantic segments; health standard terms in the semantic segments are identified, and the semantic direction of the text is determined by combining semantic modifiers, and text health features are encapsulated and generated, including:

[0024] The text content in the text source record set is standardized. The standardization process includes removing consecutive repeated characters, unifying full-width and half-width characters, converting synonyms into health standard terms, and retaining semantic modifiers related to health status judgment.

[0025] The synonym conversion involves converting non-standard health expressions in patient texts into corresponding standard health terms according to a medical terminology mapping table for diseases corresponding to standardized management paths.

[0026] Based on punctuation marks, line breaks, and semantic pause words pre-configured in the standardized management path, the normalized text content is divided into semantic segments to obtain at least one semantic segment.

[0027] The semantic fragments are matched with a medical thesaurus and a thesaurus to determine the health keywords corresponding to the health standard terms; an evidence summary is generated based on the semantic fragments containing the health keywords.

[0028] Based on the term type of the health standard terms corresponding to the health keywords, the text category is determined, and the semantic direction of the text is determined by combining the semantic modifiers within the preset word distance before and after the health keywords;

[0029] The health standard terms, evidence summaries, text categories, and text semantic directions are written into the text health feature record, and the text health feature record is written into the text health feature.

[0030] Furthermore, the step of extracting path node-related feature records based on the standard disease course feature table, determining the source direction of the path node-related feature records, and encapsulating evidence summaries to obtain a disease course evidence unit table, including:

[0031] Based on the standard disease course feature table, the node association time range is determined, and the text health features, indicator change features and path execution results of disease course time markers falling within the node association time range are extracted to generate path node association feature records.

[0032] Based on the path node association feature records, text source direction, indicator source direction and path source direction are generated respectively, and evidence integrity markers and consistency status are determined, and consistency status feature groups are encapsulated and generated.

[0033] Based on the consistency status feature group, view destination markers are generated, and text evidence summaries, indicator evidence summaries, and path execution summaries are extracted. Combined with the node association time range, a disease course evidence unit table is generated.

[0034] Furthermore, the step of generating text source direction, indicator source direction, and path source direction based on path node association feature records, and determining evidence integrity markers and consistency status, includes:

[0035] Read the path node identifier, text health feature set, indicator change feature set, and path execution feature set from the path node association feature record;

[0036] When the text health feature set is not empty, the text source direction is generated based on the text semantic direction in the text health feature set; when the text health feature set is empty, the text source direction is recorded as missing.

[0037] When the indicator change feature set is not empty, the indicator source direction is generated based on the indicator change direction in the indicator change feature set and whether the indicator value exceeds the indicator reference range; when the indicator change feature set is empty, the indicator source direction is recorded as missing.

[0038] When the path execution feature set is not empty, read the path execution direction from the path execution result that matches the path node identifier, and use it as the path source direction. When the path execution feature set is empty and the path node is not a missing node in the path execution record, record the path source direction as missing.

[0039] Based on the number of non-missing source directions among the text source direction, indicator source direction, and path source direction, an evidence integrity marker is generated, and a consistency state is generated based on the direction value relationship between the non-missing source directions.

[0040] Further, the step of generating a doctor-side health trend view and a patient-side health trend view based on the disease course evidence unit table, and generating a dual-track health trend view based on the doctor-side health trend view and the patient-side health trend view, includes:

[0041] The view destination markers in the disease course evidence unit table are merged to generate abnormal summary view items and view items to be verified. The view items are then sorted according to the time range of the evidence to generate a doctor-side health trend view.

[0042] Based on the doctor's health trend view, generate the patient-side constraint index, determine the patient-side deterministic trend output marker based on the patient-side constraint index and the disease course evidence unit table, combine the view destination marker to generate the patient-side view item, and generate the patient-side health trend view.

[0043] Perform consistency and source checks on the doctor-side and patient-side health trend views, generate page description data, and determine the dual-track view based on dual-track constraints. Figure 1 Consistent markers are used to encapsulate and generate a dual-track health trend view.

[0044] Further, the step of generating a patient-side constraint index based on the doctor-side health trend view, determining patient-side deterministic trend output markers based on the patient-side constraint index and the disease course evidence unit table, generating patient-side view items by combining the view destination markers, and generating a patient-side health trend view includes:

[0045] Read the view items to be verified in the doctor's health trend view, extract the path node identifiers in the view items to be verified, and generate the patient's constraint index;

[0046] Read the progress evidence units from the progress evidence unit table that are marked as patient recovery progress or patient record reminders, and generate a patient-side candidate set.

[0047] The path node identifiers of the disease course evidence units in the patient-side candidate set are compared with the path node identifiers in the patient-side constraint index to determine the patient-side deterministic trend output marker.

[0048] When the patient-side deterministic trend output flag is set to 1, a rehabilitation progress view item or a record reminder view item is generated based on the view destination flag.

[0049] When the patient-side deterministic trend output flag is set to 0, a neutral record prompt view item is generated;

[0050] The recovery progress view items, record reminder view items, and neutral record prompt view items are sorted according to the start time of the evidence time range corresponding to each view item to generate a patient-side health trend view.

[0051] Furthermore, the consistency and source checks are performed on the doctor-side health trend view and the patient-side health trend view, page description data is generated, and the dual-track view is determined based on the dual-track constraint back lookup. Figure 1 Consistent markers, encapsulated to generate a dual-track health trend view, including:

[0052] Perform consistency checks on patient identifiers and cycle numbers in the health trend views on the doctor's end and the health trend views on the patient's end;

[0053] After the consistency check passes, the path node identifiers in the doctor's health trend view and the patient's health trend view are extracted respectively, and the disease evidence unit with the same path node identifier is searched in the disease evidence unit table to perform source verification.

[0054] After the source verification is passed, the doctor's workstation page description data is generated based on the doctor's health trend view, and the patient's mobile page description data is generated based on the patient's health trend view.

[0055] Using the view items to be checked in the doctor's health trend view as the inspection objects, extract their path node identifiers and query the patient's constraint index to determine the dual-track view. Figure 1 Coherence markers;

[0056] Combine doctor workstation page description data, patient mobile page description data, and dual-track view. Figure 1 Incidence markers, patient identifiers, and cycle numbers are encapsulated together to generate a dual-track health trend view.

[0057] A health trend view generation system integrating multi-source heterogeneous feature mining is used to implement the above-mentioned health trend view generation method integrating multi-source heterogeneous feature mining. The system includes a follow-up encapsulation module, a feature fusion module, an evidence encapsulation module, and a view generation module.

[0058] The follow-up encapsulation module is used to acquire the original follow-up data, perform original record parsing, period division, and time correspondence establishment on the original follow-up data in sequence, and encapsulate it to generate a periodic follow-up data packet.

[0059] The feature fusion module is used to extract text health features, indicator change features and path execution results based on the periodic follow-up data package, and fuse the text health features, indicator change features and path execution results to generate standard disease course feature records to construct a standard disease course feature table.

[0060] The evidence encapsulation module is used to extract path node-related feature records based on the standard disease course feature table, determine the source direction of the path node-related feature records and encapsulate the evidence summary to obtain a disease course evidence unit table.

[0061] The view generation module is used to generate a doctor-side health trend view and a patient-side health trend view based on the disease course evidence unit table, and to generate a dual-track health trend view based on the doctor-side health trend view and the patient-side health trend view.

[0062] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0063] This invention improves the problem of scattered follow-up records and difficulty in organizing them according to the same follow-up cycle in the whole course of disease management by sequentially parsing the original records, dividing the period, and establishing the time correspondence of the original follow-up data, and encapsulating them into periodic follow-up data packages. This allows the scattered original follow-up data to be formed into a unified processing object according to the period and time correspondence.

[0064] This invention extracts text health features, indicator change features, and path execution results based on periodic follow-up data packages, and integrates the three types of features to generate a standard disease course feature record to construct a standard disease course feature table. This allows text descriptions, indicator changes, and path execution status to be expressed in a unified record, reducing the burden on doctors to manually compare and organize follow-up information from different sources.

[0065] This invention extracts path node-related feature records from a standard disease course feature table, determines their source direction, and encapsulates evidence summaries to obtain a disease course evidence unit table. This enables multi-source disease course evidence related to the same path node to form a identifiable and summarizable data unit, improving the problem of scattered abnormal or recovery evidence making it difficult to form a view input.

[0066] This invention generates a doctor-side health trend view and a patient-side health trend view based on the disease course evidence unit table, and further generates a dual-track health trend view. This allows the same disease course evidence to be presented in a differentiated manner to the doctor and patient, thus improving the information asymmetry problem caused by the inconsistency between the doctor's focus and the patient's understanding in the whole disease course management. Attached Figure Description

[0067] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0068] Figure 1 This is a flowchart illustrating the principle of the health trend view generation method that integrates multi-source heterogeneous feature mining according to the present invention.

[0069] Figure 2 This is a structural diagram of the source registration data of this invention;

[0070] Figure 3 This is a schematic diagram illustrating the construction of a time-based indicator observation window according to the present invention.

[0071] Figure 4 This is a functional module diagram of the health trend view generation system that integrates multi-source heterogeneous feature mining according to the present invention. Detailed Implementation

[0072] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] Example 1:

[0074] Please see Figure 1 As shown, this embodiment provides a method for generating a health trend view by fusing multi-source heterogeneous feature mining, including:

[0075] Step S1000: Obtain raw follow-up data For the original follow-up data The process sequentially executes raw record parsing, period division, and establishment of time correspondence, and encapsulates the data to generate periodic follow-up data packets. .

[0076] Specifically, this step involves obtaining the raw follow-up data. As input, the original follow-up data The original records are parsed, the period is divided, and the time correspondence is established sequentially to obtain the periodic follow-up data package. This step is applicable to the integration and processing of multi-source follow-up data in the whole-course management system. It is used to register the sources, divide the cycles, and establish time correspondences of doctor-patient interaction records, health indicator detection records, and pathway execution records, generating a periodic follow-up data package containing multi-source follow-up records and their time correspondences. This provides a unified data input for subsequent steps.

[0077] Further, step S1000 includes:

[0078] Step S1100: parse the raw follow-up data. Each original record Patient identification Collection time Source type and the original content Combine to generate source registration data .

[0079] In the specific implementation process, this step first reads the original follow-up data. Each record in the database is treated as a raw record, including a chat message, a metric upload record, and a path execution record. .in, This represents the sequence number of the original record, and its value is an integer greater than or equal to 1.

[0080] Next, from the original records Patient identifiers are parsed from the included patient-related information, time information, source information, and business content information. Collection time Source type and the original content Among them, patient identification. The unique identifier assigned to each patient by the full-process disease management system; data collection time. For this original record The time of the corresponding business event. If the time of the business event does not exist, the time when the system received the record is used, accurate to the minute; Source type. The value can be the text source, the indicator source, or the path source; original content. This corresponds to the business content recorded.

[0081] Further, see Figure 2 , Figure 2 This is a structural diagram of the source registration data of this invention.

[0082] Then, for different source types Corresponding original content Categorize and organize the text. Regarding text sources, the original content... This includes text content, sender identity, and message identifier; for the source of the indicator, the original content... This includes the indicator name, indicator value, unit of measurement, and data collection method; for the source path, the original content... This includes the path node name, planned execution time, actual execution time, and execution status. The execution status can be one of three values: completed, completed after timeout, or not completed.

[0083] Finally, it will be identified by the patient. Collection time Source type and the original content A source registration record is added to the source registration data. The source registration data From raw follow-up data Each original record It consists of separately generated source registration records.

[0084] For example, a patient sends a text message in a group saying "I'm a little out of breath walking today," and the pulse oximeter uploads a pulse oximetry record around the same time. This step extracts the patient identifier, collection time, source type, and original content from both original records, registering the text message as the text source and the pulse oximetry record as the indicator source. After registration, both records retain the same patient identifier while retaining their respective collection time and original content, which can be used for subsequent association of follow-up data from different sources based on patient identifier and collection time.

[0085] Step S1200, based on the initial file creation time and follow-up period step Calculate source registration data Periodic numbering of registration records from various sources and numbered according to periodicity and source type Source registration data Classify and generate periodically aligned data. .

[0086] In the specific implementation process, this step first follows the patient identification... Source registration data The patients were grouped to obtain a set of registration records for each patient.

[0087] Next, read the patient's initial registration date. The follow-up cycle length is configured according to a standardized management pathway that matches the patient's current stage of disease management. The initial filing time is mentioned above. This refers to the starting time for patients to enter the full-course management system; the standardized management path refers to a set of pre-set course management rules based on disease type, treatment stage, follow-up frequency requirements, and management goals, which includes at least the follow-up cycle step. Configuration details include path nodes, execution requirements, and time constraints; the follow-up period step size. The value can be either 7 days or 30 days. 7 days is used for the weekly follow-up pathway, and 30 days is used for the monthly follow-up pathway.

[0088] Based on the initial filing time and follow-up period step The timeline after patient registration is divided into continuous follow-up time intervals. Each follow-up time interval corresponds to a cycle number. This indicates which follow-up cycle the source registration record belongs to after the patient's file was established.

[0089] Then, based on the collection time Calculate the period number to which each source registration record belongs. The period number The specific calculation logic is as follows: based on the patient's first registration time. The data collection time will be used as the baseline time. Subtract the initial filing time The time difference between the occurrence of the source registration record and the establishment of the file is obtained; the time difference is divided by the follow-up period step size. Finally, the result of the division is rounded down to obtain the period number of the record. If the acquisition time Earlier than the initial filing time If so, the source registration record will not be included in the periodic alignment data. .

[0090] Subsequently, patients with the same identification will be... and the same period number The registration records are grouped into the same periodic dataset. Here, the periodic dataset represents the patient identifier. The corresponding patient in the first A multi-source record set formed within a follow-up time interval. The first... The follow-up time interval is based on the patient's initial record establishment time. and follow-up period step Confirmed, its start time is The end time is .

[0091] For source registration records grouped into the same periodic dataset, further classification is based on source type. Classify the text sources and collect the resulting collection of text source records. Indicator source record set and path source record set Write to this periodic dataset. Specifically, when the source type When a source is specified, the source registration record is written to the text source record set. When the source type When an indicator is a source, the registration record of that source is written into the indicator source record set. When the source type When a path originates from another path, the path origination record is written into the path origination record set. .

[0092] Each periodic dataset contains patient identifiers. and period number When a certain source type When there are no source registration records within the follow-up time interval, the record set corresponding to the source type is set to an empty set, and the empty set is marked in the periodic dataset to indicate that the data of the source type is missing within the period, rather than being a data processing omission.

[0093] Finally, this step outputs periodically aligned data. The periodic alignment data It is a collection consisting of several periodic datasets, each periodic dataset including patient identifiers. Periodic numbering Text source record set Indicator source record set and path source record set .

[0094] For example, the current standardized management pathway for patients is configured with a 7-day follow-up cycle. Within a specific follow-up time interval, participants uploaded their blood pressure records on day 1, sent a description of discomfort on day 3, and completed their medication adherence check-in on day 5. This step is based on the collection time of these three records. and the patient's first registration time The same period number was obtained through calculation. Blood pressure records are then written into the indicator source record set. Write inappropriate descriptions into the text source record set Write the medication attendance record into the path source record set. This generates a periodic dataset for the patient within the specified follow-up time interval. This periodic dataset is used in subsequent steps to analyze the correspondence between symptom descriptions, objective indicators, and pathway implementation within the same follow-up time interval.

[0095] Step S1300: Determine the periodic alignment data Minimum time distance between the record to be matched and the target indicator record Based on the minimum time distance Corresponding to the preset time limit Establish the time correspondence between the records to be matched and the target indicator records, and encapsulate and generate periodic follow-up data packages. .

[0096] In the specific implementation process, this step receives the periodic alignment data from step S1200. First, read the text source record set from each periodic dataset. Indicator source record set and path source record set Next, using the collection time of the text source record or path source record as a reference time, the indicator source record closest to the reference time is searched within the same follow-up time interval. Then, the time correspondence between the text source record, path source record, and indicator source record is determined based on the proximity of the collection times, and encapsulated into a periodic follow-up data package. .

[0097] Specifically, for a single text source record or path source record within the same patient and the same follow-up time interval, the text source record or path source record is designated as the record to be matched, and the collection time of the record to be matched is recorded as... At the same time, the first [number] [unit] within the same follow-up time interval [will be included]. The collection time of each indicator source record is recorded as follows: Calculate the time distance between the two. .

[0098] The time distance Indicates the record to be matched with the first... The time distance between the source records of the indicator is taken as the collection time of the record to be matched. Subtract the collection time of the source record of the aforementioned indicator The absolute value; This represents a count index variable used to identify the set of source records for the indicator in the current follow-up time interval. The source records of the indicators in the data.

[0099] Select time distance The record with the smallest source metric is used as the target metric record for the record to be matched, and this smallest time distance is denoted as . When the minimum time distance Less than or equal to the preset corresponding time limit If so, a time correspondence is established between the record to be matched and the target indicator record. This time correspondence includes the patient identifier of the record to be matched. Source type Patient identifiers recorded by target indicators The indicator name, indicator value and unit of measurement, and the collection time of the record to be matched. The time of data collection for the target indicator. and minimum time distance The preset corresponding time limit The time frame is 24 hours, determined based on the configuration of the effective duration for different indicator data within the standardized management path. If the minimum time distance... Greater than the preset corresponding time limit If not, no time correspondence is established.

[0100] Finally, the text source records, indicator source records, path source records, and the established time correspondence are encapsulated into a periodic follow-up data package. .

[0101] For example, a patient reports "chest tightness and shortness of breath" in a follow-up text at 8 PM, and their home blood pressure monitor uploads a blood pressure and heart rate record at 8:10 PM. Since the two records were collected close together, this step establishes a time correspondence between the text source record and the indicator source record. Subsequent analysis can then combine "chest tightness and shortness of breath" with changes in blood pressure and heart rate around the corresponding time to make judgments. Conversely, if the patient reports "chest tightness and shortness of breath" at 8 PM, but the most recent blood pressure record was collected three days ago, then that blood pressure record cannot be used as the corresponding indicator record for the current symptom occurrence. This step does not establish a time correspondence between the two to avoid subsequently using indicator records exceeding the corresponding time limit as the basis for current symptoms.

[0102] Step S2000: Based on the periodic follow-up data packet Extracting health features from text Characteristics of indicator changes The path execution results will include text health characteristics. Characteristics of indicator changes The results of path execution are integrated to generate standard disease course feature records to construct a standard disease course feature table. .

[0103] Specifically, this step uses the periodic follow-up data package from step S1300. As the basis for processing, feature extraction is performed on data objects belonging to different dimensions to obtain a standard disease course feature table. This step involves associating text records, health indicator records, and path execution records uploaded or generated by patients at different time points, according to the same patient, the same follow-up period, and similar occurrence times, to form a standard disease progression feature record that represents the correspondence between symptom descriptions, indicator changes, and management behaviors. This provides structured input for subsequently generating disease evolution trends and personalized health recommendations.

[0104] Further, step S2000 includes:

[0105] Step S2100, process the periodic follow-up data packet. The collection of text source records in Normalization and semantic segmentation are performed to obtain semantic segments; health standard terms are identified in the semantic segments, and the semantic direction of the text is determined by combining semantic modifiers. Encapsulate and generate text health features .

[0106] In the specific implementation process, this step first takes the periodic follow-up data packet output in step S1300. Read the text source record set Each text source record is treated as a pending text record. For each pending text record, its patient identifier is read. Periodic numbering Collection time Message identifiers and text content.

[0107] Next, the text content undergoes normalization processing. This normalization includes removing consecutive repeated characters, standardizing full-width and half-width characters, converting synonymous expressions into standard health terms, and retaining semantic modifiers related to health status assessment. These semantic modifiers include negation words, degree words, time words, mitigation words, emphasis words, and completion words, used to determine the semantic direction of the standard health terms within the text.

[0108] The synonym conversion involves converting non-standard health expressions in patient texts into corresponding standard health terms according to a medical terminology mapping table for diseases corresponding to the standardized management path. The medical terminology mapping table includes a pre-configured medical glossary and a synonym mapping table. The medical glossary consists of symptom terms, drug names, indicator names, and lifestyle terms for diseases corresponding to the standardized management path, and each standard health term is configured with a term type. The synonym mapping table records the correspondence between commonly used patient expressions and standard health terms. The term types include symptom terms, medication feedback terms, and lifestyle terms. For example, in the hypertension management path, "dizziness," "chest tightness," "palpitations," and "shortness of breath" are symptom terms; "missed dose" and "taken medication" are medication feedback terms; and "low-salt diet" and "exercise" are lifestyle terms. The synonym mapping table uniformly maps patient text expressions such as "a little short of breath" and "shortness of breath" to the standard health term "shortness of breath."

[0109] Then, the standardized text content is segmented into semantic segments according to punctuation marks, line breaks, and semantic pause words pre-configured in the standardized management path, resulting in at least one semantic segment. The semantic pause words are used to indicate pauses or transitions between adjacent health statements in the patient's text, and include words such as "but," "however," "in addition," "also," and "then."

[0110] Next, health keywords are identified in each semantic segment. Specifically, based on the shortest and longest lengths of entries in the medical thesaurus and the thesaurus, the shortest length is used as the minimum truncation length of the candidate text segment, and the longest length is used as the maximum truncation length of the candidate text segment, thereby determining the truncation length range of the candidate text segment; then, candidate text segments are truncated from the semantic segments, and the candidate text segments are matched with the medical thesaurus and the thesaurus; if the candidate text segment is the same as a health standard entry in the medical thesaurus, or can be mapped to a health standard entry in the medical thesaurus through the thesaurus, then the candidate text segment is determined as a health keyword, and its corresponding health standard entry is determined as the health standard entry corresponding to the health keyword.

[0111] Subsequently, the text category is determined based on the term type of the health standard terms corresponding to the health keywords. The text category includes symptom description, medication feedback, or lifestyle feedback. Specifically, when a health standard term belongs to the symptom category, the text category is determined as symptom description; when a health standard term belongs to the medication feedback category, the text category is determined as medication feedback; and when a health standard term belongs to the lifestyle category, the text category is determined as lifestyle feedback. If more than two types of health standard terms are identified in the same text source record, corresponding text health feature records are generated according to different text categories.

[0112] For each text health feature record to be generated, identify the patient from the text source record. Periodic numbering Message identifier and collection time The patient identifier, cycle number, message identifier, and text occurrence time fields are written into the text health feature record, and the health standard terms and text categories corresponding to the health keywords are written into the text health feature record.

[0113] Then, an evidence summary is generated based on the semantic fragment containing health keywords, and this evidence summary is used as the evidence field for the corresponding text health feature record. The length of the evidence summary is limited to 5 to 60 Chinese characters; if the semantic fragment exceeds 60 Chinese characters, then 20 Chinese characters before and after the health keywords are retained to form the evidence summary.

[0114] Subsequently, the semantic direction of the text was determined by combining the semantic modifiers within the preset word spacing before and after the health keywords. and the semantic direction of the text This serves as the directional field for recording the corresponding text health features. The preset word distance is within 8 Chinese characters before and after the health keyword, used to limit the scope of semantic modifiers on the health keyword. Indicates the first A text health feature record; the semantic direction of the text The value can be 1, 0 or -1.

[0115] Specifically, when semantic modifiers such as "no longer," "no longer," "relieved," and "improved," indicating the disappearance or reduction of symptoms or the completion of management actions, appear within the preset word spacing before and after health keywords, the semantic direction of the text will be adjusted. The value is 1; when semantic modifiers such as "worsened," "significant," "relapsed," "missed dose," and "incomplete" appear, indicating worsening symptoms, recurrence of symptoms, or incomplete management, the semantic direction of the text is... The value is -1; when no semantic modifiers indicating improvement or abnormality appear, and the text only indicates data upload, inspection completion, or general record, the text semantic direction is... The value is 0.

[0116] Finally, each text health feature record is written into the text health feature. Among them, the text health features Follow-up data packets from the same period It consists of at least one text health feature record generated internally.

[0117] For example, in a hypertension management pathway, a patient sends a message at 20:30 on October 6, 2025: "My dizziness has been more pronounced these past two days than last week, and I also missed a dose of medication last night." This step identifies "dizziness" and "missed dose." A symptom description record is generated for "dizziness," and a medication feedback record is generated for "missed dose." Since "pronounced more pronounced" indicates worsening dizziness, and "missed dose" indicates failure to take medication on time, the semantic direction of the two records is... Both records are -1. Both records retain the patient identifier, cycle number, message identifier, and text occurrence time for this text.

[0118] Step S2200, based on text health features In periodic follow-up data packets In the middle, determine the indicator observation window, and read the indicator source record set within the indicator observation window. The system matches the indicator records and calculates the change in the indicator based on these records. and the direction of indicator change Encapsulation and generation of indicator change characteristics .

[0119] In the specific implementation process, this step first reads the text health features from step S2100. And one of the text health feature records is used as the current text feature record to be processed.

[0120] Next, according to the patient identifier in the current text feature record to be processed. and period number The periodic follow-up data packet output in step S1300 In the middle, read the set of matching indicator source records. If no matching set of indicator source records is found. If the health feature record for this text is not generated, the indicator change features will not be recorded. The health characteristics of the text are recorded and retained in step S2300.

[0121] Then, based on the health standard terms and text categories in the current text feature record to be processed, the name of the indicator to be analyzed is determined in the symptom-indicator correspondence table in the standardized management path. The symptom-indicator correspondence table records the correspondence between health standard terms, text categories, and the name of the indicator to be analyzed. For example, "dizziness" corresponds to systolic blood pressure, diastolic blood pressure, and heart rate; "palpitations" corresponds to heart rate; and "shortness of breath" corresponds to blood oxygen saturation and heart rate.

[0122] If a health standard term corresponds to more than two indicator names to be analyzed, then each indicator name will be processed separately. If no corresponding indicator name exists, then no indicator change feature corresponding to that text health feature record will be generated. The health characteristics of the text are recorded and retained in step S2300.

[0123] Subsequently, the time correspondence established in step S1300 is used to determine the indicator observation window. Specifically, based on the message identifier of the currently pending text feature record, the periodic follow-up data packet is used. In the time correspondence, search for a matching target indicator record. If a matching target indicator record exists, then record the collection time of that target indicator record. The central time as an indicator observation window If no matching target indicator record exists, the occurrence time of the text in the current text feature record to be processed will be used as the center time of the indicator observation window. .

[0124] The indicator observation window is represented as follows: .in, This indicates the duration of looking back, with a value of 2 hours. This indicates the duration of observation, with a value of 4 hours. For indicators to be analyzed that are collected only once a day, the observation window for that indicator can be expanded to [a larger value]. Whether to expand depends on the collection frequency of the indicator to be analyzed in the standardized management path.

[0125] Further, see Figure 3 , Figure 3 This is a schematic diagram illustrating the construction of a center-time-based indicator observation window according to the present invention. For example... Figure 3 As shown, the bold vertical dashed line represents the center time of the indicator observation window. The bold vertical line to the left of the center time represents the start time of the forward look-back duration, and the bold vertical line to the right of the center time represents the end time of the backward look-back duration. The blue double arrows represent the forward look-back duration formed by extending forward from the center time, the black double arrows represent the backward observation duration formed by extending backward from the center time, and the red double arrows represent the indicator observation window determined by both the forward look-back duration and the backward observation duration.

[0126] Then, the indicator record corresponding to the name of the indicator to be analyzed is read in the indicator observation window. This step involves selecting the patient identifier. Periodic numbering These are indicator records that match the current text feature record being processed, have the same indicator name as the current indicator being analyzed, and whose collection time falls within the indicator observation window. If more than two indicator records are read, they are first sorted in ascending order of collection time, and the indicator value with the latest collection time is recorded as... The earliest collected index value is recorded as The difference between the two is used as the change in the indicator. If only one indicator record is read, the change in the indicator cannot be calculated because there are no comparable values ​​for the preceding and following indicators. Then the change in the indicator is not calculated. It directly generates the direction of indicator change based on the indicator value recorded by the indicator and the indicator reference range. If no indicator record is read, no indicator change will be generated. and the direction of indicator change The health characteristics of the text are recorded and retained in step S2300. This is the serial number of the record of the indicator change.

[0127] The reference range for the indicators is provided by the standardized management path configuration, including the indicator name, unit, lower reference limit, and upper reference limit. For example, the heart rate reference range can be configured to 60 to 100 beats / min, and the lower reference limit for blood oxygen saturation can be configured to 95%. If the same indicator has different units of measurement, it is first converted to a unified unit according to the unit conversion rules in the standardized management path before being included in the calculation.

[0128] Finally, based on the number of indicator records and the amount of indicator change... And the reference range of the indicator, generating the direction of indicator change. The direction of change of the indicator. The value can be 1, 0, or -1; 1 indicates a tendency to recover, 0 indicates unclear changes or the indicator is within the reference range, and -1 indicates a tendency to be abnormal or the indicator is outside the reference range.

[0129] Specifically, if more than two indicator records are read, the value of the indicator with the latest collection time will be selected. When returning from outside the reference range to within the reference range, the direction of the indicator change will be... The value is 1; it represents the latest value of the index. When the indicator changes from within the reference range to outside the reference range, or continues to move further away from the reference range from outside the reference range, the direction of the indicator change will be... The value is -1; when the indicator remains within its reference range and the change does not exceed 10% of the width of the reference range, the direction of the indicator change will be... The value is 0. The reference range width of the indicator is the difference between the upper and lower reference limits; for indicators with only a lower reference limit, 10% of the lower reference limit is used as the benchmark for judging the magnitude of change.

[0130] If only one indicator record is read, the direction of indicator change will not be recorded because there is no comparison between the preceding and following indicator values. The value is set to 1. At this point, if the recorded value of the indicator is within the indicator's reference range, the direction of the indicator change will be... The value is 0; when the value recorded by this indicator is lower than the lower reference limit or higher than the upper reference limit, it will... The value is -1.

[0131] Generate indicator change characteristics At that time, write the patient identifier. Periodic numbering The message identifier of the text feature record to be processed, the name of the indicator to be analyzed, the indicator observation window, and the indicator change. and the direction of indicator changes .

[0132] For example, a patient sends the text "worsening palpitations" at 8 PM. In step S1300, a time correspondence was established between this text and the heart rate record collected at 8:15 PM. This step uses 8:15 PM as the center time of the indicator observation window and reads the heart rate records from the two hours before to the four hours after. Since the indicator to be analyzed corresponding to "palpitations" is heart rate, this step selects the heart rate records within this observation window. The reading results show that there are multiple heart rate records within the window. Arranged in chronological order of collection time, the earliest heart rate record is 86 beats / min, and the latest heart rate record is 116 beats / min. The heart rate reference range is 60 to 100 beats / min. The latest heart rate record is higher than the upper limit of the reference range, and the heart rate has increased from within the reference range to outside the reference range. Therefore, this step generates the indicator change characteristics corresponding to the health characteristic record of this text. The indicator to be analyzed is heart rate, and the direction of indicator change is... This indicates that the indicator is trending abnormally.

[0133] Step S2300, based on the periodic follow-up data packet The set of path source records in Determine the path execution result, and combine it with text health features. Characteristics of indicator changes Perform correlation, generate standard disease course feature records, and determine disease course time markers. Encapsulate and generate a standard disease course feature table .

[0134] In the specific implementation process, this step first reads the periodic follow-up data packet output by S1300. The set of path source records in One of the path source records is then designated as the current path to be processed. For each current path to be processed, its path node name and planned execution time are parsed. Actual execution time The execution status is also specified. The path node name represents a specific execution item in the standardized management path, including medication node, follow-up visit node, exercise node, and diet record node. The execution status values ​​include completed, completed after timeout, and not completed, and are derived from the path source record registered in step S1100.

[0135] Next, determine the allowed completion time limit based on the path node name. The allowed completion time limit This is used to determine whether a path node is completed within the execution time constraints specified in the standardized management path. Specifically, the allowable completion time limit for the medication administration node. Take 6 hours, the allowable completion time limit corresponding to the follow-up visit. Take 1 day as the allowed completion time limit for each movement node. Take 12 hours as the allowed completion time limit for each food tracking node. Take 12 hours.

[0136] Then, based on the execution status and planned execution time of the current pending path record. Actual execution time and the allowed completion time limit Generate path execution direction .in, Indicates the first Each path node; the path execution direction The value can be 1 or -1. For path nodes with an execution status of "completed", calculate the actual execution time. With planned execution time The time difference between them. If the time difference is less than or equal to the allowed completion time limit corresponding to the path node. Then the path execution direction will be changed. A value of 1 indicates that the path node was completed on time. For path nodes with an execution status of "timeout completed" or "incomplete", the path execution direction will be changed. A value of -1 indicates that the path node was not completed within the specified time. This means the system time at the time this step was executed had not yet reached the planned execution time. For path nodes, it does not determine whether they are completed on time, nor does it generate a path execution direction. The path node will be marked as not having reached its planned execution time; the path node not having reached its planned execution time will not be written into the standard disease course characteristic table for this cycle. .

[0137] Subsequently, the current pending path record and path execution direction will be used. The path execution result and the text health features generated in step S2100 The characteristics of indicator changes generated in step S2200 Perform correlation, generate standard disease course characteristic records, and write them into the standard disease course characteristic table. When the path execution result is written to the standard disease course characteristic record, the path execution direction will be... Write the path execution direction field into the standard disease course feature record.

[0138] The specific execution logic of the association is as follows: based on patient identifiers and period number As a matching criterion, textual health features belonging to the same patient and the same follow-up time interval were selected. Characteristics of indicator changes The execution result of the path.

[0139] For a text-based health feature record, first look up the corresponding indicator change features based on its message identifier. If found, then the change characteristics of that indicator will be recorded. Write the indicator change feature field to the same standard disease course feature record. If not found, write "data missing" to the indicator change feature field of that standard disease course feature record.

[0140] Next, within the same patient and the same follow-up time interval, find the path execution result closest to the text occurrence time of the health feature record. Specifically, calculate the text occurrence time and the planned execution time of the path node. The time interval between them; and determine whether there is an actual execution time. If it exists, continue calculating the text occurrence time and the actual execution time of the path node. The time distance between the two values ​​is used as the path matching time distance; if no actual execution time exists... Then directly compare the time the text occurred with the planned execution time. The time distance between them is determined as the path matching time distance. If the path matching time distance is less than or equal to the upper limit of the path matching time corresponding to the path node name... If no path execution result exists within the same patient and follow-up time interval, or if path execution results exist but their path matching time distances are all greater than the corresponding path matching time limit, then the path execution result will be written to the path execution feature field of the same standard disease course feature record. If no path execution result matching the time condition is found, then it is determined that no such result was found, and the "data missing" field is entered into the path execution characteristic field of the standard disease course characteristic record. The upper limit of the path matching time is... This is used to limit the time correspondence between path nodes and text health features: medication nodes are set to 6 hours, exercise nodes to 12 hours, diet record nodes to 12 hours, and follow-up visit nodes to 3 days.

[0141] If no text health feature record is matched in the execution result of a certain path within the same follow-up time interval, a standard disease course feature record is generated separately based on the execution result of that path, and the missing data is written into the text health feature field and indicator change feature field of the standard disease course feature record.

[0142] For indicator change features generated in step S2200 but not written into the same standard disease course feature record as the text health feature record. If its patient identification and period number , and periodic follow-up data packets If consistent, then the change characteristics of this indicator will be used. Generate a separate standard disease course feature record and write the missing data into the text health feature field and path execution feature field.

[0143] Then, according to the first Generate disease time stamps from the fields already written in the standard disease course feature records. .in, Indicates the first A standard record of disease progression characteristics; the time markers of the disease progression. Used to represent the occurrence time corresponding to the standard disease course characteristic record. Specifically, if the first... The text health feature is written into the text health feature field of the standard disease course feature record. Then, based on the text health characteristics The corresponding text occurrence time serves as a time marker for the course of the disease. If the first If the text health feature field of a standard disease course feature record is missing data, but the path execution feature field contains the path execution result, then the planned execution time of the path node in the path execution result will be used. As a time marker of the course of the disease If the first The standard disease course characteristic record only wrote the indicator change characteristics in the indicator change characteristic field. Then, based on the characteristics of the change of this indicator The corresponding indicator observation window center time As a time marker of disease course .

[0144] Finally, the disease course time was marked. Write to the The time stamp field of the standard disease course characteristic record is added, and the standard disease course characteristic record is written into the standard disease course characteristic table. The standard disease course characteristic table includes... This includes text health feature fields, indicator change feature fields, path execution feature fields, path execution direction fields, disease course time stamp fields, and feature source fields. The feature source field is based on the... The content written to each source field in a standard disease course feature record is determined, and is used to record the data source contained in that standard disease course feature record. Specifically, if the text health feature field is written with text health features... If the feature source field contains the text source, then the indicator change feature field should be written with the indicator change feature. If the feature source field contains the indicator source, then if the path execution feature field contains the path execution result, then the feature source field contains the path source. If the data written to a certain source field is missing, then that source is not written to the feature source field. For text health feature fields, indicator change feature fields, or path execution feature fields with missing data, their missing data status is retained, and other fields already written in the same standard disease course feature record are also retained.

[0145] For example, the planned execution time of a patient's medication administration. The scheduled time for medication was 8:00 AM, but the patient did not complete the dose. At 11:00 AM, the patient sent a text message stating "Dizziness worsened." In the corresponding monitoring window, the patient's blood pressure rose from 135 / 85 mmHg to 162 / 98 mmHg. This discrepancy occurred because the text message occurred at a time when the scheduled medication administration time was different from the planned time. The 3-hour difference does not exceed the 6-hour time limit for path matching corresponding to the medication node. Therefore, the text health feature, blood pressure index change feature, and medication path execution result are written into the same standard disease course feature record. In this standard disease course feature record, the text health feature field is set to "worsened dizziness," the index change feature field is set to "increased blood pressure," and the path execution feature field is set to "medication node not completed." Correspondingly, the text semantic direction is -1, the index change direction is -1, and the path execution direction is -1. This standard disease course feature record indicates that near this medication node, the patient simultaneously experienced worsened symptoms, increased blood pressure, and failure to take medication on time.

[0146] Step S3000, according to the standard disease course characteristic table Extract path node association feature records Record the associated features of path nodes Source direction determination and evidence summary encapsulation are performed to obtain the disease course evidence unit table. .

[0147] Specifically, this step uses the standard disease course characteristic table from S2300. As input, first generate path node association feature records. Then record based on the path node association features Determine the consistent state feature group And encapsulate and generate a disease course evidence unit table. This step is used to associate text health features within the same path node over a given time range. Characteristics of indicator changes The results of the path execution are organized into data units for subsequent view generation, providing input for the subsequent generation of health trend views for doctors and patients.

[0148] Further, step S3000 includes:

[0149] Step S3100, based on the standard disease course characteristic table Determine the time range associated with nodes and extract disease course time markers. Text health features falling within the time range associated with the node Characteristics of indicator changes Based on the path execution results, generate path node association feature records. .

[0150] In the specific implementation process, this step first follows the patient identification... and period number Read the standard disease course characteristic table from step S2300 The standard disease course characteristics are recorded, and all standard disease course characteristics records of the same patient within the same follow-up period are collected as a set of periodic characteristics records.

[0151] Next, the path execution feature fields are read from the periodic feature record set. If a path execution result exists in a standard disease course feature record, the path node name and planned execution time are extracted from that path execution result. Actual execution time Execution status and execution direction And generate path node identifiers. The path node identifiers are derived from the patient identifier. Periodic numbering Path node name and planned execution time Combined generation is used to uniquely identify a path node within the same patient and the same cycle.

[0152] If the set of periodic feature records does not contain a path execution result, but the standardized management path configuration contains an execution plan for the path node that should be executed in that period, then read the path node name and planned execution time of that path node execution plan. Candidate path nodes are generated and marked as nodes with missing path execution records. The path node execution plan refers to the pre-determined path node execution arrangement in the standardized management path configuration for the current patient and current follow-up period, which includes the path node name and planned execution time. This process allows for the aggregation of textual and indicator evidence around planned path nodes even in cases where no check-in or path execution record has been uploaded, preventing the inability to generate pathological evidence for that path node due to missing path records.

[0153] Then, the node association time range is determined based on the path node name. The node association time range is represented by the node's base time. Centered on this, text health features are extracted that are temporally related to the path node. Characteristics of indicator changes And the time range of the execution result of the path. For cases where actual execution time exists... For path nodes, if the execution status is "completed" or "timed out", the actual execution time will be used. As the reference time of the node For path nodes whose execution status is incomplete, or for candidate path nodes where the path execution record is missing, the planned execution time will be used. As the reference time of the node The node reference time Used to uniformly determine the extraction range of text health features, indicator change features, and path execution results.

[0154] Determine the reference time of the node. Then, based on the associated time half-width corresponding to the path node name. The time ranges for node association are determined as follows: 6 hours for medication nodes, 12 hours for exercise nodes, 12 hours for diet record nodes, and 3 days for follow-up visit nodes. These values ​​are consistent with the upper limit of path matching time in step S2300, and are used to limit the association time between path nodes and text health features. Characteristics of indicator changes The effective correlation range between them. Based on the node reference time. and associated time half-width The time range associated with the node is represented as follows: .

[0155] Subsequently, disease course time markers were extracted from the periodic feature record set. One or more text health features falling within the time range associated with the node Characteristics of indicator changes and path execution results. For text health features Read its health standard terms, text category, and text semantic direction. and evidence summary; for the characteristics of indicator changes Read the name of the indicator to be analyzed and the change in the indicator. and the direction of indicator change For the path execution result, read its path node name, execution status, and path execution direction. .

[0156] Finally, using the same path node identifier and its corresponding node association time range as units, the text health features falling within the node association time range are analyzed. Characteristics of indicator changes Based on the path execution results, generate a path node association feature record. Each path node is associated with a feature record. Including patient identification Periodic numbering Path node identifier, path node name, node base time The data includes the node association time range, text health feature set, indicator change feature set, path execution feature set, and feature source field.

[0157] The text health feature set is used to store one or more text health features that fall within the time range associated with the node. The indicator change feature set is used to store one or more indicator change features that fall within the time range associated with the node. The path execution feature set is used to store one or more path execution results that fall within the time range associated with the node. The feature source field is based on the standard disease course feature table in step S2300. Sure.

[0158] For example, a patient has a medication-taking node within a certain cycle, with a planned medication time of 8:00 AM on the current day. The system did not receive a check-in record for this medication-taking node. This medication-taking node is a missing node in the path execution record; therefore, the planned medication time of 8:00 AM is used as the node's base time. The half-width of the associated time corresponding to the medication-taking node is 6 hours, so the node's associated time range is from 2:00 AM to 2:00 PM on the current day. Within this time range, the patient uploaded a blood pressure record at 10:30 AM, and the result was higher than the set range; the patient also filled in a text record of "significant dizziness" at 11:00 AM. This step organizes the medication-taking node, the blood pressure change feature at 10:30 AM, and the text health feature at 11:00 AM into a single path node associated feature record. If there are no exercise records or other text records within this time range, a data missing marker is written to the corresponding field, which does not affect the retention of the extracted blood pressure record and text record.

[0159] Step S3200: Record based on path node association features Generate text source directions respectively Indicator source direction and path source direction It also identifies evidence integrity markers and consistency statuses, and encapsulates and generates consistency status feature groups. .

[0160] In the specific implementation process, this step first reads the associated feature record of each path node generated in step S3100. It reads the text health feature set, indicator change feature set, and path execution feature set, and extracts the text semantic direction respectively. Direction of indicator change and path execution direction .

[0161] Next, the text source direction is generated separately. Indicator source direction and path source direction .

[0162] For the text health feature set, first determine if the set is empty. If the text health feature set is empty, then determine the text source direction. Mark as missing and end the text source direction for this set. Generation Processing. If the text health feature set is not empty, continue to determine the number of text health features in the set. If the text health feature set contains only one text health feature, then read the text semantic direction of that text health feature. and the semantic direction of the text Directly used as the source direction of the text .

[0163] If the text health feature set contains more than two text health features, then the text semantic direction is counted separately. The number of text health features is 1, -1, and 0, and the text source direction is generated based on the statistical results. Specifically, if the semantic direction of the text... The number of text health features with a value of -1 is greater than the number of text semantic orientation features. If the number of text health features is 1, then the text source direction will be... Set to -1; if the text semantic direction The number of text health features with a value of 1 is greater than the number of text semantic orientation features. If the number of text health features is -1, then the text source direction will be... Set to 1; if the text semantic direction If the number of 1s and -1s are equal, then the text source direction will be changed. Set to 0. This means that two or more text health features satisfy the text semantic direction. When the value is 0, since there are no text health features pointing to recovery or anomalies, the text source direction will also be... Set it to 0.

[0164] For a set of indicator change characteristics, first determine if the set is empty. If the set of indicator change characteristics is empty, then determine the direction of indicator source. Mark as missing and end the index source direction for this set. Generation Processing. If the set of indicator change features is not empty, continue to determine the number of indicator change features in the set. If the set of indicator change features contains only one indicator change feature, then read the direction of indicator change for that feature. And the corresponding indicator value. If the direction of change of this indicator... If the value is recorded as -1 and the corresponding indicator value exceeds the reference range of the indicator configured in the standardized management path, then the indicator source direction will be changed. Set to -1; if the direction of change of this indicator If denoted as 1, then the direction of the indicator source will be... Set it to 1; if the direction of change of this indicator If recorded as 0, then the source direction of the indicator will be... Set it to 0.

[0165] If the set of indicator change features contains more than two indicator change features, then the indicator source direction is generated according to the anomaly priority rule. Specifically, first determine whether there is a direction of change in the indicator. This is recorded as -1, and the corresponding indicator value exceeds the reference range of the standardized management path configuration. If such an indicator exists, the source direction of the indicator will be... Set it to -1. If the above-mentioned indicator change characteristics do not exist, then determine whether there is an indicator change direction. The characteristic of the indicator change is denoted as 1; if it exists, the direction of the indicator source is... Set it to 1. If there is no direction of indicator change... Recorded as -1 or The characteristic of indicator change is denoted as 1, and the direction of indicator change for each characteristic of indicator change is... If all values ​​are 0, then the direction of the indicator source will be... The value is set to 0. The anomaly priority rule is used when more than two indicator change characteristics exist within the same node's associated time range, prioritizing the anomaly indicator that exceeds the indicator reference range to determine the source direction of the indicator. This allows the indicator to participate in subsequent consistency status judgments.

[0166] For the path execution feature set, if the path execution feature set is not empty, then read the path execution result that matches the current path node identifier, and set the path execution direction in the path execution result. As the direction of path source If the path execution feature set is empty, but the current path node is associated with feature records... If the corresponding path node is a missing node in the path execution record generated in step S3100, then the path source direction is... A value of -1 indicates that no valid execution record has been formed for the planned path node. If the path execution feature set is empty, and the current path node is not a node with a missing path execution record, then the path source direction will be... Record as missing.

[0167] Then, based on the direction of the text source Indicator source direction and path source direction Generate evidence integrity markers. Specifically, if , and If none of them are missing, then the evidence completeness is marked as complete from three sources; if , and If any two categories are not missing, the evidence integrity is marked as valid from two sources; if only one category is not missing, the evidence integrity is marked as valid from one source; if... , and If all are missing, the evidence integrity is marked as having no valid source, and no consistent state tending towards anomaly or recovery is generated.

[0168] Subsequently, a consistent state is generated. Specifically, after removing the source directions with missing values, if... , and If at least two source directions are -1, and there is no case where the source direction is 1, then the consistency state is consistent in terms of source and tends towards anomaly; if , and If at least two source directions are 1, and there is no case where the source direction is -1, then the consistency state is consistent from the source and tends to recover; if , and If both 1 and -1 exist simultaneously, the consistency state is "inconsistent source"; if only one source direction is not missing, or if all non-missing source directions are 0, the consistency state is "single source, pending supplementation". , and If all are missing, the consistency status is recorded as insufficient evidence.

[0169] To prevent records with a source direction of 0 from affecting anomaly or recovery determination, this step treats sources with a source direction of 0 as neutral sources when generating the consistency state. These neutral sources do not independently trigger source consistency and tend towards an anomaly state, nor do they independently trigger source consistency and tend towards a recovery state; however, neutral sources are retained in the consistency state feature group. This is used in the subsequent step S3300 to generate the disease course evidence unit table to explain the source of evidence.

[0170] Finally, this step records the associated features of the current path node. Text source direction Indicator source direction Path source direction Evidence integrity markers and consistency status are written into the consistency status feature group. .

[0171] For example, a patient sends "I'm not out of breath walking today" within the time frame associated with the node corresponding to the motion node, thereby obtaining the direction of the text source. When blood oxygen saturation is below the reference lower limit of the standardized management pathway configuration within the same time frame, the source direction of the indicator is determined. The motion node has been completed within the allowed time limit, thus obtaining the path source direction. This step determines that both recovery and abnormal directions exist simultaneously among the three source directions, and generates a consistency status of "inconsistent source." Through this processing, the whole-course management system will not only rely on the improvement description or path in the patient's text to generate a recovery judgment, but will also retain abnormal blood oxygenation in the consistency status feature group. This provides the subsequent step S3300 with a table of medical evidence units for doctors to verify.

[0172] Step S3300, based on the consistency state characteristic group Generate view destination markers and extract text evidence summaries, indicator evidence summaries, and path execution summaries. Combine these with the node-associated time ranges to encapsulate and generate a disease course evidence unit table. .

[0173] In the specific implementation process, this step first reads the consistency state feature group output by step S3200. .

[0174] Next, based on the consistency status and evidence integrity markers, a view destination marker is generated. This view destination marker determines the position in subsequent step S4000 where the disease course evidence unit will be generated in the dual-track health trend view; its values ​​include doctor anomaly summary, doctor pending verification, patient recovery progress, and patient record reminders.

[0175] The specific generation logic of the view destination marker is as follows: When the consistency status is consistent from the same source and tends towards anomaly, and the evidence integrity marker is two-source valid or three-source complete, a doctor's anomaly summary marker is generated. This process indicates that at least two types of data sources simultaneously point to anomalies, serving as the abnormal evidence that doctors prioritize viewing.

[0176] If the consistency status is consistent from the source and trending towards anomaly, but the evidence integrity flag is marked as single-source valid, no doctor anomaly summary flag is generated; instead, a doctor pending verification flag is generated. This process is used to prevent single-source anomalies from directly entering the doctor's anomaly summary.

[0177] When the consistency status is "source inconsistency," a doctor-pending verification marker is generated. This process indicates that there are opposing trends between different sources; for example, the text description tends to recover while the indicator change tends to be abnormal, requiring further verification by the doctor.

[0178] When the consistency status is consistent with the source and tends towards recovery, and the evidence integrity marker is two-source valid or three-source complete, a patient recovery progress marker is generated. This process indicates that at least two types of data sources point to recovery, making it suitable for displaying recovery progress on the patient's end.

[0179] When the consistency status is consistent in origin and tending towards recovery, but the evidence integrity flag is single-source valid, a patient record reminder flag is generated. This indicates that there is currently a record of recovery direction, but the source of evidence is insufficient, and the patient needs to be prompted to continue supplementing indicators, text, or path execution records.

[0180] When the consistency status is "single source pending supplementation," a patient record reminder flag is generated. This indicates that there is currently only a single valid source, or that all valid sources are neutral, and no abnormality or recovery judgment can be formed.

[0181] When the consistency status is insufficient evidence, no doctor abnormal summary mark, doctor pending verification mark, or patient recovery progress mark is generated; if the path node originates from the missing path execution record node in step S3100, a patient record reminder mark is generated to prompt the patient to supplement the execution record corresponding to the path node.

[0182] Then, text evidence summaries, indicator evidence summaries, and path execution summaries are generated. For text evidence summaries, if the text health feature set is empty, the text evidence summary is written with missing data; if it is not empty, it is selected first. The text summary, and then the selection The text summary was selected last. The text summary. If multiple summaries exist under the same priority, they are ranked according to the text occurrence time and the node reference time. The evidence is sorted from most recent to oldest, with a maximum of 3 entries retained. The length of each text evidence summary follows the limit of 5 to 60 Chinese characters set in step S2100.

[0183] For indicator evidence summaries, if the indicator change feature set is empty, the indicator evidence summary will be considered as having missing data; if it is not empty, it will be selected first. Furthermore, indicators whose values ​​are below the lower reference limit or above the upper reference limit should be selected next. The indicators were finally selected. The metrics. If multiple metrics exist under the same priority, they are ranked according to the center time of the metric observation window. With node reference time The time intervals are sorted from most recent to oldest, with a maximum of three items retained. The evidence summary for each indicator includes the name of the indicator being analyzed and the direction of its change. and the state of the indicator values; there is a change in the indicator values. At the same time, the change in the indicator is written. .

[0184] For path execution summaries, if the path execution result is read, then the path node name, execution status, and path execution direction are considered. Generate a path execution summary; if no path execution result is read and the current path node is a missing path execution record node generated in step S3100, then write "No execution record formed for the plan node" into the path execution summary; if no path execution result is read and the current path node is not a missing path execution record node, then write "Data missing" into the path execution summary.

[0185] Subsequently, an evidence time range is generated. This evidence time range is taken from the current path node associated feature record in step S3100. Node association time range The aforementioned time range of evidence is used to define the time interval from which each piece of textual evidence, indicator evidence, and path execution evidence in this disease course evidence unit originates.

[0186] Finally, this step will group the consistent state characteristics. View destination markers, text evidence summaries, indicator evidence summaries, path execution summaries, and evidence time ranges are written into the disease course evidence unit table. .

[0187] Specifically, this step aggregates textual evidence, indicator evidence, and path execution evidence along the same path node, generating a unified view destination marker. For records with consistent anomalies from multiple sources, the system categorizes them into a doctor's anomaly summary, making it easier for doctors to view situations supported by simultaneous symptom, indicator, or execution anomalies. For records with single-source anomalies or inconsistent sources, the system categorizes them as requiring doctor verification, reducing false alarms caused by deviations in single records or single types of data. For records with consistent recovery from multiple sources, the system categorizes them into patient recovery progress, ensuring that the recovery content displayed on the patient's end is supported by textual, indicator, or path execution results. For nodes with insufficient evidence or missing execution records, the system categorizes them into patient record reminders, guiding patients to supplement the records. Thus, evidence from different sources can be used together under the same path node, making it easier for doctors to view anomaly verification content and for patients to clearly understand recovery progress and content requiring supplementation.

[0188] Step S4000, based on the disease course evidence unit table Generate a health trend view for doctors. and patient-side health trend view And based on the doctor's health trend view and patient-side health trend view Generate a dual-track health trend view .

[0189] Specifically, this step uses the disease course evidence unit table from S3300. As input, a doctor-side health trend view is generated based on the view destination markers. and patient-side health trend view Subsequently, a health trend view for doctors was analyzed. and patient-side health trend view Perform patient identification, cycle numbering, and source verification, and encapsulate and generate a dual-track health trend view. This step converts the disease course evidence units into page data for both doctors and patients, providing output data for subsequent terminal display.

[0190] Further, step S4000 includes:

[0191] Step S4100, based on the disease course evidence unit table The view destination markers in the data are merged to generate anomaly summary view items and view items to be verified, and then sorted according to the time range of the evidence to generate a doctor-side health trend view. .

[0192] In the specific implementation process, this step first follows the patient identification... and period number Read the disease course evidence unit table from step S3300 Identify the same patient Numbering in the same period The corresponding disease course evidence units are grouped into the same doctor-side cycle view candidate set. .

[0193] Next, scan the candidate set of cycle views on the doctor's end. The view destination markers for each disease course evidence unit are defined. When the view destination marker is a physician's abnormal summary, the corresponding disease course evidence unit is written into the abnormal summary candidate set. When the view destination is marked as a doctor to be checked, it is added to the candidate set to be checked. When the view destination is marked as patient recovery progress or patient record reminder, it will not be written to the exception summary candidate set. and candidate set to be checked Instead, its path node identifiers are written into the trend observation candidate set. The trend observation candidate set. Path nodes used to record disease evidence units generated for the same patient and within the same follow-up period, but which have not entered the abnormal summary area or the area to be verified on the doctor's end.

[0194] Then, the candidate set of anomaly summaries The disease course evidence units are merged according to the path node identifier to generate anomaly summary view items. Each anomaly summary view item includes path node identifier, consistency status, evidence integrity marker, text evidence summary, indicator evidence summary, path execution summary, and evidence time range.

[0195] If the same path node identifier corresponds to one disease course evidence unit, then the above fields of that disease course evidence unit are directly written into the abnormal summary view item. If the same path node identifier corresponds to more than two disease course evidence units, then the text evidence summary, indicator evidence summary, and path execution summary are merged respectively; a maximum of 3 summaries are retained for each type, and the first 3 are selected from the closest to the furthest according to the time distance between the evidence time corresponding to each summary and the baseline time of the path node; the evidence integrity mark is retained in the order of three-source complete, two-source valid, and single-source valid, and the evidence integrity mark that is ranked first is retained; the consistency status is written as consistent in source and tending to be abnormal; the evidence time range is the union of the evidence time ranges of each disease course evidence unit.

[0196] Subsequently, the candidate set for verification was processed. The disease course evidence units are merged according to the path node identifiers to generate view items to be verified. Each view item to be verified includes a path node identifier, consistency status, verification type, text evidence summary, indicator evidence summary, path execution summary, and evidence time range. The verification type is determined based on the consistency status, and its values ​​include inconsistent source verification, single-source evidence supplementation, and missing path execution record verification. Specifically, when the consistency status is inconsistent source, the verification type is recorded as inconsistent source verification; when the consistency status is single-source evidence supplementation, the verification type is recorded as single-source evidence supplementation; when the consistency status is insufficient evidence and the corresponding path node is a missing path execution record node, the verification type is recorded as missing path execution record verification.

[0197] If the candidate set to be checked If the same path node identifier corresponds to only one disease course evidence unit, then the above fields of that disease course evidence unit are written into the view item to be verified. If the same path node identifier corresponds to more than two disease course evidence units, then the text evidence summary, indicator evidence summary, and path execution summary of each disease course evidence unit are merged; if the consistency status is the same, the record with the evidence integrity mark sorted first and the evidence time range closer to the path node's base time is retained; if the consistency status is different, the corresponding verification type and summary content are retained respectively in the same view item to be verified.

[0198] Finally, the abnormal summary view items and the view items to be verified are sorted according to the start time of the evidence time range from earliest to latest, generating a health trend view for doctors. The doctor-side health trend view It includes an anomaly summary area, a trend observation area, and a check-in area. The anomaly summary area is used to write anomaly summary view items, the check-in area is used to write view items to be checked, and the trend observation area is used to write the trend observation candidate set. The path node information and the corresponding time range of the evidence.

[0199] If the same patient is numbered in the same cycle If there are no disease evidence units marked as doctor's abnormal summary or doctor's pending verification in the view destination, then the abnormal summary area and the pending verification area are empty, and it is recorded that no doctor's abnormal summary view item and pending verification view item were generated in this cycle.

[0200] Specifically, this step is based on the disease course evidence unit table. The view navigation markers in the system write abnormal evidence into the abnormal summary area on the doctor's end, and display content with inconsistent sources or insufficient evidence in the verification area, while simultaneously showing the corresponding text, indicators, and path execution information. Therefore, when reviewing the same follow-up period, doctors can directly see abnormal situations and content to be verified, reducing the need to search and compare multiple records. For example, if "dizziness," "elevated blood pressure," and "medication not completed" occur simultaneously near a medication administration point, it is displayed as an abnormal summary; if the patient describes improvement but the indicators remain abnormal, it is displayed as content to be verified for further doctor assessment.

[0201] Step S4200, based on the doctor's health trend view Generate patient-side constraint index Based on patient-side constraint index and disease course evidence unit table Determine the patient-side deterministic trend output label Combine the view destination markers to generate patient-side view items and generate a patient-side health trend view. .

[0202] In the specific implementation process, this step first extracts the doctor's health trend view output in step S4100. In the process, all view items to be verified in the verification area are read, and the path node identifiers in each view item are extracted to generate a patient-side constraint index. The path node identifier is composed of the patient identifier. Periodic numbering Path node name and planned execution time Generate by combination. Therefore, the patient-side constraint index. This is used to limit the disease course evidence units under the same patient, same follow-up period, and same path node corresponding to the view item to be verified in the doctor's end, so that it does not generate deterministic trend statements such as recovery, deterioration, or normalization on the patient's end. The patient-end constraint index It includes at least one constraint index record, each constraint index record corresponds to a view item to be verified in the doctor's end, and writes the path node identifier of the view item to be verified in the doctor's end.

[0203] Next, the disease course evidence unit table output from step S3300 In the process, the view destination is read from the disease progress evidence units marked as patient recovery progress or patient record reminders to generate a patient-side candidate set. For the patient-side candidate set For each piece of evidence in the disease progression, its path node identifier is read and compared with the patient-side constraint index. Perform a match.

[0204] In the candidate set of patients When performing constraint judgments on disease course evidence units, the path node identifier of the disease course evidence unit is compared with the patient-side constraint index. The path node identifiers of each constraint index record are compared. When the path node identifiers of the two are completely identical, it means that the disease course evidence unit and the view item to be verified in the doctor's end belong to the same patient, the same follow-up period, and the same path node, and it is determined that the disease course evidence unit is subject to the patient-side constraint index. Constraints.

[0205] If the path node identifier of this disease course evidence unit matches the patient-side constraint index If the path node identifiers of any constraint index record are inconsistent, then it is determined that the disease course evidence unit is not subject to the patient-side constraint index. Constraints. For example, although two records correspond to the same patient identifier. However, its periodic number The differences may lie in the path node name or the planned execution time. If they are different, the corresponding path node identifiers are different. Therefore, the constraint rules for the view items to be verified in the doctor's end are not applied to this disease course evidence unit, in order to avoid the view items to be verified in the doctor's end of other follow-up periods or other path nodes affecting the generation of the current patient's view. The constraint rules include restrictions that the patient end does not generate deterministic trend descriptions such as recovery, deterioration, or normalization.

[0206] To indicate whether the patient is allowed to output definitive trend statements such as recovery, deterioration, or normality, this step generates patient-side definitive trend output markers. The value selection rules are as follows: Extract the path node identifier of the disease course evidence unit and match it with the patient-side constraint index. The path node identifiers for writing each constraint index record are compared.

[0207] If there exists a constraint index record whose path node identifier is exactly the same as the path node identifier of the disease course evidence unit, then it is determined that the disease course evidence unit is subject to the patient-side constraint index. Constraints, and label the patient-side deterministic trend output. Set to 0. Where, This indicates that the disease course evidence unit does not output definitive trend statements such as recovery, deterioration, or normalization on the patient's end.

[0208] If the patient-side constraint index If no constraint index record exists that matches the path node identifier of this disease course evidence unit, then it is determined that this disease course evidence unit is not subject to patient-side constraint indexes. Constraints, and label the patient-side deterministic trend output. Set to 1. Where, This indicates that the disease course evidence unit can generate patient-side view content according to its view destination markers.

[0209] Then, a marker is output based on the patient-side deterministic trend. The view destination markers are used to generate patient-side view items. These patient-side view items include three categories: recovery progress view items, record reminder view items, and neutral record prompt view items.

[0210] When the patient-side deterministic trend output is marked When the value is 1, and the view destination is marked as the patient's recovery progress, a recovery progress view item is generated. This recovery progress view item is used to display the patient's recovery-related records for that path node, including the path node name, text evidence summary, indicator evidence summary, path execution summary, and evidence time range. This view item does not display the verification instructions used by the doctor.

[0211] When the patient-side deterministic trend output is marked When the value is 1, and the view destination is marked as a patient record reminder, a record reminder view item is generated. This record reminder view item is used to prompt for the completion of missing records under this path node. The missing records are determined based on the evidence integrity marker in the disease course evidence unit, including text records, indicator records, or path execution records.

[0212] When the patient-side deterministic trend output is marked When the value is 0, regardless of the view destination marker's value, no patient-side view item containing a deterministic trend description is generated; instead, a neutral record prompt view item is generated. This neutral record prompt view item indicates to the patient that the relevant record for that path node is still under verification, including the path node name, evidence time range, and prompt content. The prompt content is: "This record is being verified this period; please continue recording the corresponding path node." This prompt content does not contain deterministic trend descriptions such as "recovery," "deterioration," or "normal."

[0213] Finally, the recovery progress view items, record reminder view items, and neutral record prompt view items are arranged from earliest to latest according to the start time of the evidence time range corresponding to each view item, to generate a patient-side health trend view. If no recovery progress view item, record reminder view item, or neutral record prompt view item is generated in this cycle, then the patient's health trend view will be displayed. If left blank, no specific disease progression information will be displayed.

[0214] Specifically, in practical use, when a certain disease course evidence unit has inconsistent sources, insufficient evidence, or incomplete records, this step first sends it to the doctor for verification, while the patient only displays a neutral record prompt. For example, if a patient says, "I'm not out of breath walking today," but their blood oxygen saturation is low in the same period, the doctor displays information pending verification, while the patient displays, "This record for this period is being verified; please continue recording the corresponding path node." This allows the doctor to retain a verification entry point while preventing the patient from making definitive statements.

[0215] Step S4300: View the health trend on the doctor's end. and patient-side health trend view Perform consistency and source checks, generate page description data, and determine the dual-track view based on dual-track constraints. Figure 1 Consistent markers are used to encapsulate and generate a dual-track health trend view. .

[0216] In the specific implementation process, this step first reads the health trend view from the doctor's end. Patient identification in and period number And read the patient's health trend view. Patient identification in and period number .

[0217] When the patient identifiers on the doctor's end and the patient's end are... and period number If they match, then determine the doctor's health trend view. and patient-side health trend view View data belonging to the same patient within the same follow-up period, and the view composited state. Mark as pending synthesis and continue with subsequent processing.

[0218] When the patient identifiers on the doctor's end and the patient's end are... or period number If there is inconsistency, the view composition state will be changed. This is considered a synthetic anomaly; stop generating the dual-track health trend view. And generate view synthesis verification records.

[0219] The view synthesis and verification record is used to record the health trend view on the doctor's end. and patient-side health trend view Whether they belong to the same patient and the same cycle, the record fields include: doctor's patient ID, patient's patient ID, doctor's cycle number, patient's cycle number, verification result, verification inconsistency field, and verification time.

[0220] The verification result is either "pass" or "fail". The verification result is "pass" when the doctor's patient identifier matches the patient's patient identifier, and the doctor's cycle number matches the patient's cycle number; the verification result is "fail" when the doctor's patient identifier does not match the patient's patient identifier, or the doctor's cycle number does not match the patient's cycle number.

[0221] The inconsistency field is used to record the name of the field that caused the verification to fail. When only the patient identifier is inconsistent, the inconsistency field is recorded as "Patient Identifier"; when only the cycle number is inconsistent, the inconsistency field is recorded as "Cycle Number"; when both the patient identifier and cycle number are inconsistent, the inconsistency field is recorded as "Patient Identifier and Cycle Number"; when the verification result is successful, the inconsistency field is empty. The verification time is the system time when the view synthesis verification record is generated, accurate to the minute.

[0222] When the verification result is unsuccessful, the view synthesis verification record is saved as the verification output of this step for subsequent review of the doctor's health trend view. and patient-side health trend view Match?; At this point, the dual-track health trend view will terminate. The generation process is described, and the view synthesis verification record is output.

[0223] Next, based on the disease course evidence unit table output in step S3300... Health trend view for doctors and patient-side health trend view Source verification is performed. This source verification confirms whether the view items on both the doctor's and patient's ends originate from the disease progress evidence unit table. Specifically: Extract the health trend view from the doctor's end. and patient-side health trend view The path node identifier, and in the disease course evidence unit table Search for whether there are disease evidence units with the same path node identifier.

[0224] At the start of source validation, a set of view items that failed source validation is created. And initialize it as an empty set. If the path node identifier of a certain view item cannot be found in the disease course evidence unit table. If a corresponding record is found, the view item is added to the set of view items that failed source validation. .

[0225] The collection of view items that failed source verification The record fields include: view terminal type, partition where the view item is located, path node identifier, and reason for failure. The view terminal type is either doctor's terminal or patient's terminal. The partition where the view item is located records the display area of ​​the view item in the corresponding terminal view, and its value is determined by the view terminal type: when the view terminal type is doctor's terminal, the partition where the view item is located is an anomaly summary area, trend observation area, or area to be verified; when the view terminal type is patient's terminal, the partition where the view item is located is a recovery progress area, record reminder area, or neutral record prompt area. The reason for failure is a missing disease progress evidence unit.

[0226] If the source verification fails, the view item collection If not empty, then the view will be composited. This is recorded as a synthesis anomaly; the set of view items that failed source validation is retained. And stop generating dual-track health trend views. If the source validation fails, the view item collection... If empty, then the view source validation status is... The source verification is considered successful, and the terminal page description data continues to be generated.

[0227] Then, based on the doctor's health trend view Generate doctor workstation page description data. This generation process converts the view items generated in step S4100 into page fields recognizable by the doctor workstation interface. Specifically, it reads the doctor's health trend view. The data is divided into an anomaly summary area, a trend observation area, and a pending verification area, and is written into the anomaly summary display field, trend observation display field, and pending verification display field in the description data of the doctor's workstation page, respectively. The anomaly summary display field corresponds to the doctor's anomaly summary view item; the trend observation display field corresponds to the changes in path nodes arranged according to the evidence time range within the same period; and the pending verification display field corresponds to the doctor's pending verification view item.

[0228] Then, based on the patient-side health trend view Generate patient mobile terminal page description data. This generation process converts the view items generated in step S4200 into page fields recognizable by the patient mobile terminal interface. Specifically, it reads the patient terminal health trend view. The rehabilitation progress view item, record reminder view item, and neutral record prompt view item are generated and written into the rehabilitation progress display field, record reminder display field, and neutral record prompt field in the patient's mobile terminal page description data, respectively. Specifically, the rehabilitation progress display field corresponds to the view item whose view destination is marked as the patient's rehabilitation progress; the record reminder display field corresponds to the view item whose view destination is marked as the patient's record reminder; and the neutral record prompt field corresponds to the neutral record prompt view item formed after hitting the patient's terminal constraint index.

[0229] Subsequently, a dual-track constraint review process is performed. This process checks whether the view items to be verified on the doctor's end have already generated neutral record prompts on the patient's end, in order to reduce the situation where the same path node prompts for verification on the doctor's end but outputs a definitive recovery trend on the patient's end.

[0230] Specifically, using the doctor's health trend view The view item to be checked is the inspection object. Extract its path node identifier and query the patient-side constraint index generated in step S4200. If the path node identifier exists in the patient-side constraint index. In the middle, and the patient's health trend view If a corresponding neutral record prompt view item exists, then the dual-track view item to be verified will be... Figure 1 Consistency markers are recorded as constrained. If the path node identifier does not exist in the patient-side constraint index... In the middle, or the patient's health trend view If no corresponding neutral record prompt view item exists, then the dual-track view item to be checked will be... Figure 1 The consistency marker is marked as pending review.

[0231] Among them, the dual-track vision Figure 1 The values ​​for the consistency flag only include "constrained" and "pending review". "Constrained" means that the patient has used neutral recording prompts for the same path node and has not shown the patient a definite recovery trend; "pending review" means that the view item to be reviewed on the doctor's side has not yet formed a corresponding constraint display on the patient's side and needs to be reviewed later.

[0232] Finally, the doctor's workstation page description data, the patient's mobile page description data, and the dual-track view were combined. Figure 1 Pathogenic markers, patient identification Periodic numbering Together with the view generation time, a dual-track health trend view is generated. The view generation time is the system time at the time this step is completed and encapsulated, accurate to the minute.

[0233] If the patient identifier and cycle number verifications pass, the source verification passes, and the page description data generation is complete, then the view composition status will be changed. The synthesis is now complete, and a dual-track health trend view is output through both the doctor's workstation interface and the patient's mobile terminal interface. .

[0234] If the view is in composite state If a synthesis error occurs, the interface will only output the corresponding error record; the error record includes view synthesis verification records or a set of view items that failed source verification. It does not output description data for the doctor's workstation page or the patient's mobile page.

[0235] Example 2:

[0236] This embodiment, based on Embodiment 1, provides a health trend view generation system that integrates multi-source heterogeneous feature mining, such as... Figure 4 As shown, the system includes a follow-up encapsulation module, a feature fusion module, an evidence encapsulation module, and a view generation module;

[0237] The follow-up encapsulation module is used to acquire raw follow-up data. For the original follow-up data The process sequentially executes raw record parsing, period division, and establishment of time correspondence, and encapsulates the data to generate periodic follow-up data packets. .

[0238] The feature fusion module is used to perform fusion based on periodic follow-up data packets. Extracting health features from text Characteristics of indicator changes The path execution results will include text health characteristics. Characteristics of indicator changes The results of path execution are integrated to generate standard disease course feature records to construct a standard disease course feature table. .

[0239] The evidence encapsulation module is used to encapsulate evidence based on a standard disease course characteristic table. Extract path node association feature records Record the associated features of path nodes Source direction determination and evidence summary encapsulation are performed to obtain the disease course evidence unit table. .

[0240] The view generation module is used to generate a view based on the disease course evidence unit table. Generate a health trend view for doctors. and patient-side health trend view And based on the doctor's health trend view and patient-side health trend view Generate a dual-track health trend view .

[0241] The parts of the technical solutions provided in the embodiments of this application that are consistent with the implementation principles of corresponding technical solutions in the prior art have not been described in detail to avoid excessive elaboration.

[0242] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for generating health trend views by integrating multi-source heterogeneous feature mining, characterized in that, include: Obtain the raw follow-up data, and sequentially perform raw record parsing, period division, and time correspondence establishment on the raw follow-up data, and encapsulate it to generate periodic follow-up data packets; Based on the periodic follow-up data package, extract text health features, indicator change features and path execution results, integrate the text health features, indicator change features and path execution results to generate standard disease course feature records to construct a standard disease course feature table; Based on the standard disease course feature table, path node associated feature records are extracted, and the source direction of the path node associated feature records is determined and evidence summary is encapsulated to obtain the disease course evidence unit table. Generate a doctor-side health trend view and a patient-side health trend view based on the disease course evidence unit table, and generate a dual-track health trend view based on the doctor-side health trend view and the patient-side health trend view. The process involves extracting path node-related feature records from a standard disease progression feature table, determining the source direction of these records, and encapsulating them with evidence summaries to obtain a disease progression evidence unit table, including: Based on the standard disease course feature table, the node association time range is determined, and the text health features, indicator change features and path execution results of disease course time markers falling within the node association time range are extracted to generate path node association feature records. Based on the path node association feature records, text source direction, indicator source direction and path source direction are generated respectively, and evidence integrity markers and consistency status are determined, and consistency status feature groups are encapsulated and generated. Based on the consistency status feature group, view destination markers are generated, and text evidence summaries, indicator evidence summaries, and path execution summaries are extracted. Combined with the node association time range, a disease course evidence unit table is generated.

2. The method for generating a health trend view by fusing multi-source heterogeneous feature mining according to claim 1, characterized in that, The process of acquiring raw follow-up data, sequentially performing raw record parsing, period division, and time correspondence establishment on the raw follow-up data, and encapsulating it to generate a periodic follow-up data packet includes: The patient identifier, collection time, source type, and original content of each original record in the original follow-up data are analyzed and combined to generate source registration data; The period number of each source registration record in the source registration data is calculated based on the initial filing time and the follow-up period step, and the source registration data is classified according to the period number and source type to generate period-aligned data; Determine the minimum time distance between the record to be matched and the target indicator record in the periodic alignment data. Based on the minimum time distance and the preset corresponding time upper limit, establish the time correspondence between the record to be matched and the target indicator record, and encapsulate and generate a periodic follow-up data package.

3. The method for generating a health trend view by fusing multi-source heterogeneous feature mining according to claim 2, characterized in that, The minimum time distance between the record to be matched and the target indicator record in the determined periodic alignment data is used to establish a time correspondence between the record to be matched and the target indicator record based on the minimum time distance and a preset corresponding time upper limit, and then encapsulates and generates a periodic follow-up data package, including: Within the follow-up time interval corresponding to the same patient and the same cycle number, the text source record or the path source record is identified as the record to be matched; Within the follow-up time interval, the time distance between the collection time of each indicator source record and the collection time of the record to be matched is calculated, and the indicator source record with the smallest time distance is selected as the target indicator record. When the minimum time distance between the record to be matched and the target indicator record is less than or equal to the preset corresponding time limit, a time correspondence relationship between the record to be matched and the target indicator record is established; The text source records, indicator source records, path source records, and established time correspondences are encapsulated into periodic follow-up data packages.

4. The method for generating a health trend view by fusing multi-source heterogeneous feature mining according to claim 1, characterized in that, Based on the periodic follow-up data package, text health features, indicator change features, and path execution results are extracted. These features are then fused to generate standard disease course feature records to construct a standard disease course feature table, including: The periodic follow-up data package includes a set of text source records, a set of indicator source records, and a set of path source records; Normalization and semantic segmentation are performed on the set of text source records in the periodic follow-up data package to obtain semantic segments; health standard terms in the semantic segments are identified, and the semantic direction of the text is determined by combining semantic modifiers, and the text health features are encapsulated and generated. Based on the health features of the text, the indicator observation window is determined in the periodic follow-up data package. The indicator records matched in the indicator source record set are read within the indicator observation window. The indicator change amount and direction are calculated based on the indicator records, and the indicator change features are encapsulated and generated. The path execution result is determined based on the path source record set in the periodic follow-up data package. It is then associated with text health features and indicator change features to generate standard disease course feature records and determine disease course time markers. Finally, a standard disease course feature table is generated.

5. The method for generating a health trend view by fusing multi-source heterogeneous feature mining according to claim 4, characterized in that, The text source record set in the periodic follow-up data package is normalized and segmented into semantic segments to obtain semantic segments; health standard terms in the semantic segments are identified, and the semantic direction of the text is determined by combining semantic modifiers, and text health features are encapsulated and generated, including: The text content in the text source record set is standardized. The standardization process includes removing consecutive repeated characters, unifying full-width and half-width characters, converting synonyms into health standard terms, and retaining semantic modifiers related to health status judgment. The synonym conversion involves converting non-standard health expressions in patient texts into corresponding standard health terms according to a medical terminology mapping table for diseases corresponding to standardized management paths. Based on punctuation marks, line breaks, and semantic pause words pre-configured in the standardized management path, the normalized text content is divided into semantic segments to obtain at least one semantic segment. The semantic fragments are matched with a medical thesaurus and a thesaurus to determine the health keywords corresponding to the health standard terms; an evidence summary is generated based on the semantic fragments containing the health keywords. Based on the term type of the health standard terms corresponding to the health keywords, the text category is determined, and the semantic direction of the text is determined by combining the semantic modifiers within the preset word distance before and after the health keywords; The health standard terms, evidence summaries, text categories, and text semantic directions are written into the text health feature record, and the text health feature record is written into the text health feature.

6. The method for generating a health trend view by fusing multi-source heterogeneous feature mining according to claim 1, characterized in that, The method based on path node association feature records generates text source direction, indicator source direction, and path source direction respectively, and determines evidence integrity markers and consistency status, including: Read the path node identifier, text health feature set, indicator change feature set, and path execution feature set from the path node association feature record; When the text health feature set is not empty, the text source direction is generated based on the text semantic direction in the text health feature set; when the text health feature set is empty, the text source direction is recorded as missing. When the indicator change feature set is not empty, the indicator source direction is generated based on the indicator change direction in the indicator change feature set and whether the indicator value exceeds the indicator reference range; when the indicator change feature set is empty, the indicator source direction is recorded as missing. When the path execution feature set is not empty, read the path execution direction from the path execution result that matches the path node identifier, and use it as the path source direction. When the path execution feature set is empty and the path node is not a missing node in the path execution record, record the path source direction as missing. Based on the number of non-missing source directions among the text source direction, indicator source direction, and path source direction, an evidence integrity marker is generated, and a consistency state is generated based on the direction value relationship between the non-missing source directions.

7. The method for generating a health trend view by fusing multi-source heterogeneous feature mining according to claim 1, characterized in that, The process of generating a doctor-side health trend view and a patient-side health trend view based on the disease course evidence unit table, and generating a dual-track health trend view based on the doctor-side health trend view and the patient-side health trend view, includes: The view destination markers in the disease course evidence unit table are merged to generate abnormal summary view items and view items to be verified. The view items are then sorted according to the time range of the evidence to generate a doctor-side health trend view. Based on the doctor's health trend view, generate the patient-side constraint index, determine the patient-side deterministic trend output marker based on the patient-side constraint index and the disease course evidence unit table, combine the view destination marker to generate the patient-side view item, and generate the patient-side health trend view. Perform consistency and source checks on the doctor-side and patient-side health trend views, generate page description data, and determine the consistency markers of the dual-track views based on the dual-track constraint back lookup, and encapsulate and generate dual-track health trend views.

8. The method for generating a health trend view by fusing multi-source heterogeneous feature mining according to claim 7, characterized in that, The process of generating a patient-side constraint index based on the doctor-side health trend view, determining patient-side deterministic trend output markers based on the patient-side constraint index and the disease course evidence unit table, generating patient-side view items by combining the view destination markers, and generating a patient-side health trend view includes: Read the view items to be verified in the doctor's health trend view, extract the path node identifiers in the view items to be verified, and generate the patient's constraint index; Read the progress evidence units from the progress evidence unit table that are marked as patient recovery progress or patient record reminders, and generate a patient-side candidate set. The path node identifiers of the disease course evidence units in the patient-side candidate set are compared with the path node identifiers in the patient-side constraint index to determine the patient-side deterministic trend output marker. When the patient-side deterministic trend output flag is set to 1, a rehabilitation progress view item or a record reminder view item is generated based on the view destination flag. When the patient-side deterministic trend output flag is set to 0, a neutral record prompt view item is generated; The recovery progress view items, record reminder view items, and neutral record prompt view items are sorted according to the start time of the evidence time range corresponding to each view item to generate a patient-side health trend view.

9. The method for generating a health trend view by fusing multi-source heterogeneous feature mining according to claim 7, characterized in that, The process involves performing consistency and source checks on the doctor-side and patient-side health trend views, generating page description data, determining consistency markers for the dual-track views based on dual-track constraints, and encapsulating and generating a dual-track health trend view, including: Perform consistency checks on patient identifiers and cycle numbers in the health trend views on the doctor's end and the health trend views on the patient's end; After the consistency check passes, the path node identifiers in the doctor's health trend view and the patient's health trend view are extracted respectively, and the disease evidence unit with the same path node identifier is searched in the disease evidence unit table to perform source verification. After the source verification is passed, the doctor's workstation page description data is generated based on the doctor's health trend view, and the patient's mobile page description data is generated based on the patient's health trend view. Using the view items to be checked in the doctor's health trend view as the inspection objects, extract their path node identifiers and query the patient's constraint index to determine the dual-track view consistency markers. The doctor's workstation page description data, the patient's mobile page description data, the dual-track view consistency marker, the patient identifier, and the cycle number are encapsulated together to generate a dual-track health trend view.

10. A health trend view generation system integrating multi-source heterogeneous feature mining, used to implement the health trend view generation method integrating multi-source heterogeneous feature mining as described in any one of claims 1-9, characterized in that, The system includes a follow-up encapsulation module, a feature fusion module, an evidence encapsulation module, and a view generation module; The follow-up encapsulation module is used to acquire the original follow-up data, perform original record parsing, period division, and time correspondence establishment on the original follow-up data in sequence, and encapsulate it to generate a periodic follow-up data packet. The feature fusion module is used to extract text health features, indicator change features and path execution results based on the periodic follow-up data package, and fuse the text health features, indicator change features and path execution results to generate standard disease course feature records to construct a standard disease course feature table. The evidence encapsulation module is used to extract path node associated feature records based on the standard disease course feature table, determine the source direction of the path node associated feature records and encapsulate the evidence summary to obtain a disease course evidence unit table. The view generation module is used to generate a doctor-side health trend view and a patient-side health trend view based on the disease course evidence unit table, and to generate a dual-track health trend view based on the doctor-side health trend view and the patient-side health trend view.

Images