A health grading early warning information processing method and system

Abstract

The application discloses a health grading early warning information processing method and system, belongs to the technical field of intelligent health monitoring and health early warning information processing, and aims at solving the problems that fixed threshold and single sign judgment cannot adapt to individual differences, composite abnormality recognition is insufficient, and alarm distribution and privacy control are not accurate. Through multi-source sign data preprocessing, scene-specific long-term and short-term baseline modeling, abnormal deviation calculation, multi-sign correlation fusion, hierarchical routing and grading desensitization, the application realizes the technical effects of accurate identification of health abnormal information, grading early warning, hierarchical treatment and privacy protection.

Description

Technical Field

[0001] This invention relates to the field of intelligent health monitoring and health early warning information processing, and in particular to a method and system for processing health graded early warning information. Background Technology

[0002] With the development of wearable devices, community-based elderly care platforms, and chronic disease management systems, vital sign data such as heart rate, blood pressure, blood oxygen saturation, body temperature, and blood glucose can be continuously collected and used for health risk alerts. Existing health monitoring and early warning systems typically compare the collected vital sign data with uniform fixed thresholds, or make independent judgments on individual vital signs, and then push the trigger results to guardians, community doctors, or platform administrators.

[0003] The above methods are ill-suited to individual differences among monitored subjects in terms of baseline vital signs, circadian rhythms, activity levels, sleep patterns, postprandial states, and chronic disease status. When the uniform threshold is set too strictly, physiological fluctuations caused by activity, postprandial levels, or individual baseline differences are easily misjudged as abnormal; conversely, when the threshold is set too leniently, risk changes that have already deviated significantly from an individual's baseline may be missed. Furthermore, complex abnormalities where multiple vital signs do not reach individual warning thresholds but the overall combination presents a risk profile are difficult to identify promptly using methods that assess individual signs independently.

[0004] In multi-party collaborative scenarios, existing systems suffer from insufficient granularity in alarm distribution. When alarms are pushed to a single recipient, the processing chain is prone to breakage; indiscriminately pushing alarms to multiple recipients can easily lead to information overload, duplicate processing, or oversight by management. Alarm content contains privacy fields such as name, contact information, address, location, and physical condition details. Without a hierarchical anonymization mechanism based on roles, jurisdiction, and authorization relationships, privacy leaks and compliance risks are increased.

[0005] Therefore, there is a need for a health classification early warning information processing method and system that can overcome the shortcomings of the existing technologies. Summary of the Invention

[0006] One objective of this invention is to propose a health grading early warning information processing method and system. Addressing the problems in existing technologies such as fixed thresholds failing to adapt to individual differences, single-sign judgment struggling to identify complex anomalies, alarm distribution failing to balance severity and recipient responsibilities, and lack of tiered control over privacy fields, this invention proposes a technical solution combining standardized data collection, scenario-specific personalized long and short-term baselines, anomaly deviation calculation, multi-sign association fusion, routing matrix distribution, and access control with desensitization. This invention achieves the technical effects of reducing scenario-based false alarms, improving the ability to identify progressive and complex anomalies, enhancing alarm handling matching accuracy, and reducing the risk of privacy leaks.

[0007] This invention provides a method for processing health grading early warning information, characterized by comprising: S1, acquiring vital sign data and auxiliary data of the monitored subject and generating standardized monitoring records; S2, determining physiological or behavioral scenarios based on the standardized monitoring records, and establishing personalized baseline parameters including personalized long-term baseline, personalized short-term baseline, and personalized standard deviation of fluctuation; S3, calculating personalized abnormal deviation degree based on the degree of deviation of measured vital sign values ​​from the personalized baseline parameters, and controlling the updating of personalized baseline parameters according to the early warning status and abnormal markers; S4, performing nonlinear analysis on the personalized abnormal deviation degree. The system performs gender transformation and weighted fusion to obtain a comprehensive health score. This comprehensive health score is then mapped to a normal status identifier or a multi-level warning level identifier. A health warning record is generated using any multi-level warning level identifier as the alarm trigger level. S5. When an alarm is triggered, the alarm distribution object is determined in the routing matrix based on the warning level identifier, the receiver role level divided by handling level and function, and the role's applicable scope or authorization matching rules. S6. The alarm content is hierarchically desensitized and authorized according to the permission level of the alarm distribution object. The corresponding alarm information is then pushed to the alarm distribution object that has passed the authorization verification.

[0008] Optionally, S1 includes:

[0009] Collect vital signs fields such as heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature and blood glucose, as well as auxiliary fields such as activity level, wearing status, post-meal event markers, sleep period markers, location administrative division code, sampling time, device identifier and data quality identifier;

[0010] The edge agent converts data packets from different device protocols into a unified field structure and establishes a binding relationship based on the monitored object identifier and the device identifier;

[0011] The transformed data is processed by aligning sampling time, unifying units, marking missing fields, filtering outliers, and smoothing with a sliding window to obtain the standardized monitoring records containing field source and data quality identifiers.

[0012] Optionally, S2 includes:

[0013] The standardized monitoring records are divided into resting scenario, activity scenario, sleep scenario, and post-meal scenario based on activity level, wearing status, sampling time period, and post-meal event markers.

[0014] Within each scenario, the personalized long-term baseline and personalized standard deviation of fluctuation are calculated using historical normal state records, and the personalized short-term baseline is calculated using records that were not marked as abnormal within the most recent continuous time window.

[0015] When the number of historical normal state records for a certain scenario does not reach the preset sample size threshold, the adjacent scenario baselines and the preset population reference baselines for the same monitored object are fused according to the sample size weights to form the initial personalized long-term baseline for that scenario.

[0016] Optionally, S3 includes:

[0017] For vital signs with bilateral abnormal directions, the long-term deviation component is obtained by dividing the absolute difference between the measured value and the personalized long-term baseline by the upper limit element in the binary set composed of the personalized standard deviation of fluctuation and the preset lower limit protection value.

[0018] For vital signs with a unilateral abnormal direction, the effective difference between the measured value and the personalized long-term baseline is selected according to the preset abnormal direction, and the effective difference is divided by the upper limit element in the binary set composed of the personalized fluctuation standard deviation and the preset lower limit protection value to obtain the long-term deviation component.

[0019] The difference between the personalized short-term baseline and the personalized long-term baseline is calculated as the short-term drift component using the same anomaly direction and the same divisor.

[0020] The personalized abnormal deviation degree is obtained by fusing the long-term deviation component and the short-term drift component according to the preset deviation weight;

[0021] Furthermore, controlling the update weights of the personalized long-term baseline and the personalized short-term baseline includes: when the previous frozen calculation cycle is in a normal state or an alarm-free state, the quality identifiers of each participating field are valid, the wearing status is valid, and no vital signs trigger an abnormality marker, the current standardized monitoring record is added to the baseline update sample according to the update weight corresponding to the data quality identifier.

[0022] When any vital sign has triggered an abnormality marker, the previous frozen calculation period is in an alarm triggered state, or its warning level identifier is a level 2 or higher warning level identifier, or the wearing status is invalid wearing, the weight of the current standardized monitoring record for baseline update is reset to the weight value corresponding to the warning level identifier in the abnormal update weight table, or the current standardized monitoring record is suspended from participating in baseline update;

[0023] Records suspended from participating in baseline updates are retained in the abnormal event cache for subsequent review, but not for the personalized baseline calculation of the current version; the comprehensive health score and warning level identifier generated in the current period are only written to the baseline update cache of the next version and do not participate in the deviation calculation of the current period.

[0024] Optionally, S4 includes:

[0025] A monotonically nonlinear mapping is performed on the personalized abnormal deviation of each vital sign to obtain a sub-item abnormal score whose value is within a preset scoring range and whose increasing value indicates an increasing degree of abnormality. Based on the sub-item abnormal score and the corresponding abnormal direction, the association rule table is read to determine the association rule pre-trigger identifier and its corresponding associated risk field. Based on the chronic disease label of the monitored object, the chronic disease correction parameters for the corresponding vital signs are read from the chronic disease adjustment table. A dimensionless activity correction parameter is generated based on the deviation of activity volume from the activity volume baseline and the lower limit protection value of the activity volume in the scenario. The scenario activity volume baseline is obtained from the activity volume statistics in historical normal state records in step S2. Both the chronic disease correction parameter and the activity correction parameter have preset value ranges, and the sign direction, applicable vital sign fields, applicable abnormal direction, and disabling conditions of the activity correction parameter are determined by the activity adjustment table. No activity deduction is performed for risk components corresponding to decreased blood oxygen, abnormal body temperature, abnormal blood glucose, or the association rule pre-trigger identifier. The sub-item abnormal scores, the chronic disease correction parameters, and the activity correction parameters determined by the vital sign fields are input into a weight fusion table to calculate the basic comprehensive score. The association rule pre-triggers the generation of the multi-signal association gain component, then adds the basic comprehensive score to the multi-signal association gain component and performs clipping according to the upper and lower boundaries of the preset score interval to obtain the comprehensive health score; according to the warning level mapping table, the comprehensive health score is mapped to a normal state identifier or a multi-level warning level identifier, the multi-level warning level identifier includes N warning levels arranged from high to low severity, where N is an integer not less than 2; when the mapping result is any of the multi-level warning level identifiers, a health warning record is generated and processed. In the hierarchical routing process, automatic distribution is not triggered when the mapping result is a normal status identifier or an alarm-free status. When the record-level confidence level does not meet the preset confidence threshold conditions configured by the confidence parameter table, the key field has low confidence, the number of valid fields does not meet the lower limit conditions of the number of fields configured by the scoring security parameter table, the mapping table is not hit, or the rule table is unreadable, the warning level identifier or security status field is set to pending manual review, mapping abnormal, or score retention status, and automatic distribution is not triggered. When the manual review interface confirms that an alarm is required, a health warning record with the source of manual review is generated.

[0026] Furthermore, the multi-sign associated gain components are determined through an association rule table, which includes a sign combination field, a combination trigger condition field, a gain coefficient field, and an applicable scenario field.

[0027] When multiple abnormal scores in the heart rate score, blood pressure score formed by taking the maximum value of the systolic blood pressure score and diastolic blood pressure score, blood oxygen saturation score, body temperature score and blood glucose score do not reach the corresponding single alarm threshold but satisfy the combination relationship defined by the combined triggering condition field, or when at least one of the abnormal scores in the sub-item does not reach the corresponding single alarm threshold and together with the preset scenario conditions, satisfies the combination relationship defined by the combined triggering condition field, the multi-signal association gain component is generated according to the corresponding gain coefficient;

[0028] The combined relationships include at least the combination of the direction of decreasing blood oxygen saturation and the direction of increasing heart rate, the combination of the direction of increasing body temperature and the direction of increasing heart rate, the combination of the direction of increasing systolic or diastolic blood pressure and the direction of increasing heart rate, and the combination of signs and scenarios that are satisfied by both the direction of abnormal blood glucose and the postprandial scenario.

[0029] Optionally, S5 includes:

[0030] A routing matrix is ​​established, indexed by the warning level identifier, the receiver role hierarchy, and the applicable scope or authorization matching rules corresponding to the receiver role. The receiver role hierarchy is divided according to the alarm handling level and the receiver's function, and is not limited to the specific name of the receiving institution. The receiver role hierarchy includes, but is not limited to, guardians, community doctors, personnel from elderly care service institutions, street or community management personnel, and superior supervisors. Receivers with the same handling level or the same handling function are all assigned to the corresponding receiver role hierarchy. The guardian role is matched based on the guardianship authorization relationship or other personal authorization relationship; the community doctor or receiver with similar medical service handling functions is matched based on the medical service relationship or service jurisdiction; and the elderly care service institution personnel or receivers with similar functions are matched based on the medical service relationship or service jurisdiction. For recipients of similar service and disposal functions, the matching basis is the institutional disposal relationship. For recipients of similar grassroots management or higher-level supervision functions, the matching basis is the administrative scope, service scope, or management authority. When an alarm is triggered, the scope or authorization matching corresponding to the recipient's role is performed based on the administrative division code of the monitored object, guardianship authorization relationship or other personal authorization relationship, medical service relationship, institutional disposal relationship, service scope, management authority, and the administrative division code under the recipient's jurisdiction. The push strategy corresponding to the role level is read according to the warning level identifier. When the same alarm meets multiple push strategies at the same time, a distribution queue is generated according to the lexicographical order rules of the disposal time limit, role level priority, and receipt penalty status corresponding to the warning level identifier.

[0031] Optionally, S6 includes:

[0032] The alarm content is divided into fields for identity, contact information, address, location, vital signs, rating, warning level, and handling suggestions.

[0033] Based on the permission level of the alarm distribution object, the field retention rules and field replacement rules are read from the desensitization rule table, and hierarchical desensitization is performed on the identity field, contact information field, address field, and location field;

[0034] Before pushing the alarm, verify the role permissions, scope of application permissions, authorization validity period and authorization relationship with the recipient role of the alarm distribution object. The scope of application permissions of the guardian role is satisfied by the guardianship authorization relationship, and the scope of application permissions of the community doctor, elderly care service institution disposal personnel, street or community management personnel and superior supervision personnel are satisfied by the medical service relationship, institution disposal or jurisdiction management authority.

[0035] When the verification passes, an anonymized alarm message is pushed and written to the audit log; when the verification fails, only a rejection record is written and the output of privacy fields is prohibited.

[0036] On the other hand, the present invention also provides a health classification early warning information processing system, comprising:

[0037] The data access preprocessing module is used to acquire vital sign data and auxiliary data of the monitored objects and generate standardized monitoring records; the scenario baseline modeling module is used to determine physiological or behavioral scenarios and establish personalized long-term baselines, personalized short-term baselines, and personalized fluctuation standard deviations; the anomaly deviation calculation module is used to calculate personalized anomaly deviation degrees and control baseline update weights; the scoring and early warning module is used to generate sub-item anomaly scores, comprehensive health scores, and early warning level identifiers; the hierarchical routing module is used to determine the alarm distribution targets based on the early warning level identifier, the receiver role level, and the applicable scope or authorization matching criteria of the corresponding role; and the privacy control push module is used to perform hierarchical desensitization of alarm content, authorization verification, alarm push, and audit records.

[0038] The beneficial effects of this invention are:

[0039] 1. This invention establishes personalized long-term baselines, short-term baselines, and standard deviations of fluctuation for the monitored subjects under different physiological or behavioral scenarios. In abnormal or warning states, it reduces or suspends abnormal data from participating in baseline updates, so that warning judgment no longer depends on a uniform fixed threshold. This can reduce false alarms caused by scenario-based fluctuations such as activity, sleep, and post-meal, while avoiding abnormal data from polluting subsequent baselines.

[0040] 2. This invention incorporates the deviation of measured values ​​from the long-term baseline and the drift of short-term baselines from the long-term baseline into the abnormal deviation score. It generates a comprehensive health score through sub-item abnormality scoring, chronic disease correction, activity correction, and multi-sign association gain components. It can identify complex abnormalities where individual signs have not exceeded the limit, but multiple signs or signs and scenarios have shown risk.

[0041] 3. This invention utilizes a routing matrix composed of warning levels, receiver role levels, and the applicable scope or authorization matching criteria corresponding to the receiver roles. Before pushing, it performs field-level desensitization and authorization verification, so that different receivers receive alarm content that matches their handling responsibilities and permissions, thereby reducing information overload caused by indiscriminate push and reducing the risk of privacy leakage. Attached Figure Description

[0042] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0043] Figure 1 This is a flowchart of a method for processing health classification and early warning information.

[0044] Figure 2 This is a flowchart of step S4 of the present invention. Detailed Implementation

[0045] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0046] refer to Figures 1-2 A method for processing health classification early warning information, comprising:

[0047] S1. Acquire vital sign data and auxiliary data of the monitored object and generate standardized monitoring records; S2. Determine the physiological or behavioral scenario based on the standardized monitoring records and establish personalized baseline parameters including personalized long-term baseline, personalized short-term baseline, and personalized standard deviation of fluctuation; S3. Calculate personalized abnormal deviation based on the degree of deviation of measured vital sign values ​​from personalized baseline parameters, and control the updating of personalized baseline parameters according to the warning status and abnormal markers; S4. Perform nonlinear transformation and weighted fusion on personalized abnormal deviation to obtain a comprehensive health score, map the comprehensive health score to a normal status identifier or a multi-level warning level identifier, and generate a health warning record using any multi-level warning level identifier as the alarm trigger level; S5. When an alarm is triggered, determine the alarm distribution object in the routing matrix according to the warning level identifier, the receiver role level divided by handling level and function, and the role's applicable scope or authorization matching rules; S6. Perform hierarchical desensitization and authorization verification of the alarm content according to the permission level of the alarm distribution object, and push the corresponding alarm information to the alarm distribution object that has passed the authorization verification.

[0048] In this specific embodiment, S1 includes:

[0049] In this specific embodiment, the data access preprocessing module of the monitoring center establishes communication connections with the wrist-worn wearable device, arm-type blood pressure device, continuous glucose monitoring device, and edge agent. The edge agent maintains a device binding table for each monitored object, and the record fields of the device binding table include the monitored object identifier. Equipment identification Equipment type Binding start time Binding status and protocol version When the edge agent receives a device message, it first determines whether it is in accordance with the following criteria: Retrieve the binding table, only Records that are valid and whose message time falls after the binding start time are written to the collection cache, while messages that do not match the binding relationship are written to the isolation cache and marked as pending review.

[0050] The edge proxy converts fields from different device protocols into a unified standard field structure, which includes heart rate. systolic blood pressure diastolic blood pressure Blood oxygen saturation ,body temperature ,blood sugar Activity level Wearing status Post-meal event markers Sleep period marking Locating the administrative division code Sampling time Equipment identification and data quality identifiers The protocol conversion table is stored in the local configuration area of ​​the edge agent. Each record in the table includes the device type, original field name, target field name, unit conversion factor, allowed range, and missing padding values. The edge agent is configured according to the protocol version. Read the corresponding record to complete the field mapping;

[0051] Post-meal event tagging Sleep period markers are written from continuous glucose monitoring device events, user meal record interfaces, or the monitoring center's meal event table. The data is written from the wearable device's sleep state or the platform's schedule. When multiple sources exist within the same period, the module determines the source priority according to the order of continuous blood glucose events, user confirmation records, and platform schedules, and writes the source type accordingly. and ;when When the file is missing, expired, or of low confidence of origin, the module will... Set it to 0 and write the post-meal mark to be reviewed, so that S2 does not enter the post-meal scenario and S4 does not trigger the post-meal blood glucose combination rule;

[0052] The monitoring center follows the preset sampling cycle. Time alignment is performed on data from multiple devices monitoring the same object, and the sampling time is adjusted accordingly. Falling into the same period window The fields are merged into periodic records. When multiple valid values ​​appear for the same field within the same period window, the valid value closest to the center time of the period is taken as the field value. When fields such as heart rate, blood pressure, blood oxygen saturation, body temperature, or blood glucose are missing, the system does not use the previous value as a substitute, but instead... Write missing identifier in This enables subsequent baseline and scoring calculations to identify fields that lack valid measurement sources;

[0053] For fields with inconsistent units, the preprocessing module performs standardization according to a unit conversion table. Specifically, body temperature is standardized to degrees Celsius, blood glucose to millimoles per liter, blood pressure to millimeters of mercury, and activity level to cycle steps or equivalent activity counts. For vital signs fields... If the converted value Not within the physiologically permissible range corresponding to the device type The data quality identifier for this field is then specified. Set it to invalid for out-of-bounds and store the original value in the original message index instead of using it in subsequent scoring calculations;

[0054] To suppress transient noise, the preprocessing module performs sliding window smoothing on the same field with valid quality identifiers. The window length is configured in the preprocessing parameter table according to the device type and field type. The set of valid values ​​within the window is denoted as Smoothing values ​​according to Generate, and when the number of valid samples in the window is lower than the preset threshold for the number of valid samples. The original measurement value is retained and The field is set to a low confidence flag, which reduces the weight of the field in subsequent scoring steps.

[0055] The preprocessing module also generates a processing chain identifier for each cycle record. The processing chain identifier consists of the protocol conversion table version, unit conversion table version, quality filtering rule version, and smoothing window version; when any table version is missing or inconsistent with the device protocol version... If there is a mismatch, the module writes the period record to the queue to be reviewed and outputs a quality indicator. Subsequent steps only read For fields that are valid or have low confidence, protocol upgrades, device replacements, or field mapping failures will not be mistakenly used as real vital sign fluctuations in early warning calculations;

[0056] After preprocessing, the data is fed into the preprocessing module to generate standardized monitoring records. And write it into the time series database of the monitoring center. The fields include , , Smoothing values ​​of various vital signs Activity level Wearing status Post-meal event markers Mark source after meal Sleep period marking Data quality identification The original message index and edge agent identifier; the standardized monitoring record serves as the sole input object for S2 scenario baseline modeling, and original messages that fail binding verification, unit conversion, or quality filtering are not directly entered into subsequent calculations.

[0057] In this specific embodiment, S2 includes:

[0058] The scene baseline modeling module reads the standardized monitoring records written by S1. According to the identification of the monitored object Maintain physiological or behavioral scene identifiers in the scene status table. The fields of the scenario status table include activity level. Wearing status Sampling period Post-meal event markers Sleep period marking and location administrative division code ;when When the wearing is invalid, this period will not participate in the scene judgment sample update, and will only output the invalid wearing scene identifier for subsequent scoring steps to perform quality control;

[0059] During the effective wearing period, the scenario baseline modeling module generates scenario identifiers using a deterministic scenario determination model. The model uses state vectors For input, and according to priority rules Perform matching: when True and activity level Below the sleep activity threshold Output sleep scene at time; when Preset post-meal duration Output the post-meal scenario when the content is true; output the post-meal scenario when the activity level is true. Not lower than the activity threshold The active scenario is output during the active period; the resting scenario is output during the remaining effective wearing periods; and each threshold is displayed. , and duration Stored in the scene parameter table, read according to the age group of the monitored object and the device type, and read the population reference parameter record when no match is found;

[0060] The scene determination model also maintains a scene confidence field. and conflict identifier field When both sleep and post-meal scenarios are met within the same period, the module outputs the scenario that comes first according to the aforementioned priority and writes the scenario that comes later into the conflict flag field; when the activity level field is missing but the sampling period is... and When sufficient to determine the sleep scenario, Take a downgraded confidence value and prohibit this period from entering the activity baseline sample. When all scenario determination conditions are unavailable, the module outputs a scenario pending review status. This prevents the cycle from participating. , , and renew;

[0061] For each vital sign and scene The module reads a set of records from the historical normal state record library that have not triggered anomaly flags and whose data quality identifiers are valid. With a long window Personalized long-term baseline calculation within the records and personalized standard deviation of fluctuation In this specific embodiment, , ,in Containing only scene identifiers equals and fields Valid history records, long-term window length Provided by the baseline parameter table and with version identifier ;

[0062] when When written to the scene status table, the scene baseline modeling module retains the standardized monitoring records for that period. and the original scene input However, it will use its sample access identifier. Set to zero; this record is only for manual review and will not be entered into the historical normal state record database to avoid data with uncertain scenarios being incorrectly classified into rest, activity, sleep, or post-meal scenarios.

[0063] The module also uses the most recent consecutive time window Internal read of valid record set not marked as abnormal Calculate personalized short-term baseline ;like Empty or the number of valid records is less than the short-term sample threshold. Then the short-term baseline Use the previous version's short-term baseline; if no previous version's short-term baseline exists, let And write a sample insufficiency flag, which in S3 causes the weight of the short-term drift component to adopt the short-term default weight, so as to avoid the trend judgment formed by a single random measurement value.

[0064] In a certain scenario Number of historical normal status records The preset sample size threshold has not been reached. At that time, the module uses the baseline of the adjacent scene. and population reference baseline Generate an initial personalized long-term baseline, calculated according to the following rules: ,in And was cut into interval, The population key is formed by the combination of age group, gender, and chronic disease labels of the monitored subjects; when the individualized standard deviation of fluctuation... When samples are insufficient, the module calculates the standard deviation of adjacent scenes and the group reference standard deviation. and preset lower limit protection value The order is determined And write the cold start flag into the personalized baseline table;

[0065] For the activity level field, the scenario baseline modeling module uses the same scenario. The set of activity levels recorded in the internal history of normal states Calculate the baseline of activity levels in the scenario The Adopting a long-term window of vital signs The same statistical window and sample cleaning rules, and with Scene and baseline version Write it into the activity baseline table; if The number of samples is lower than the active sample threshold. The module generates data based on baseline activity levels in adjacent scenes and reference values ​​for crowd activity. When a hit is missed, the lower limit of activity level protection is used. As The minimum available baseline for S4 to read It has a confirmed source and a frozen version;

[0066] The scene baseline modeling module performs sample cleaning before each baseline generation, first removing samples... For invalid wearing, For records that are invalid due to exceeding the limit or have already been written to the exception event cache by S3, then sort them according to the sampling time. The sorting retains records that fall within the window; if multiple device sources exist for the same field within the same period, the module prioritizes the source ranked first in the data quality weight table, and uses the device identifier when weights are equal. The stable order ensures that the same baseline version is obtained when the same batch of input records is repeatedly calculated;

[0067] The cleaned record set is arranged according to Vital signs Scene and baseline version Form candidate sample keys, the module samples samples at the same time Same device identifier Only the record with the highest quality weight is retained for duplicate records; if the number of samples after cleaning is still lower than [the previous record], [the remaining record will be removed]. or The corresponding baseline field is entered into a cold start or a branch that continues to be used, and the reason for the deficiency is written into the baseline update log;

[0068] The scene parameter table is periodically calibrated by the monitoring center according to device type and object grouping. The table fields include device type, age group, and sleep activity threshold. Activity threshold Postprandial duration The module reads only the frozen scene parameter versions in each calculation cycle, including the confidence downgrade value, version number, and effective time. and will Together , and Write the frozen version into the scene status table. Both S3 and S4 read this frozen version to avoid inconsistencies between the scene label and the baseline version due to parameter refresh within the same cycle.

[0069] The population reference baseline table was generated by the monitoring center based on de-identified historical samples before deployment. The table fields include the population key. Vital signs Scene Reference mean Reference standard deviation Sample size, statistical time range, and version number; when the population key If no match is found, the module gradually relaxes the key fields according to the order of age group, gender, and chronic disease label. If no match is found, the vital sign is marked as baseline pending review, and the automatic upgrade of the warning level label based solely on this field is prohibited in S3.

[0070] Both the population reference baseline table and the scenario parameter table have a version rollback field. This field is used when the statistical time range of the new version is incomplete or the sample size is lower than the population sample threshold. If field range validation fails, the module continues to read the previous frozen version and writes the failed version number to the parameter exception queue; this rollback branch does not change the versions already frozen in the current period. and This ensures that S3 still has a stable scene and baseline source when calculating deviation;

[0071] The scene baseline modeling module generates each field , , Sample size Activity baseline Baseline sample size for activity levels Baseline version Insufficient sample size and cold start flags are written into the personalized baseline table. The personalized baseline table is then used to... Vital signs and scene The key is a composite key; the personalized baseline table is frozen to the current version before the end of each calculation cycle, S3 only reads the frozen version and writes the update suggestion to the next version cache, and S4 only reads the data from the same frozen version. This avoids the simultaneous participation of measured outliers within the same period in both the comparison benchmark and the anomaly determination.

[0072] In this specific embodiment, S3 includes:

[0073] The abnormal deviation calculation module reads the current standardized monitoring records. And the S2-frozen personalized baseline table for each vital sign Obtain measured values Scene identification Long-term baseline Short-term baseline Personalized fluctuation standard deviation and preset lower limit protection value The preset lower limit protection value Used to prevent the deviation from being amplified when the standard deviation of a stable object approaches zero. Stored in the deviation parameter table according to the vital signs field and having the same units as that vital sign;

[0074] For vital signs exhibiting bilateral abnormal orientations, the abnormal deviation calculation module follows... Calculate the long-term deviation component according to Calculate the short-term drift component, where The bilateral abnormality direction is applicable to body temperature, blood glucose, and fields marked in the configuration table as requiring monitoring of both upper and lower limits. The numerical unit is... After normalization, dimensionless components are formed;

[0075] For vital signs with unilateral abnormal orientation, the module reads the orientation parameters from the abnormal orientation table. ,when When adopted and ,when When adopted and The long-term deviation component and the short-term drift component are respectively and This converts abnormalities in different directions, such as increased heart rate, increased blood pressure, and decreased blood oxygen, into dimensionless components with increased numerical values, indicating increased risk.

[0076] The abnormal deviation calculation module is based on Obtain personalized anomaly deviation, weight and From the deviation weight table and satisfying When S2 writes the short-term insufficient sample flag, Set to short-term default weight and allocate the remaining weights to This allows for comparisons with long-term stable states even when recent samples are insufficient;

[0077] The personalized abnormal deviation It is a dimensionless non-negative number, and in , and When unchanged, long-term deviation components or short-term drift components Increasing will inevitably make Increase; the module writes this monotonic direction into the deviation vector. Direction field This allows S4 to confirm the numerical direction of the abnormal scores for each item, avoiding misinterpretation of deviation as health or credibility.

[0078] The deviation weight table is established by the monitoring center before deployment based on the vital signs field, which includes vital signs. Long-term deviation from weight Short-term drift weights Short-term default weights Applicable scenarios, version number, and effective date; modules are categorized by... and If the table lookup fails and no match is found, the default record is used. and and in A default weight identifier is written into the data so that subsequent audits can distinguish between true trend fusion and judgments based solely on long-term deviations.

[0079] The anomaly calculation module first generates individual anomaly markers before determining the baseline update weights. The Determined by the frozen version of the individual anomaly threshold table; the individual anomaly threshold table fields include vital signs. Scene Abnormal direction Deviation threshold Sub-item scoring thresholds Boundary inclusion rules, version number And the effective time, the module is frozen at the start of the current cycle. Read threshold, according to Alternatively, it can be pre-computed via the S4 homologous nonlinear mapping. generate Generate when the threshold is not reached. ;

[0080] When a single anomaly threshold table is not found, the version is unreadable, or the boundary containment rule is missing, the module will not add the record to the historical normal state record library, but will instead... Set to pending review status and output the exception marker and reason for pending review; all After S3 is calculated, the deviation vector follows... Freeze. S4 only reads the freeze flag for score tracing and combination rule judgment to avoid S4 score results from adversely changing the baseline update admission conditions for this period.

[0081] The module simultaneously controls the update weights of both the long-term and short-term baselines; at the start of the current period, the module reads the comprehensive health score from the previous frozen calculation period. and warning level labels As a baseline update state, do not read data that has not yet been generated by S4 in this cycle. and ;like The status is normal or no alarm has been triggered; the quality identifiers of all participating fields are currently valid. For effective wearing and any vital signs If all values ​​are 0, then the current record is added to the baseline update sample, and the weights are updated. Taken from the data quality weight table; if any vital sign... 1. Any Status pending review The alarm status has been triggered or the warning level is level 2 or above. If the wear is deemed invalid, the entire record will be subject to abnormal weighting or branch suspension. Take the anomaly update weight table and Values ​​corresponding to record-level exception states When this value is equal to zero, the record is suspended from participating in baseline updates;

[0082] Records suspended from participating in baseline updates will not be deleted; the module will record them as exception events. Write to the exception event cache. include , Vital signs field Measured values Long-term deviation components Short-term drift components Personalized abnormal deviation Triggering cause and baseline version The exception event cache is used for manual review and S4 scoring traceability, but is not written to the current version's historical normal status record library.

[0083] The data quality weight table includes fields for quality indicators. The fields for the abnormal update weight table include the field type, weight value, applicable device type, and version number. Fields for this include the identifier of the previous frozen warning level. Record-level abnormal states Wearing status, update weight, and pause flag; the aforementioned From the entire period The summary shows that as long as any one of them is correct. If the value is 1 or pending review, it indicates an abnormal or pending review value; if no table is found, the module will... Set to zero and output a pending review and update status, thus forming a safe state, that is, the current record is not allowed to change the personalized baseline when the quality or anomaly level cannot be confirmed;

[0084] For records that have been suspended from participating in baseline updates, the module also generates a review task identifier. The review task fields include the cause of the anomaly, the affected vital signs fields, the current baseline version, and the recommended review time limit. When the review result confirms that the record is due to equipment error, the record will not be included in the historical normal state record library. When the review result confirms that the record is a real but recovered normal sample, the record can only be included in the candidate sample pool after the next frozen version. The baseline version that has already been used for scoring will not be modified in reverse.

[0085] The module uses a shadow update method to verify the baseline update results and candidate baselines. and First, write the next version cache and compare it with the current frozen baseline; then each module calculates the long-term drift. and short-term drift ,like Exceeding the long-term baseline drift review threshold or Exceeding the short-term baseline drift review threshold The module retains the current frozen baseline and outputs the baseline drift review status. Only the cached version that passes the review can become the frozen baseline for the next cycle.

[0086] The abnormal deviation calculation module calculates the input source and divisor for each component. The weight version and anomaly direction parameters are written to the calculation chain log; when the S4 or review end needs to reproduce the comprehensive score later, it can be restored according to the calculation chain log. , , , and The value version is selected to avoid reproduction deviation caused by mixing baselines from different periods;

[0087] The direction for determining personalized anomaly deviation is fixed in the deviation direction table as a monotonically increasing risk level. A value of zero indicates that neither the measured value nor the short-term baseline has formed a valid abnormal deviation relative to the long-term baseline. The larger the value, the higher the risk. A smaller value indicates lower risk; this value is a negative penalty when applied to the overall health risk category, rather than a positive return item used to select the highest-return target, and is included as a positive risk score component. This module is used by S4 for mapping and outputting alert thresholds; it writes the direction, field validity, and whether it participates in S4 scoring. In Fields, S4 only for The deviation is used to perform a nonlinear mapping;

[0088] After completing the calculation, the anomaly deviation calculation module outputs the deviation vector. The Stored by vital signs field , , Single anomaly marker Validity identification and updated weight and will Write to the early warning calculation cache; S4 only reads from the early warning calculation cache related to the current period. Baseline version and abnormal threshold version Consistent deviation vector, generated in this period S4 and It is only written to the next version cache as the baseline update state for the next cycle, ensuring that the input versions of the scoring chain and the baseline update chain are closed.

[0089] In this specific embodiment, S4 includes:

[0090] The scoring and early warning module reads the deviation vector output by S3. Standardized monitoring records of S1 For each vital sign Perform a monotonic nonlinear mapping to obtain sub-item anomaly scores. In this specific implementation, the mapping rule is as follows: ,in vital signs The sensitivity coefficient is stored in the scoring parameter table. Cut in The larger the value in the range, the greater the degree of abnormality of the field relative to the personalized scenario baseline. Before generating the activity correction parameters, the module first reads the association rule table based on the sub-item abnormality score and the corresponding abnormality direction, and generates the association rule pre-trigger identifier and the set of association risk fields.

[0091] The scoring and early warning module is based on the chronic disease labels of the monitored subjects. Read chronic disease adjustment parameters from the chronic disease adjustment table The chronic disease adjustment table includes fields for chronic disease label, vital signs, correction parameters, applicable scenarios, and version number; for activity impact, the module reads the data from the activity baseline table frozen in S2. Scene and baseline version Corresponding scenario activity baseline First, take the lower limit of the activity level. , then calculate ,in This is the lower limit protection value for activity level. From the activity adjustment form, This refers to dimensionless activity deviation; the activity adjustment table is based on vital signs. Abnormal direction and scene Give the deduction factor Maximum deduction ratio and disabling conditions; high activity levels only generate non-positive activity correction items for the activity-explained direction of heart rate or blood pressure increase fields. For fields corresponding to decreased blood oxygen, abnormal body temperature, abnormal blood sugar, low-confidence data fields, or pre-triggered markers of association rules, Set to 0 or transfer to manual review;

[0092] The module inputs the sub-item abnormal scores for each vital sign, chronic disease correction parameters, and activity correction parameters determined by each field into a weighted fusion table to calculate the baseline comprehensive score. ,in Scoring for dimensionless fields The weights are determined by vital signs, and the sum of the weights for the current version of the same object is 1. and and The data are all on the same percentage scale; when the data quality indicator of a certain vital sign field is missing or low confidence, the module reduces the value of that field according to the quality weight table. The reduced weights will be redistributed according to the original weight ratios of the valid fields; the set of key fields is configured by the scoring safety parameter table according to the scenario and chronic disease label. If the record-level confidence is lower than the preset confidence threshold, the data quality of the key fields is marked as low confidence, or the number of valid fields is lower than the minimum field number threshold, the score will be adjusted accordingly. If so, the output status will be set to pending manual review instead of automatically escalating the warning level indicator;

[0093] To identify situations where an individual item does not exceed the limit but the combination presents a risk, the scoring and early warning module generates multi-signal correlation gain components based on the aforementioned association rules and pre-trigger identifiers. Each record in the association rule table includes a vital sign combination field, a combination trigger condition field, a gain coefficient field, an applicable scenario field, and a rule version. The trigger condition is jointly limited by the sub-item abnormal score threshold, the abnormal direction, and preset scenario conditions; when the blood oxygen saturation decreases, the score is... Score based on the direction of heart rate rise Simultaneously meeting the corresponding triggering conditions, or scoring the direction of body temperature rise. Score based on the direction of heart rate rise When the corresponding triggering conditions are met simultaneously, the module reads the corresponding gain coefficient and accumulates it. ;

[0094] For combinations of blood pressure and blood sugar, the module first uses an abnormal score for the systolic blood pressure component. Abnormal scores for diastolic blood pressure component Blood pressure score And process them according to the same association rule table. The corresponding combinations of the direction of blood pressure rise and the direction of heart rate rise, as well as the combinations of signs and scenarios that satisfy the abnormal blood glucose direction and the postprandial scenario. If multiple combination rules are triggered simultaneously, ,in For the first The gain coefficient of the triggered rule. Then, according to the upper bound of the gain in the association rule table Cut off the data to avoid repeated amplification of the overall health score by multiple related combinations in the same period;

[0095] The scoring and early warning module calculates a comprehensive health score. And according to the warning level mapping table Mapped to warning level identifier The warning level mapping table includes the score range, handling time limit, and whether an alarm has been triggered for normal or no-alarm status, as well as Level 2, Level 3, and Level 4 warnings. The "whether an alarm has been triggered" field is false for normal or no-alarm status. Warning levels above Level 2 are identified as the alarm trigger level. The interval boundaries are configured by the monitoring center and include a version number. When located at the intersection of two interval boundaries, the interval rule that includes the lower boundary but not the upper boundary is adopted, and the highest interval includes its upper boundary, ensuring that the same score has only one warning level label;

[0096] The fields of the weighted fusion table include vital signs. Scene Chronic disease label Field quality level, weight The module is defined by its weight version and effective time. , and Used as the lookup key; when the quality level of a certain field is missing, the corresponding key for that field... Each input is not included in the summation; after the weights are redistributed, it is equal to... The score within the interval or the correction parameter within the preset interval, therefore It is a dimensionless score and is not directly added to the original measurements with units, such as blood pressure, body temperature, and blood sugar.

[0097] When the record-level confidence level is lower than the preset confidence threshold, the data quality of key fields is marked as low confidence, or the number of valid fields is lower than the minimum field number threshold, At that time, the scoring and early warning module enters the record-level low-confidence safety branch and generates... and ,in The security status value serves as the identifier for the warning level field. This is a reference score identifier used only for interface display and not an automatic alarm score; this security branch will still write missing fields, low-confidence key fields, number of valid fields, and trigger reasons. However, it will not send an automatic distribution request to S5 unless the manual review interface marks the record as requiring a manual alert.

[0098] The association rule table is generated by the medical rule configuration terminal and published by the monitoring center before deployment. Each combination trigger condition in the table specifies the participating fields, anomaly direction, sub-item anomaly scoring threshold, applicable scenario, and mutually exclusive rules. When multiple rules exist for the same vital sign combination, the module sorts them according to rule priority, accuracy of applicable scenario, and rule version number, and takes the rule that appears first in the list. Calculate to avoid the same combination being counted repeatedly;

[0099] The early warning level mapping table is configured by the monitoring center according to the operational procedures. Fields include mapping version, lower score boundary, upper score boundary, early warning level identifier, handling time limit, whether an alarm has been triggered, and manual review requirements. The mapping table is frozen by version for use within the same calculation period. If no interval is hit, the module will Set to a mapping exception safe state And stop automatic distribution, and write the record to the scoring rule exception queue;

[0100] When any version of the association rule table, weight fusion table, or warning level mapping table becomes unreadable, the scoring and warning module adopts a safe output state. and will Write the security status bit to the warning level identifier field; this status retains the sub-item anomaly score and deviation vector but does not generate an automatic alarm distribution object; when the table version is restored, the module will be in a frozen state. , The scores are recalculated based on the rule version, without using baseline or activity data that has already entered the next cycle to replace the current cycle's input;

[0101] for , and The scoring and early warning module generates unified safety status records for three safety status categories. The This includes security status type, trigger rule version, set of missing or abnormal fields, and reserved fields. Version, Reserved Version and manual review entry identifier; the module will Health warning records Associated storage, and Set to automatic dispatch disabled state, so that S5 can only read trend displays or manual review tasks and cannot generate automatic alarm dispatch objects; when the manual review interface returns... Artificial early warning level Manual triggering identifier and reviewer identification At that time, the module generates health alert records with manual review. ,Will Set to manual confirmation and distribution permission status, and with As the routing input for S5;

[0102] After the scoring is completed, the scoring and early warning module outputs a health warning record. The include , , Sub-item abnormal score set, chronic disease correction parameters, activity correction parameters, multi-sign association gain components Comprehensive health score Warning level indicators Security status field Scoring parameter version and baseline version When the "whether an alarm has been triggered" field in the warning level mapping table is true and When automatically distributing allowed states, It is sent to the S5 hierarchical routing module; when When manually confirming the distribution permission status, Sent to S5 and carrying the source for manual verification, and Otherwise, it will only be written to the health trend database, manual review queue, or scoring rule exception queue for subsequent processing.

[0103] In this specific embodiment, S5 includes:

[0104] The hierarchical routing module reads the health alert records output by S4. And read the candidate receiver set from the receiver directory table. The recipient directory table includes fields for recipient identifier. Role hierarchy Jurisdiction Administrative Division Code Set of receiveable warning levels The processing time limit, type, duty status, and contact information; the candidate recipient set is first determined by... Filtering, i.e., keeping only belong And the on-duty status is a valid recipient record;

[0105] Routing matrix Indexed by warning level identifier, role hierarchy, and the applicable scope or authorization matching rules corresponding to the recipient's role, each matrix record includes whether a push notification was sent, the push time limit, whether a receipt is required, escalation delay, and de-identification policy identifier; the hierarchical routing module... In , The system identifies the monitored entities and performs role-based matching for each candidate recipient. Guardians are matched based on their guardianship authorization relationship; community doctors are matched based on their medical service relationship or service jurisdiction; personnel from elderly care service institutions are matched based on their institution's handling relationship; and street or community management personnel and higher-level supervisors are matched based on their jurisdiction according to the administrative division hierarchy table. Once a match is established, [the system will...]. Set to 1 otherwise set to 0;

[0106] For the guardian role, the module reads the authorization relationship in the guardianship authorization table. If the authorization relationship is valid and has not expired, it will... Set to 1, no administrative division code is required; for the community doctor role, the module reads the contract, follow-up, or chronic disease service relationship table, and determines whether a medical service relationship or service jurisdiction exists. When Set to 1; For personnel handling matters at elderly care service institutions, the module reads the institution handling relationship table, and when the monitored individual's admission, care, or service records point to that institution, it will... Set to 1; only for street or community management personnel and higher-level supervisors will the module read the administrative division hierarchy table and confirm according to jurisdiction or management authority. This results in the distribution of candidate records. Its fields include , , , Push time limit and desensitization strategy identifier;

[0107] When the same alarm satisfies multiple push policies simultaneously, the hierarchical routing module generates a lexicographical sorting key according to routing priority. ,in The corresponding response time limits for each warning level are normalized into dimensionless level numbers. Given by the role hierarchy priority table, A preset penalty level will be applied if the recipient fails to complete the processing as required by historical receipts; the module will be configured accordingly. The distribution queue is generated in ascending lexicographical order to avoid directly adding the different scales of processing time limit, role number and receipt penalty;

[0108] The range matching factor in the routing matrix is ​​denoted as And determined according to the recipient's role: the guardian's role is confirmed by the guardianship authorization form. The last digit is 1, and the recipient is not required to have a jurisdictional administrative division code; the role of the community doctor is confirmed by matching the medical service relationship or service jurisdiction; the personnel handling the elderly care service institution are confirmed by the institution's handling relationship; only street or community management personnel and superior supervisors read the administrative division hierarchy table to perform jurisdictional matching.

[0109] The hierarchical routing module also maintains upgrade rules. When a level 4 or level 3 alert is not acknowledged within the preset acknowledgement time, the module copies the alert to the candidate set of the next higher role level according to the upgrade delay field of the routing matrix, and writes the upgrade source and original recipient identifier into the copy record. When a level 2 alert has been confirmed by the guardian and acknowledged by the community doctor, the module merges the duplicate push records of the same monitored object in the same period into a single disposal chain record to avoid duplicate pushes to the same role.

[0110] Both the receiver directory table and the routing matrix have version numbers, and the hierarchical routing module is in the same alarm event. The module freezes the directory version and matrix version. If the contact information of the candidate recipient is empty, the duty status is invalid, or the push channel is unavailable, the module removes the recipient from the current distribution queue and writes the reason for the unavailable channel. At the same time, it continues to check the next candidate record at the same role level to ensure that the channel failure does not cause the entire routing matrix to fail.

[0111] After filtering based on role level, role scope of application or authorization matching criteria, and channel availability... When empty, the hierarchical routing module enters the routing security state. The event number, warning level identifier, administrative division code of the monitored object, and reason for failure are written into the manual dispatch queue, and S6 is prohibited from generating any push content containing privacy fields; manual dispatchers can only see the de-identified event summary, and S5 role matching will be re-executed after the recipient directory, role scope record or authorization relationship is supplemented;

[0112] When generating the distribution queue, the module also records the matching evidence for each candidate recipient, including the matrix record of the warning level hit, the role level hit record, the role application scope matching path, and the authorization relationship record; among them, the role application scope matching path record of street or community management personnel and superior supervisors is recorded as the administrative division level chain, the community doctor records the medical service relationship or service jurisdiction hit record, the elderly care service institution disposal personnel records the institution disposal relationship, and the guardian records the guardianship authorization relationship; the matching evidence is passed to S6 as an immutable field of the alarm routing table, and S6 selects the de-identification strategy accordingly and retains the push basis in the audit log to prevent subsequent changes in the permission table from affecting the interpretation of the completed alarm distribution;

[0113] For route upgrades, the module will use the original event number. With upgrade serial number Combine them into an upgrade event key, and retain the receipt status, push time and upgrade reason of the previous recipient in the upgrade event; when the upper-level role receives the receipt, the module will synchronize the receipt status back to the original event chain without regenerating the health score or changing the warning level identifier of the S4 output, ensuring that the route upgrade only affects the handling flow and does not change the risk judgment.

[0114] The records in the manual scheduling queue only contain event numbers. Warning level indicators The data includes the jurisdictional location, failure reason, and available role level, but does not include name, phone number, or detailed address. When dispatchers add temporary recipients, the module still needs to re-execute role permissions, role scope or jurisdiction, and authorization validity period verification. Only after the verification result is successful will a new distribution candidate record be generated. If the verification fails, the data will be retained. The status remains unchanged and the output of privacy fields is still disabled.

[0115] The hierarchical routing module also performs a merging judgment on consecutive alarms for the same monitored object, with the merging key being determined by... Warning level indicators, set of main abnormal signs and time windows Composition; in Alarms that recur within the same period and whose warning level has not been raised are only added to the existing handling chain and the receiver set is not expanded again; when the warning level is raised or the set of main abnormal signs changes, the module generates a new distribution queue and retains the association number with the old handling chain.

[0116] The hierarchical routing module outputs a set of alarm distribution objects. And write it to the alarm routing table. It only includes recipient records matched by role hierarchy, alert level, and role scope or authorization; each record carries a de-identification policy identifier, push deadline, receipt requirements, upgrade rules, and routing matrix version, serving as input for S6 to perform hierarchical de-identification and authorization verification, and is not included in the list of records to be processed. The recipient will not receive the alarm content; when the input record is At that time, the hierarchical routing module uses Replacement for automatic calculation Read the routing matrix and manually verify the source field. Write the review timestamp to the distribution queue The manual confirmation distribution path still executes the same role applicability or authorization matching rules and passes the manual source field to the S6 audit log.

[0117] In this specific embodiment, S6 includes:

[0118] The privacy control push module reads the alarm distribution object set output by S5. and health alert records output by S4 First, organize the alarm content into field-based alarm objects. The fields include identity fields. Contact information field Address field Location field Vital signs field Rating fields Warning level identifier field , Disposal suggestion field and event number Field-based alarm objects do not generate complete natural language text before entering the push queue; all privacy controls are performed at the field layer.

[0119] After the field-based alarm object is generated, the privacy control push module performs field cleaning to remove empty fields, expired contact information, and fields with the current event number. Inconsistent cached fields, and for vital sign fields and rating fields Additional source version; when multiple candidate sources exist for the same field, the module follows the S4 health alert record. S1 Standardized Monitoring Records The order of manually entered records is determined by the source. If there are still parallel sources, the order is determined by the source timestamp. Select and write the field source identifier in descending order;

[0120] The module reads field retention rules and field replacement rules from the desensitization rule table based on the desensitization policy identifier and permission level in the alarm distribution object record. The desensitization rule table includes field name, permission level, retention granularity, replacement template, and visibility conditions. For the recipient who is a guardian and whose guardianship authorization is valid, the identity field and contact information field are retained according to the authorization scope. For street or community management personnel, the identity field is replaced with the monitored object number, the contact information field is replaced with a hidden template, and the address field is retained to the community or neighborhood granularity. For superior supervisory personnel, the location field is retained to the administrative division granularity and the name, telephone number, and detailed address are hidden.

[0121] When the same field matches multiple de-identification rules, the module selects a unique rule in the order of stricter permission level restrictions, coarser location granularity, and updated replacement template version. If a unique rule still cannot be determined, a default rejection policy is adopted, setting the identity field, contact information field, detailed address field, and precise location field to hidden templates, and recording rule conflict indicators in the audit log. This sorting and conflict branching ensures that the selection of de-identification rules has a definite output and does not leak privacy fields due to rule overlap.

[0122] Before data anonymization, the privacy control push module performs authorization verification, and the verification input includes the recipient's identifier. Role hierarchy Administrative division code under the jurisdiction of the recipient Administrative division code of the monitored object Authorization Validity Period And the relationship of role authorization The verification rules are as follows: When any of the factors is 0 If the value is 0, the module will prevent the output of identity, contact information, address, and precise location fields, and will write a rejection record.

[0123] In authorization verification Determined by the role permission table, By the validity period of the authorization Determined by comparison with push time; According to the recipient role branch, the guardian role is taken as 1 when the guardianship authorization form is confirmed to be valid and no longer requires administrative jurisdiction matching. The community doctor is determined by the medical service relationship or service jurisdiction matching. The personnel of the elderly care service institution are determined by the institution's handling relationship. The street or community management personnel and the superior supervision personnel are determined by the administrative division matching evidence transmitted by S5. Similarly, based on the recipient's role branch, the guardian's role is determined by reading the guardianship authorization form. The community doctor reads the medical service relationship table to obtain... The personnel handling the situation at the elderly care service institution obtained the information from the institution's handling relationship table. Street or community management personnel and higher-level supervisors can obtain the jurisdiction management authority table. Furthermore, the role of a non-guardian who is not subject to guardianship authorization is not applicable. As a necessary multiplicative factor;

[0124] when When the value is 1, the module generates de-identification alarm objects according to the de-identification rule table. The module selects a push channel based on the recipient's role level. The channel selection table includes role level, warning level identifier, preferred channel, backup channel, number of retry attempts, and receipt requirements. The module attempts the preferred channel, backup channel, and manual scheduling channel in that order. If any channel returns a successful status, the module stops further attempts. If all channels fail, only the channel failure audit record is written and the event is returned to the S5 upgrade rules for processing.

[0125] Guardians receive information requiring immediate action via app messages or SMS; community doctors receive medical treatment information including vital signs, scoring, and treatment suggestions; community management personnel receive alert level indicators, jurisdictional location granularity, and treatment status; and higher-level supervisors receive aggregated alert level indicators and jurisdictional statistical fields. The content pushed through each channel is anonymized by the alert recipients. Rendering, not without prior anonymization. Generate directly;

[0126] After the push is completed, the module writes to the audit log. The audit log fields include the event number. Recipient Identifier The audit log includes the following information: permission level, desensitization policy identifier, authorization verification result, list of pushed fields, list of hidden fields, push channel, push time, and receipt status. If the authorization verification fails, the audit log only records the reason for rejection and the recipient identifier, and does not record the original text of the hidden privacy fields. The audit log is saved by appending and is uniquely indexed by the event number and the recipient identifier to facilitate subsequent compliance traceability.

[0127] When the push channel returns a failure, the recipient device is offline, or the receipt format does not meet the requirements of the channel selection table, the privacy control push module does not expand the scope of visible fields, nor does it switch to a recipient with higher privileges; the module only returns the channel failure status, event number, and role level to S5, so that S5 can perform upgrades or manual scheduling according to the frozen routing matrix, ensuring that communication anomalies do not bypass the desensitization rules;

[0128] When the alarm distribution target received by S6 comes from a manually confirmed distribution path, the privacy control push module, in addition to performing field desensitization and authorization verification, will also manually verify the source. , The review timestamp is written to the audit log; this log uses the same recipient identifier as the automatic alarm audit log. The authorization verification result field ensures that manually confirmed alarms are also subject to hierarchical desensitization and authorization control constraints.

[0129] The privacy control push module also monitors the recipient's feedback based on the receipt requirements carried by S5. When the recipient submits a processing receipt within the push time limit, the receipt status is written back to the alarm routing table. If no receipt is submitted after the push time limit, the module only returns the event number, warning level identifier, role level, and jurisdiction granularity information to S5 for upgrade routing, without returning the complete privacy fields. This makes alarm distribution, de-identification authorization, push receipt, and upgrade processing form a closed information processing link.

[0130] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

[0131] This invention transforms vital sign data from raw collection records into a comprehensive health score that can be used for alarm classification and response distribution through a continuous data processing chain between personalized scenario baselines, abnormal deviations, and multi-sign correlation gain components. This score reflects not only the deviation of the monitored object from its own stable state, but also recent trends, multi-sign combination relationships, and the relationship between signs and scenarios. Therefore, it can better support the identification of health risks in community-based elderly care, chronic disease management, and multi-stakeholder collaborative response.

[0132] The improved technical solution of this invention does not simply change the alarm threshold, but rather achieves synergistic improvements in the baseline modeling structure, baseline update rules, composite scoring mechanism, and distribution control mechanism. It distinguishes between stable health states and recent trends by using long-term and short-term baselines, maintains baseline reliability by downweighting or pausing updates for abnormal data, and controls alarm flow and content granularity through routing matrices and hierarchical desensitization. This allows for a closed loop of health anomaly identification, alarm handling, and privacy protection within the same information processing chain.

Claims

1. A method for processing health classification early warning information, characterized in that, include: S1. Acquire vital sign data and auxiliary data of the monitored object and generate standardized monitoring records; S2. Determine the physiological or behavioral scenario based on standardized monitoring records, and establish personalized baseline parameters including personalized long-term baseline, personalized short-term baseline, and personalized standard deviation of fluctuation; S3. Calculate the personalized abnormal deviation degree based on the degree of deviation of the measured vital signs values ​​from the personalized baseline parameters, and control the updating of personalized baseline parameters according to the warning status and abnormal markers; S4. Perform nonlinear transformation and weighted fusion on the personalized abnormal deviation to obtain a comprehensive health score. Map the comprehensive health score to a normal status identifier or a multi-level warning level identifier, and generate a health warning record using any multi-level warning level identifier as the alarm trigger level. S5. When an alarm is triggered, determine the alarm distribution object in the routing matrix according to the warning level identifier, the receiver role level divided by handling level and function, and the role's applicable scope or authorization matching rules. S6. Perform hierarchical desensitization and authorization verification on the alarm content according to the permission level of the alarm distribution object, and push the corresponding alarm information to the alarm distribution object that has passed the authorization verification.

2. The health classification early warning information processing method according to claim 1, characterized in that, Step S1 includes: collecting vital sign fields such as heart rate, systolic blood pressure, diastolic blood pressure, blood oxygen saturation, body temperature, and blood glucose, as well as auxiliary fields such as activity level, wearing status, post-meal event marker, sleep period marker, location administrative division code, sampling time, device identifier, and data quality identifier; the edge agent converts data packets under different device protocols into a unified field structure and establishes a binding relationship according to the monitored object identifier and device identifier; the converted data is processed by sampling time alignment, unit unification, missing field marking, outlier filtering, and sliding window smoothing to obtain the standardized monitoring record containing field source and data quality identifier.

3. The health classification early warning information processing method according to claim 1, characterized in that, Step S2 includes: dividing the standardized monitoring records into resting scenarios, activity scenarios, sleep scenarios, and post-meal scenarios based on activity level, wearing status, sampling period, and post-meal event marking; within each scenario, calculating the personalized long-term baseline and personalized standard deviation of fluctuation using historical normal state records, and calculating the personalized short-term baseline using records not marked as abnormal within the most recent continuous time window; when the number of historical normal state records for a certain scenario does not reach the preset sample size threshold, fusing the adjacent scenario baselines and the preset population reference baseline for the same monitored object according to the sample size weight, as the initial personalized long-term baseline for that scenario.

4. The health classification early warning information processing method according to claim 1, characterized in that, Step S3 includes: for vital signs with bilateral abnormal directions, dividing the absolute difference between the measured value and the personalized long-term baseline by the upper limit element of the binary set composed of the personalized standard deviation of fluctuation and the preset lower limit protection value to obtain the long-term deviation component; for vital signs with unilateral abnormal directions, selecting the effective difference between the measured value and the personalized long-term baseline according to the preset abnormal direction, and dividing the effective difference by the upper limit element of the binary set composed of the personalized standard deviation of fluctuation and the preset lower limit protection value to obtain the long-term deviation component; calculating the difference between the personalized short-term baseline and the personalized long-term baseline as the short-term drift component according to the same abnormal direction and the same divisor; and fusing the long-term deviation component and the short-term drift component according to the preset deviation weight to obtain the personalized abnormal deviation degree.

5. The health classification early warning information processing method according to claim 1, characterized in that, Step S3, controlling the update weights of the personalized long-term baseline and personalized short-term baseline, includes: when the previous frozen calculation period was in a normal state or an alarm-free state, the quality identifiers of all participating fields were valid, the wearing status was valid, and no vital signs triggered an abnormality marker, the current standardized monitoring record was added to the baseline update sample according to the update weight corresponding to the data quality identifier; when any vital sign triggered an abnormality marker, the previous frozen calculation period was an alarm-triggered state, or its warning level identifier was a level 2 or higher warning level identifier, or the wearing status was invalid wearing, the weight of the current standardized monitoring record for baseline updates was reset to the weight value corresponding to the warning level identifier in the abnormal update weight table, or the current standardized monitoring record was suspended from participating in baseline updates; records suspended from participating in baseline updates were retained in the abnormal event cache for subsequent review and not used for the current version of personalized baseline calculation; the comprehensive health score and warning level identifier generated in the current period were only written to the baseline update cache of the next version and did not participate in the deviation calculation of the current period.

6. The health classification early warning information processing method according to claim 1, characterized in that, Step S4 includes: performing a monotonically nonlinear mapping on the personalized abnormal deviation of each vital sign to obtain a sub-item abnormal score whose value is within a preset scoring range and whose value increases to indicate an increased degree of abnormality; reading the association rule table based on the sub-item abnormal score and the corresponding abnormal direction to determine the association rule pre-trigger identifier and its corresponding association risk field; reading the chronic disease correction parameters of the corresponding vital signs from the chronic disease adjustment table based on the chronic disease label of the monitored object; generating dimensionless activity correction parameters based on the degree of deviation of activity volume from the activity volume baseline and the lower limit protection value of the activity volume in the scenario, wherein the scenario activity volume baseline is obtained by statistical analysis of activity volume in historical normal state records in step S2; both the chronic disease correction parameters and the activity correction parameters have preset value ranges, and the sign direction, applicable vital sign fields, applicable abnormal directions, and disabling conditions of the activity correction parameters are determined by the activity adjustment table; not performing activity deduction for risk components corresponding to decreased blood oxygen, abnormal body temperature, abnormal blood sugar, or the association rule pre-trigger identifier; inputting each sub-item abnormal score, the chronic disease correction parameters, and the activity correction parameters determined according to the vital sign fields into a weight fusion table to calculate the basic comprehensive score. The multi-signal correlation gain component is generated according to the association rule pre-trigger identifier. Then, the basic comprehensive score is added to the multi-signal correlation gain component, and clipped according to the upper and lower boundaries of the preset score interval to obtain the comprehensive health score. The comprehensive health score is mapped to a normal state identifier or a multi-level warning level identifier according to the warning level mapping table. The multi-level warning level identifier includes N warning levels arranged from highest to lowest severity, where N is an integer not less than 2. When the mapping result is any of the multi-level warning level identifiers, a health warning record is generated. Upon entering the hierarchical routing process, automatic distribution is not triggered when the mapping result is a normal status identifier or an alarm-free status. When the record-level confidence level does not meet the preset confidence threshold conditions configured by the confidence parameter table, the key field has low confidence, the number of valid fields does not meet the minimum field number conditions configured by the scoring security parameter table, the mapping table is not matched, or the rule table is unreadable, the warning level identifier or security status field is set to pending manual review, mapping anomaly, or score retention status, and automatic distribution is not triggered. When the manual review interface confirms that an alarm is required, a health warning record with a manual review source is generated.

7. The health classification early warning information processing method according to claim 6, characterized in that, The multi-sign association gain component is determined through an association rule table, which includes a sign combination field, a combination trigger condition field, a gain coefficient field, and an applicable scenario field. When multiple abnormal scores in the heart rate score, blood pressure score (formed by a score that is not less than the other score among the systolic and diastolic blood pressure scores), blood oxygen saturation score, body temperature score, and blood glucose score do not reach the corresponding single alarm threshold but satisfy the combination relationship defined by the combination trigger condition field, or when at least one of the abnormal scores does not reach the corresponding single alarm threshold and together with the preset scenario conditions, satisfies the combination relationship defined by the combination trigger condition field, the multi-sign association gain component is generated according to the corresponding gain coefficient. The combination relationship includes at least the combination of the direction of blood oxygen saturation decrease and the direction of heart rate increase, the combination of the direction of body temperature increase and the direction of heart rate increase, the combination of the direction of systolic or diastolic blood pressure increase and the direction of heart rate increase, and the combination of signs and scenarios that are satisfied by the abnormal blood glucose direction and the postprandial scenario.

8. The health classification early warning information processing method according to claim 7, characterized in that, Step S5 includes: establishing a routing matrix indexed by the warning level identifier, the receiver role hierarchy, and the applicable scope or authorization matching rules corresponding to the receiver role; the receiver role hierarchy is divided according to the alarm handling level and the receiver function, and is not limited to the specific receiver institution name; the receiver role hierarchy includes, but is not limited to, guardians, community doctors, elderly care service institution personnel, street or community management personnel, and superior supervisors. Receivers who undertake the same handling level or the same handling function are all assigned to the corresponding receiver role hierarchy; among them, the guardian role is matched based on the guardianship authorization relationship or other personal authorization relationship, the community doctor or receiver who undertakes the same medical service handling function is matched based on the medical service relationship or service jurisdiction, and the elderly care service institution personnel... For recipients who undertake similar service and disposal functions, the matching basis is the institutional disposal relationship. For recipients who undertake similar grassroots management or higher-level supervision functions, the matching basis is the administrative scope, service scope, or management authority. When an alarm is triggered, the scope or authorization matching corresponding to the recipient's role is performed based on the administrative division code of the monitored object, guardianship authorization relationship or other personal authorization relationship, medical service relationship, institutional disposal relationship, service scope, management authority, and the administrative division code under the recipient's jurisdiction. The push strategy corresponding to the role level is read according to the warning level identifier. When the same alarm meets multiple push strategies at the same time, a distribution queue is generated according to the lexicographical order rules of the disposal time limit, role level priority, and receipt penalty status corresponding to the warning level identifier.

9. The health classification early warning information processing method according to claim 8, characterized in that, Step S6 includes: dividing the alarm content into identity, contact information, address, location, vital signs, rating, warning level identifier, and handling suggestion fields; reading field retention rules and field replacement rules from the desensitization rule table according to the permission level of the alarm distribution object, and performing hierarchical desensitization on the identity, contact information, address, and location fields; verifying the role permissions, applicable scope permissions, authorization validity period, and authorization relationship with the recipient's role of the alarm distribution object before pushing; wherein the applicable scope permissions of the guardian role are satisfied by the guardianship authorization relationship, and the applicable scope permissions of the community doctor, elderly care service institution handling personnel, street or community management personnel, and superior supervisors are satisfied by the medical service relationship, institution handling relationship, or jurisdiction management authority; if the verification passes, pushing the desensitized alarm information and writing it to the audit log; if the verification fails, only a rejection record is written and the output of privacy fields is prohibited.

10. A health classification early warning information processing system, used to execute the health classification early warning information processing method according to any one of claims 1 to 9, characterized in that, include: The data access preprocessing module is used to acquire vital sign data and auxiliary data of the monitored objects and generate standardized monitoring records; The scenario baseline modeling module is used to determine physiological or behavioral scenarios and establish personalized long-term baselines, personalized short-term baselines, and personalized standard deviations of fluctuation. The anomaly deviation calculation module is used to calculate the personalized anomaly deviation degree and control the baseline update weights; The scoring and early warning module is used to generate sub-item abnormal scores, comprehensive health scores, and early warning level indicators; The hierarchical routing module is used to determine the alarm distribution target based on the warning level identifier, the receiver role level, and the applicable scope or authorization matching criteria of the corresponding role; the privacy control push module is used to perform hierarchical desensitization, authorization verification, alarm push, and audit logging of alarm content.

Images