Long-term care disability state evaluation method and system, storage medium, and electronic device

Abstract

The application discloses a long-term nursing disability state evaluation method and system, a storage medium and an electronic device, and comprises application identity authorization, mobile terminal sound collection, multi-source sensing enhancement, data desensitization processing, minimum nursing evidence unit generation, disability consistency evaluation, review auxiliary decision and grading authorization output modules. In the 90 to 180 day waiting verification period, the evaluation system extracts acoustic features, event labels and time stamps locally through a mobile terminal, only uploads desensitized features, and by default retains the original audio locally; solidifies single service into minimum nursing evidence units, calculates ADL dimension nursing dependence frequency, continuity and consistency, generates disability consistency scores and review path suggestions. The application takes a mobile phone as a default collection carrier, does not require special hardware, and balances privacy protection and evaluation objectivity.

Description

Technical Field

[0001] This invention relates to the fields of long-term care insurance, home care service verification, social insurance administration informatization, and privacy-protecting computing technology. Specifically, it relates to long-term care disability status assessment methods, systems, storage media, and electronic devices, particularly a low-cost continuous self-verification and verification auxiliary assessment system and method for disability status, which is designed for a 90- to 180-day waiting period for verification and uses the existing mobile terminals of caregivers or their families as the default data collection carrier. Background Technology

[0002] The long-term care insurance system aims to provide care security for insured individuals who become chronically disabled due to old age, illness, or injury. According to relevant regulations from the National Healthcare Security Administration, the disability assessment and review period for long-term care insurance does not exceed three months (90 days), although some regions extend this period to 180 days based on their specific circumstances. Meanwhile, day care facilities, as an important form of long-term care insurance service, allow disabled individuals to receive care at designated institutions during the day and return home at night, and are also included in the benefit coverage.

[0003] Current disability assessment and care service verification typically rely on in-home visits, telephone follow-ups, paper documents, institutional service records, location tracking, or video spot checks. This model suffers from at least the following structural deficiencies in its actual implementation: First, a single assessment at a specific point in time is insufficient to reflect a long-term condition. On the day of the assessment, an applicant may exhibit higher or lower levels of activity than usual due to factors such as temporary medication use, assistance from relatives, or fluctuations in mental state, leading to an underestimation or overestimation of persistent disability; conversely, temporary disability may be mistakenly identified as long-term disability.

[0004] Second, there is a lack of continuous evidence of status between application acceptance and in-home assessment or review. Processing agencies or insurance companies often struggle to obtain data on daily care dependence, activity changes, service occurrences, and abnormal fluctuations during the waiting period. The existing system does not objectively record the applicant's daily disability status between application acceptance and in-home assessment, resulting in weak evidence for review.

[0005] Third, simple EVV (Electronic Visit Verification) can prove whether the caregiver has arrived at the service location, but it cannot verify whether the caregiver is the registered caregiver, whether the service content was actually provided, or whether there was actual interaction with the caregiver. Simple voiceprint authentication or access control systems can only achieve single identity verification, cannot form a chain of evidence of continuous service for 90 to 180 days, and are not combined with the verification of the authenticity of nursing services.

[0006] Fourth, while blockchain-based electronic medical records or nursing records can improve the credibility of data storage, they cannot solve the problems of verifying the authenticity of entry data and continuous disability. Although video surveillance or infrared home monitoring can identify activity status, they have limitations such as privacy sensitivity, high storage costs, and reliance on manual analysis. Furthermore, they cannot prevent institutions and applicants from colluding to forge service scenarios locally.

[0007] Fifth, while dedicated hardware nursing equipment solutions offer relatively objective data, their deployment and maintenance costs are high, and large-scale promotion is difficult, making it hard to cover the low-cost needs of home care and primary care daycare institutions. With the increasing aging population and the promotion of daycare models, the number of applicants continues to grow, making it difficult to achieve large-scale coverage using methods relying on manual home visits or on-site verification by institutions.

[0008] Therefore, there is an urgent need for a technical solution that can collect voice features and optional multi-source summaries within the authorized scope based on the existing mobile terminals of caregivers or family caregivers, form a continuous chain of evidence for 90 to 180 days after de-identification, and output review assistance suggestions to cover both home and day care scenarios, thereby fundamentally improving the accuracy, fairness and review efficiency of disability assessment for long-term care insurance. Summary of the Invention

[0009] In view of this, the purpose of this invention is to provide a low-cost, low-privacy-risk, continuously operable assessment system and method that can assist in the self-verification of disability status and the verification of service authenticity in long-term care insurance. This system does not directly replace manual assessment or insurance claim adjudication, but rather provides verifiable, traceable, and tiered authorization-based evidence summaries and risk diversion suggestions.

[0010] Long-term care disability status assessment system, the assessment system comprising: The application identity authorization module is used to generate authorization tokens and data collection permissions based on the identity relationships of the applicant, the person being cared for, the authorized caregiver, the caregiver, or the care institution. The mobile terminal sound acquisition module is used to collect the nursing staff's service statement voice, environmental acoustic events during the service process, and the voice response of the nursing staff through the mobile terminal during authorized service periods or preset observation periods. The data desensitization processing module is used to perform acoustic feature extraction, voice activity detection, service keyword extraction, acoustic event tag generation, and sensitive information masking on the sound data locally on the mobile terminal. Only feature vectors, event tags, and timestamps are uploaded, while the original audio is stored on the local edge device. The multi-source sensor enhancement module is used to selectively access one or more of the following: vital sign summary, desensitized video behavior tags, nursing equipment event summary, location or near-field verification data. When no data is accessed, the evaluation system operates independently based on the mobile terminal sound acquisition module. The minimum nursing evidence unit generation module is used to solidify the identity, authorization, time, location level, service statement summary, acoustic event summary, optional sensor summary, activities of daily living (ADL) dimension labels and verification information corresponding to a single nursing service or a single state observation into a minimum nursing evidence unit; The Disability Persistence Consistency Assessment Module is used to calculate the frequency of nursing dependence evidence, continuity of evidence, risk of missing evidence, consistency of activity distribution, and consistency of multi-source data in the ADL dimension based on multiple minimal nursing evidence units within an observation window of 90 to 180 days, and generate a Disability Persistence Consistency Score according to preset weights. The audit support decision-making module is used to output audit path suggestions based on the disability continuity consistency score, service authenticity risk marker, and data integrity. The hierarchical authorization output module is used to output anonymized evidence summaries and verifiable credentials that match the permissions of different authorized entities; The assessment report output module is used to extract key information from each data stream, generate a standard structured assessment report, and send it to the long-term care insurance management terminal.

[0011] Optionally, the application identity authorization module is associated with a server, which responds to the authorized subject identity identifier, authorization scope label, token validity period and token issuance timestamp, and automatically generates a dynamic authorization token in combination with anti-replay random number; the application identity authorization module also includes a nursing staff voiceprint feature template registration unit, which is used to collect the voiceprint feature vector of registered nursing staff and generate a voiceprint identity identifier.

[0012] Optionally, the mobile terminal sound acquisition module includes: Service statement voice acquisition unit, used to collect service statement voices from nursing staff; An environmental acoustic event detection unit is used to detect at least two of the following based on an acoustic event database: water flow sound, tableware collision sound, fabric friction sound, bed and chair movement sound, regular tapping sound, and nursing voice prompts. The voice response acquisition unit for cared-for individuals is used to collect the voice responses of cared-for individuals during the service process.

[0013] Optionally, the evaluation system further includes an acoustic event and service item matching unit, used to generate a list of expected acoustic events based on the service claim speech, and to match the detected acoustic event sequence with the list of expected acoustic events.

[0014] Optionally, the multi-source sensing enhancement module includes: The vital signs detection submodule is used to collect at least one statistical summary of heart rate, blood oxygen, sleep, body movement, and wearing status; The video behavior recognition submodule is used to generate human skeleton point coordinate sequences and action capability feature vectors at the edge, without uploading the original face image or complete video frame; The nursing equipment event submodule is used to receive event summaries output by intelligent nursing beds or other nursing equipment; The location verification submodule is used for near-field verification via at least one of GPS, base station, Wi-Fi, Bluetooth, QR code, or institutional virtual fence.

[0015] Optionally, the differentiated desensitization processing of the data desensitization processing module includes: extracting MFCC acoustic feature vectors and corresponding timestamps and event tags from the original sound signal locally on the mobile terminal; calculating statistical features from the original vital sign waveforms locally on the mobile terminal, and uploading the mean, standard deviation, and event frequency; performing skeletal point extraction, face blurring, and background replacement on the original video frames locally on the mobile terminal, and uploading only the skeletal point coordinate sequence and motion feature vectors; and generating Boolean values ​​and daily home time statistics summaries from the location coordinates using geofencing technology.

[0016] Optionally, the minimum unit of care evidence includes: an applicant identity identifier, an anonymous identifier of the person being cared for, an identifier of the collection role, an authorization token hash, a timestamp, a location level, a service statement summary, an acoustic event summary, a behavior tag summary, a vital signs summary, a device event summary, a service item code, an ADL dimension tag, an anomaly description, a digital signature, a hash checksum, and an index of the previous minimum unit of care evidence.

[0017] Optionally, the disability persistence consistency assessment module performs continuity, missing, and abnormal data supplementation analysis on multiple minimal nursing evidence units within a 90- to 180-day observation window, calculates the frequency of nursing dependence evidence, number of days of missing nursing services, service time distribution pattern, activity distribution consistency, and multi-source data consistency for each ADL dimension, and generates a disability persistence consistency score according to preset weights; the preset weights are configured according to regional policies or insurance terms.

[0018] Optionally, the audit support decision-making module may output quick audit suggestions, routine audit suggestions, remote certificate supplementation suggestions, sampling home visit suggestions, or key verification suggestions based on the disability continuity consistency score, nursing service authenticity score, insurance clause matching degree, and data integrity.

[0019] Optionally, the hierarchical authorization output module outputs a service completion summary and anomaly alert to the family member; outputs caregiver service records and abnormal work orders to the nursing institution; outputs service item codes and minimum nursing evidence unit hashes to the medical insurance agency; outputs ADL evidence, waiting period satisfaction, and claim confidence to the commercial insurance agency; and outputs a key on-site verification list and long-term trend report to the assessment personnel.

[0020] Optionally, the evaluation system stops collecting new data after the user revokes authorization, and retains, freezes, anonymizes, or deletes existing evidence according to audit requirements; the evaluation system performs local encryption caching, sequence number continuity recording, and retransmission integrity verification on the data collected during the network outage.

[0021] Optionally, the assessment system further includes an abnormal forgery cost identification module for identifying at least one of the following abnormal patterns: multiple caregiver accounts simultaneously serving the same applicant at different locations; long-term inconsistency between service items and vital signs or equipment events; a large number of minimal care evidence units being recorded before and after the application; and service statement voices being highly templated semantically and lacking acoustic events during the service process.

[0022] Optionally, the evaluation system further includes a voiceprint-location coupling verification unit, configured to: in the nursing scenario of a day care facility, compare the real-time voiceprint features of the caregiver with the pre-registered voiceprint template, simultaneously verify whether the caregiver's location data is within the virtual fence area of ​​the facility, and verify that the voiceprint activity detection during the service process is passed. When the three are coupled consistently, a voiceprint-location coupling verification pass identifier is generated.

[0023] Optionally, the evaluation system further includes a multi-caregiver cross-voiceprint dialogue verification unit, configured to: when at least two registered caregivers are providing services together at the same evaluation time point, extract the voiceprint identifiers and service dialogue audio streams of each caregiver, confirm the consistency between the actual caregiver present and the registered identity through dialogue voiceprint separation and cross-comparison, and detect whether the dialogue content contains preset service interaction keywords to generate cross-voiceprint dialogue verification confidence.

[0024] Optionally, the evaluation system enables different auditing entities to verify the existence and completeness of evidence without obtaining the complete original data by using verifiable digest codes, hash check codes, or authorization tokens.

[0025] A long-term care disability status assessment method, applied to the assessment system described above, includes the following steps: Authorization tokens and data collection permissions are generated based on the identity relationships of the applicant, the person receiving care, the authorized caregiver, the caregiver, or the care institution. During authorized service periods or preset observation periods, mobile terminals are used to collect nursing staff's service declaration voice, environmental acoustic events during the service process, and the voice responses of the patients. The sound data is processed locally on the mobile terminal, including acoustic feature extraction, voice activity detection, service keyword extraction, acoustic event tag generation, and sensitive information masking. Only feature vectors, event tags, and timestamps are uploaded, while the original audio is stored on the local edge device. Selectively access one or more of the following as enhanced data sources: vital sign summaries, desensitized video behavior tags, nursing equipment event summaries, location or near-field verification data; The identity, authorization, time, location level, service statement summary, acoustic event summary, optional sensor summary, ADL dimension label and verification information corresponding to a single nursing service or a single status observation are solidified into the smallest unit of nursing evidence; Within an observation window of 90 to 180 days, the frequency of nursing dependence evidence, continuity of evidence, risk of missing evidence, consistency of activity distribution, and consistency of multi-source data in the ADL dimension were calculated based on multiple minimal nursing evidence units, and a disability persistence consistency score was generated according to preset weights. Based on the aforementioned disability continuity consistency score, service authenticity risk marker, and data integrity output audit path suggestion; Output de-identified evidence summaries and verifiable credentials that match the permissions of different authorized entities; Key information from each data stream is extracted to generate a standard structured assessment report, which serves as the basis for review by the long-term care insurance management terminal.

[0026] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the evaluation method described above.

[0027] An electronic device includes a processor, a memory, and a communication interface, wherein the memory stores a program that, when executed by the processor, implements the evaluation method described above.

[0028] The beneficial effects that this invention can produce include: 1. This invention uses mobile terminal sound acquisition as the default core path, breaking through the dependence on traditional dedicated hardware. It can achieve low-cost acquisition of service declaration voice, environmental acoustic events, and voice responses of the cared-for using the existing smartphones of caregivers or their families. Through local acoustic feature extraction, event tag generation, and sensitive information masking, only anonymized features are uploaded, and the original audio is stored locally, ensuring the effectiveness of the assessment while maximizing privacy protection. It solidifies a single service into the smallest unit of nursing evidence and forms a continuous chain of evidence within a 90-180 day observation window, realizing the objectivity, quantification, and temporalization of disability status assessment, fundamentally improving the accuracy and fairness of long-term care insurance disability assessment.

[0029] 2. This invention transforms nursing service records from "post-event completion" to "automatic process generation" through a minimum nursing evidence unit generation module and a 90-180 day disability continuity consistency assessment module. It also transforms disability assessment from "single-time node judgment" to "90-180 day continuous trend analysis." Through multi-dimensional analysis of nursing dependence evidence frequency, service absence days, activity distribution consistency, multi-source data consistency, anomaly credibility, and long-term absence risk, it provides an objective, continuous, and verifiable chain of evidence for long-term care insurance benefit determination and commercial insurance claims.

[0030] 3. This invention achieves precise allocation of review resources and hierarchical protection of privacy data through a review auxiliary decision-making module and a hierarchical authorization output module. It outputs auxiliary suggestions based on the disability continuity consistency score, service authenticity and terms matching degree, so as to ensure review efficiency while minimizing the risk of privacy leakage and realizing a trust and mutual recognition mechanism with multi-party participation.

[0031] 4. This invention significantly reduces the labor costs and scheduling pressure of long-term care insurance disability assessment through coordinated control of dynamic permission authorization management, full-process intelligent data monitoring and hierarchical verification strategy, effectively alleviating the industry pain point of scarce professional assessment personnel resources; and through intelligent replacement and hierarchical resource optimization, it adapts to the continuously growing assessment business volume in the context of aging, and realizes the large-scale and standardized implementation of assessment business.

[0032] 5. This invention combines data privacy anonymization, multi-dimensional anomaly fraud prevention, time consistency verification, and standardized report output to build a full-process compliance risk control system, identify suspected abnormal risk patterns and generate review clues to ensure the authenticity and credibility of assessment results; at the same time, it realizes data traceability and responsibility definition throughout the assessment process, takes into account personal privacy security and business compliance, and greatly improves the risk control capabilities and institutional credibility of long-term care insurance disability assessment.

[0033] 6. This invention ensures data integrity and authenticity under network instability or interruption conditions through local encrypted caching, serial number continuity recording, and retransmission integrity verification mechanisms. It effectively identifies fraudulent activities such as post-event centralized data entry and improves the robustness and reliability of the system in complex network environments.

[0034] 7. This invention, through a dynamic authorization revocation mechanism and hierarchical data processing capabilities, ensures the continuity and compliance of the evaluation process while protecting users' privacy and autonomy, thus achieving a balance between privacy protection and business needs.

[0035] 8. This invention specifically designs a voiceprint-location coupling verification and multi-caregiver cross-voiceprint dialogue verification mechanism for day care facility scenarios. As an optional enhancement function, it forms a three-dimensional verification framework of "biometric features + geographical location + interpersonal mutual verification". This makes the coordination cost and technical difficulty of collusion to defraud insurance increase exponentially, while the verification cost of genuine services remains at an extremely low level, realizing a virtuous cycle incentive structure of "low burden for genuine services and high cost for fraudulent services". Attached Figure Description

[0036] Figure 1 This is a flowchart of the long-term care disability status assessment method of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] This invention provides a long-term care disability status assessment system, which includes an application identity authorization module, a mobile terminal sound acquisition module, a data desensitization processing module, a multi-source sensor enhancement module, a minimum care evidence unit generation module, a disability continuity consistency assessment module, an audit auxiliary decision-making module, a hierarchical authorization output module, and an assessment report output module.

[0039] The application identity authorization module sets up related authorization units and several authorized units based on the task chain execution process. The authorization unit issues and / or revokes dynamic authorization tokens based on timestamps and anti-replay random numbers to several authorized units within consecutive assessment time nodes according to management permissions, so as to configure the operation permissions of the corresponding data flow. This realizes fine-grained permission control of the data flow of the entire long-term care insurance assessment process, distinguishes the operation responsibilities of the management end, the verification end, and the audit end, avoids the risk of unauthorized data access and tampering, and supports the collaborative participation of multiple entities in the assessment process, gets rid of the resource limitations of a single human assessor, and improves the overall operational efficiency of the assessment process.

[0040] The mobile terminal audio acquisition module is used to collect service statement voices from caregivers, environmental acoustic events during the service process, and the voice responses of patients via mobile terminals during authorized service periods or preset observation periods. This enables the collection of disability status data using low-cost mobile terminals as the core during the verification period, retaining complete dynamic status records and completely solving the problem of insufficient verification evidence. The audio acquisition module includes a service statement voice acquisition unit, an environmental acoustic event detection unit, and a patient voice response acquisition unit. The service statement voice acquisition unit is responsible for collecting the service statement voices and daily interaction voices from caregivers, providing a data foundation for service authenticity verification. The environmental acoustic event detection unit detects at least two of the following based on an acoustic event database: water flow sounds, clinking tableware sounds, regular patting sounds, fabric rubbing sounds, bed and chair movement sounds, and nursing voice prompts. It cross-compares these with the declared service items, verifying service authenticity through physical acoustic fingerprinting. The patient voice response acquisition unit collects the patient's voice responses during the service process to assess their state of consciousness, communication ability, and degree of disability.

[0041] The data anonymization module performs differentiated anonymization processing on the collected data based on the sensitivity of each original sensor data. Specifically, this includes: extracting MFCC acoustic feature vectors and corresponding timestamps and event tags from the original sound signals locally, with the original audio files stored on a local edge device; calculating statistical features from the original vital sign waveforms locally, uploading the mean, standard deviation, and event frequency; performing skeletal point extraction, face blurring, and background replacement on the original video frames locally, uploading only the skeletal point coordinate sequence and motion feature vectors; and generating Boolean values ​​and daily home time statistics summaries for location coordinates using geofencing technology. In this process, the system only stores the anonymized feature vectors, timestamps, event tags, and various statistical summaries when uploading data, without storing any intermediate results that can be reconstructed from the original sensor data. This achieves both privacy data anonymization and feature data retention, balancing data security and evaluation effectiveness, complying with the privacy protection requirements for long-term care insurance participants, mitigating the risk of data leakage, and ensuring the compliant operation of the system.

[0042] The multi-source sensor enhancement module selectively accesses one or more of the following: vital sign summaries, desensitized video behavior tags, nursing equipment event summaries, location, or near-field verification data. When no such data is accessed, the system operates independently based on the mobile terminal's sound acquisition module. The multi-source sensor enhancement module includes a vital sign detection submodule, a video behavior recognition submodule, a nursing equipment event submodule, and a location verification submodule. The vital sign detection submodule collects at least one statistical summary of heart rate, blood oxygen, sleep, body movement, and wearing status. The video behavior recognition submodule generates a sequence of human skeletal point coordinates and a movement capability feature vector at the edge, without uploading the original face image or complete video frame. The nursing equipment event submodule receives event summaries output by the smart nursing bed or other nursing equipment. The location verification submodule performs near-field verification via at least one of GPS, base station, Wi-Fi, Bluetooth, QR code, or institutional virtual fence.

[0043] The minimum nursing evidence unit generation module is used to solidify the identity, authorization, time, location level, service statement summary, acoustic event summary, optional sensor summary, activities of daily living (ADL) dimension tags, and verification information corresponding to a single nursing service or state observation into a minimum nursing evidence unit. The minimum nursing evidence unit includes: applicant identity identifier, patient anonymity identifier, data collection role identifier, authorization token hash, timestamp, location level, service statement summary, acoustic event summary, behavioral tag summary, vital signs summary, device event summary, service item code, ADL dimension tag, anomaly description, digital signature, hash checksum, and an index of the previous minimum nursing evidence unit. A chained index structure ensures the immutability and traceability of the evidence data.

[0044] The Disability Persistence Consistency Assessment module is used to calculate the frequency of nursing dependence evidence, continuity of evidence, risk of missing evidence, consistency of activity distribution, and consistency of multi-source data for each ADL dimension based on multiple minimal nursing evidence units within a 90- to 180-day observation window, and generate a disability persistence consistency score according to preset weights. This module performs continuity, missing evidence, and anomalous data addition analysis on multiple minimal nursing evidence units within the 90- to 180-day observation window, calculating the frequency of nursing dependence evidence, number of days of missing nursing services, service time distribution patterns, consistency of activity distribution, and consistency of multi-source data for each ADL dimension, and generating a disability persistence consistency score according to preset weights; the preset weights are configured based on regional policies or insurance terms.

[0045] The audit support decision-making module is used to output audit path suggestions based on the disability continuity consistency score, service authenticity risk markers, and data integrity. This module outputs suggestions for rapid auditing, routine auditing, remote document supplementation, sampling home visits, or key verification based on the disability continuity consistency score, nursing service authenticity score, insurance clause matching degree, and data integrity.

[0046] The hierarchical authorization output module is used to output anonymized evidence summaries and verifiable credentials that match the permissions of different authorized entities. Specifically, it outputs service completion summaries and anomaly alerts to family members; it outputs caregiver service records and abnormal work orders to nursing institutions; it outputs service item codes and the smallest nursing evidence unit hash to medical insurance agencies; it outputs ADL evidence, waiting period satisfaction, and claim confidence to commercial insurance companies; and it outputs a key on-site verification list and long-term trend reports to assessment personnel.

[0047] The assessment report output module is used to extract key information from each data stream to generate a standard structured assessment report and send it to the long-term care insurance management terminal. This provides a complete, traceable, and verifiable basis for the long-term care insurance review work, enabling reviewers to quickly and accurately determine the applicant's disability status and improve the accuracy and credibility of the review results.

[0048] Furthermore, the identity authorization module is linked to a server. The server responds to the authorized entity's identity identifier, authorization scope label, token validity period, and token issuance timestamp configured in the authorization unit, and automatically generates a dynamic authorization token using an anti-replay random number. This prevents security risks such as data replay attacks and unauthorized access, ensuring the authenticity and integrity of the disabled person's privacy data and assessment data. Specifically, the identity authorization module adds an authorized unit based on the authorized entity's identity identifier configured in the authorization unit and associates it with a corresponding dynamic authorization token. This allows for flexible deletion and dynamic expansion of assessment participants, effectively incorporating caregivers, supervisors, and on-site verification personnel into the assessment process. The authorization unit dynamically adjusts the token validity period to choose to issue and / or revoke dynamic authorization tokens to authorized units. Authorized units interact with the mobile terminal's sound acquisition module based on the received dynamic authorization token, enabling data flow access. This access is dynamically activated and deactivated throughout the assessment cycle, such as only granting data interaction permissions during valid assessment periods, reducing invalid data transmission and system energy consumption.

[0049] Furthermore, the authorization scope label includes operation permissions for one or more data streams among the surrounding sound data, vital sign data, video data, and location data of disabled individuals. This enables differentiated permission control for different types of perceived data, granting data operation permissions as needed and maximizing the protection of disabled individuals' privacy. Specifically, the token validity period is configured to automatically renew according to preset assessment time nodes (e.g., daily / weekly / monthly as a complete assessment time node). Authorized units can revoke the token validity period of one or more authorized units at any assessment time node, achieving automatic renewal and dynamic revocation of permissions, adapting to continuous time-series assessment scenarios without manual intervention.

[0050] Furthermore, the system also includes a voiceprint-location coupling verification unit and a multi-caregiver cross-voiceprint dialogue verification unit. The voiceprint-location coupling verification unit is configured to: in a daycare facility setting, compare the real-time voiceprint features of caregivers with a pre-registered voiceprint template; simultaneously verify whether the caregiver's location data is within the facility's virtual fence area; and verify that the voiceprint activity detection during the service process passes. When all three are consistent, a voiceprint-location coupling verification pass identifier is generated. This unit is specifically designed for daycare settings, effectively identifying high-risk fraud patterns such as "remote substitute services" and "institutional collusion for insurance fraud" through triple coupling of voiceprint, location, and activity.

[0051] The multi-caregiver cross-voiceprint dialogue verification unit is configured to: when at least two registered caregivers are providing services simultaneously at the same assessment time point, extract the voiceprint identifiers and service dialogue audio streams of each caregiver; confirm the consistency between the actual caregivers present and their registered identities through voiceprint separation and cross-comparison; and detect whether the dialogue content contains preset service interaction keywords to generate cross-voiceprint dialogue verification confidence. This unit utilizes the natural interaction scenario of multiple caregivers present simultaneously to construct an unforgeable "interpersonal mutual verification" evidence chain through voiceprint cross-verification, significantly improving the reliability of service authenticity verification.

[0052] Furthermore, the verification logic of the voiceprint-location coupling verification unit includes three necessary steps: (a) Voiceprint consistency comparison: Extracting the voiceprint features of the nurse's declared speech during the service process, calculating the similarity score with the pre-registered voiceprint template, and determining that it is the same person if it is higher than the dynamic voiceprint threshold (e.g., 0.75); (b) Location consistency verification: Confirming that the nurse's current location is within the virtual fence area registered by the day care facility through at least two combinations of GPS positioning, Wi-Fi / Bluetooth near-field verification, or facility location QR code scanning; (c) Voiceprint activity detection: Randomly generating a 4-6 digit string at the start of the service and requiring the nurse to read it aloud, and using a live voiceprint detection algorithm (detecting natural breathing pauses, speech rate changes, and spectral dynamic features in the speech) to exclude recording replay attacks. Only when (a), (b), and (c) are all satisfied simultaneously will a valid service verification identifier for that evaluation time point be generated.

[0053] Furthermore, the verification logic of the multi-nursing staff cross-voiceprint dialogue verification unit includes four steps: (i) Dialogue voiceprint separation: Speaker separation is performed on the environmental audio stream within the same service period (using a speaker separation algorithm based on deep clustering), and voiceprint fragments of each speaker are extracted; (ii) Cross-identity comparison: The separated voiceprint fragments are compared one by one with the voiceprint template library of registered nurses on the same day to identify the actual nurses present. If the proportion of non-registered staff voiceprints exceeds 20% or registered nurses' voiceprints are missing, an anomaly is marked; (iii) Service keyword detection: Automatic speech recognition (ASR) is performed on the dialogue content to detect whether it contains preset service interaction keywords, such as "open your mouth", "eat slowly", "hold on tight", "is the water temperature suitable", "cooperate with deep breathing", etc. If the keyword density is less than 0.5 times per minute, it indicates insufficient service interaction; (iv) Anomaly marking: If the proportion of non-registered staff voiceprints exceeds the preset ratio, or registered nurses' voiceprints are missing, or the service interaction keyword density is less than the threshold, a high-risk anomaly mark is generated and an early warning is triggered.

[0054] Furthermore, the registration process for the nursing staff voiceprint feature template registration unit includes: Step S101, the nursing staff submits their qualification certificate number and identity information via a mobile terminal APP; Step S102, the system generates a random number sequence (such as "3-8-1-9-5") and prompts the nursing staff to read it aloud; Step S103, the mobile terminal records the voice locally and extracts MFCC features; Step S104, a preset dimension voiceprint feature vector is generated through the voiceprint feature extraction model; Step S105, the voiceprint feature vector is bound to the qualification certificate number and the mobile terminal device fingerprint and stored on the server; Step S106, a voiceprint identity identifier is generated and fed back to the mobile terminal. This registration process only requires the nursing staff's everyday smartphone, without the need for dedicated hardware, and a single registration takes approximately 30 seconds. It can also be completed in a quiet environment, placing a very low burden on the nursing staff.

[0055] Furthermore, the standard structured assessment report includes one or more of the following key information: ① the total collection time and effective collection time of each raw sensor data; ② the time-series curve of disability status changes and the standardized disability score corresponding to each assessment time node; ③ abnormal event records, including one or more of the following: warning event type, assessment time node, duration, and data source; ④ standardized disability score and confidence value within each assessment time node; ⑤ data integrity assessment, including at least the percentage of effective collection time of each raw sensor data to the total collection time; ⑥ suggestions for optimizing the disability assessment waiting verification cycle; ⑦ confidence level verified through voiceprint-location coupling verification via identification and cross-voiceprint dialogue among multiple caregivers.

[0056] like Figure 1As shown, the present invention also provides a method for assessing long-term care disability status, applied to the aforementioned assessment system. The assessment method includes the following steps: generating an authorization token and data collection permissions based on the identity relationships of the applicant, the person being cared for, the authorized caregiver, the caregiver, or the nursing institution; during the authorized service period or a preset observation period, collecting the service statement voice of the caregiver, the acoustic events of the service process environment, and the voice response of the person being cared for via a mobile terminal; performing acoustic feature extraction, voice activity detection, service keyword extraction, acoustic event tag generation, and sensitive information masking on the local mobile terminal, uploading only the feature vector, event tags, and timestamps, while storing the original audio on a local edge device; selectively accessing one of the following: vital sign summary, desensitized video behavior tags, nursing equipment event summary, location, or near-field verification data. Multiple data sources can be used to enhance the data flow; the identity, authorization, time, location level, service statement summary, acoustic event summary, optional sensor summary, ADL dimension labels, and verification information corresponding to a single nursing service or single status observation are solidified into the smallest unit of nursing evidence; within an observation window of 90 to 180 days, the frequency of nursing dependence evidence, evidence continuity, risk of missing data, consistency of activity distribution, and consistency of multi-source data in the ADL dimension are calculated based on multiple units of nursing evidence, and a disability continuity consistency score is generated according to preset weights; based on the disability continuity consistency score, service authenticity risk markers, and data integrity, audit path suggestions are output; de-identified evidence summaries and verifiable credentials matching the permissions of different authorized entities are output; key information from each data stream is extracted to generate a standard structured assessment report as the audit basis for the long-term care insurance management terminal.

[0057] Furthermore, the preset assessment strategy is set as follows: dynamically adjusting the disability assessment waiting verification cycle optimization suggestion based on the scoring coefficient of the standardized disability score; and / or dynamically adjusting the on-site verification frequency of the corresponding authorized unit based on the scoring coefficient of the standardized disability score; and / or, based on voiceprint-location coupling verification, verifying confidence through identification and cross-voiceprint dialogue of multiple caregivers, classifying and marking the service authenticity in the day care institution nursing scenario, the classification and marking include rapid review suggestion, routine review suggestion, remote supplementary certification suggestion, sampling on-site suggestion, or key verification suggestion.

[0058] Furthermore, the system performs local encryption caching, sequence number continuity recording, and retransmission integrity verification on the data collected during the network outage.

[0059] Furthermore, the system also includes an abnormal forgery cost identification module for identifying at least one of the following abnormal patterns: multiple caregiver accounts simultaneously serving the same applicant in different locations; long-term inconsistency between service items and vital signs or equipment events; a large number of minimal care evidence units being recorded before and after the application; and service statement voices being highly templated semantically and lacking acoustic events during the service process.

[0060] Furthermore, the system enables different auditing entities to verify the existence and integrity of evidence without obtaining the complete original data by using verifiable digest codes, hash check codes, or authorization tokens.

[0061] Example 1: Using only two ordinary mobile phones to complete 90 days of self-certification at home Simulated scenario configuration for disabled individuals: Applicant A, 75 years old, suffers from limb movement impairment due to the sequelae of a stroke and submits a disability assessment application to the local long-term care insurance agency, entering a 90-day waiting verification period. Applicant A has a full-time caregiver, B, in their family. The system configures a virtual fence for the home care scenario, extending to a radius of 200 meters from Applicant A's residential address.

[0062] Step 1: Identity Authorization and Voiceprint Registration (Day 1). Applicant A completes real-name identity verification through the assessment system client, becoming the system's primary authorizer; granting entrusted authorizer permissions to Caregiver B, with the authorization scope limited to "surrounding sound data". Caregiver B completes voiceprint feature template registration via their smartphone: opening the nursing APP, entering the qualification certificate number, the system generates a random number sequence "4-2-7-9-1", which Caregiver B reads aloud. The system extracts the voiceprint feature vector and binds it to their device fingerprint, generating a voiceprint identity identifier. The system generates a corresponding dynamic authorization token and pushes it to Caregiver B's mobile client.

[0063] Step 2: Continuous collection of multi-source data (day 1 to day 90). (1) Sound collection on mobile phone: The sound collection program runs in the background of the applicant A's mobile phone client to continuously collect the home environment sound. After extracting the acoustic feature vector locally, it is encrypted and uploaded. The collection program will automatically stop when the applicant A revokes the authorization or waits for the verification period to end. When the caregiver B provides services at home, the APP will automatically trigger voiceprint verification: a random number string will be generated and required to be read aloud. After the voiceprint activity detection, the service start time will be recorded. During the service, environmental acoustic data will be collected to detect the collision sound of tableware, the sound of water flow, the sound of fabric friction, etc., and cross-compare with the declared service items. (2) Collection of vital signs data: The heart rate / blood oxygen monitoring bracelet worn by the applicant A is connected to the system through Bluetooth BLE. It generates statistical features such as the average heart rate, average blood oxygen, and number of activity steps every day and uploads them to the cloud in encryption. The mattress-type breathing sensor is connected through Wi-Fi and generates statistical features such as the average breathing frequency, the frequency of turning over, and the number of times the person gets out of bed every day. (3) Video pose recognition data collection: The home camera performs skeletal point extraction, face blurring and background replacement processing locally, and generates action feature vectors such as average sitting and standing time, activity frequency and bed rest time every day. The original video frames do not leave the local device. (4) Location data collection: The applicant A's mobile client generates a daily summary of home time statistics and uploads it to the cloud.

[0064] Step 3: Continuous Status Analysis and Fraud Prevention Verification (Continuously running from day 15). After establishing the applicant A's personal behavioral baseline for the first 14 days, the system begins continuous status analysis: ① Generate daily disability status scores and construct a time-series curve of disability status; ② Cross-validate the scores from various data sources to confirm the consistency of the data; ③ Calculate the Time Dimension Consistency Score (LSCS). Applicant A's LSCS remains consistently high, indicating that their disability status is highly stable over time and no abnormal warnings have been triggered. Caregiver B's voiceprint-location coupling verification passed for each service, and the acoustic events during the service process matched the declared items with a accuracy rate higher than 85%.

[0065] Step Four: Evaluation Report Generation and Optimization Suggestions for Verification Cycle (Day 60). On day 60, the system generates a phased evaluation report, showing: ① Valid data collection days: 60 days, accounting for 67% of the waiting verification period; ② Disability scores across all dimensions are consistently within the severe disability range (standardized disability score ≥ 75); ③ High consistency score in multi-source data cross-validation (confidence value ≥ 0.90); ④ High LSCS score (≥ 0.85); ⑤ Voiceprint-location coupling verification continues to pass; ⑥ Overall confidence rating is "High". The system sends "Advance Evaluation Suggestions" and "On-site Verification Priority Optimization Suggestions" to the handling agency, allowing the agency to schedule on-site evaluations in advance, shortening the actual waiting period from 90 days to 60 days.

[0066] Example 2: Matching Service Claim Voice with Ambient Acoustic Events Simulated scenario configuration for disabled individuals: Applicant B, 68 years old, moderately disabled, cared for by family members. Caregiver C declares the service via a mobile app before each service session: "Today, I will assist with eating, toileting, cleaning, turning over, and back patting, estimated to take 60 minutes."

[0067] Step 1: Service Declaration Collection and Expected Acoustic Event List Generation. The system performs automatic speech recognition (ASR) on the voice of nurse C's declaration, extracts keywords "eating," "toileting," "turning over," and "patting the back," and generates an expected acoustic event list: during eating, the system should detect the sound of clattering utensils and chewing; during toileting, the system should detect the sound of running water and flushing; during turning over and patting the back, the system should detect the sound of regular patting and fabric rubbing.

[0068] Step Two: Acoustic Event Detection and Matching During Service. During the service process, the environmental acoustic event detection unit detects acoustic events in real time: during eating, it detects the sounds of clattering utensils (matched) and chewing (matched); during toilet use, it detects the sounds of running water (matched) and flushing (matched); during turning over and patting the back, it detects the sounds of regular patting (matched) and fabric rubbing (matched). The system calculates a matching degree of 92%, marking it as "High Service Authenticity".

[0069] Step 3: Example of anomaly matching. Suppose that in a certain service, caregiver C declares "assistance with bathing", but no water flow sound, brushing sound or wet area activity tag is detected during the service, and no explanation of the anomaly is submitted. The system marks this service as "missing acoustic event", with a matching degree of less than 60%, generates a "recommend remote supplementary certificate" prompt, and increases the priority of manual review for this service.

[0070] Example 3: Event Enhancement for Optional Life Monitoring and Care Devices Simulated scenario configuration for disabled persons: Applicant C, 82 years old, severely disabled, bedridden for a long time, with a smart nursing bed and heart rate monitoring bracelet at home.

[0071] Step 1: Multi-source data access authorization. Applicant C authorizes the following system accesses: nurse D's mobile phone audio data, heart rate monitoring bracelet vital sign data, and smart nursing bed equipment event data. Video camera access is not permitted (due to family privacy concerns).

[0072] Step Two: Device Event Summary Collection. The smart nursing bed reports daily: bedtime, number of back raises, number of leg raises, bed exit events, toilet use events, cleaning events, and drying events. The heart rate monitoring wristband reports daily: average heart rate, average blood oxygen saturation, number of steps taken, and abnormal heart rate markers.

[0073] Step 3: MEU Generation and Cross-Validation. The system generates MEUs daily, including: sound acquisition summary (service declaration, acoustic event), vital signs summary (heart rate, blood oxygen statistics), and device event summary (bedside, toileting, cleaning). When a caregiver declares "assisting with toileting and cleaning," but the smart nursing bed does not report a toileting or cleaning event, and no water flow sound is detected in the acoustic event, the system marks it as "multi-source inconsistency," reduces the credibility of the MEU, and generates a verification clue.

[0074] Example 4: Voiceprint Cross-verification and Collusion to Commit Insurance Fraud in Day Care Institutions Simulated scenario configuration for disabled individuals: Applicant Ding, 68 years old, was initially assessed as moderately disabled and is eligible for long-term care insurance benefits. He / She has selected "Kangyi Jiayuan" as his / her designated day care facility. Caregivers Wu and Ji are registered caregivers at the facility, responsible for applicant Ding's day care. The system configures a virtual fence around the facility for the day care scenario, extending to a radius of 50 meters from the "Kangyi Jiayuan" location, and records the facility's Wi-Fi fingerprint and Bluetooth beacon information.

[0075] Step 1: Institutional Voiceprint Registration and Scheduling Binding (Day 1). Nurses E and F both registered their voiceprint feature templates through the institution's management system upon joining, generating their own voiceprint feature vectors and binding them to their qualification certificate numbers. The system synchronizes nurses' attendance plans daily from the institution's scheduling system, associating applicant D with the nurses scheduled for that day.

[0076] Step Two: Voiceprint-Location Coupling Verification for Daycare Service (Daily Daytime Service Hours). Every day at 8:30 AM, after caregiver E arrives at the facility, they open the care app. The system automatically triggers: (a) GPS positioning comparison with the facility's virtual fence; coordinate deviation <30 meters; (b) Wi-Fi fingerprint matching to confirm connection to the "Kangyi Home" internal network; (c) Voiceprint activity detection. The system randomly generates the number string "5-1-8-3". After caregiver E reads it aloud, the voiceprint similarity score is 0.89, higher than the threshold of 0.75, thus verification is successful. The system records the service start timestamp and generates a valid service verification identifier.

[0077] Step 3: Cross-voiceprint verification by multiple caregivers (during multi-person collaborative service hours). At 10:00 AM, caregivers had already joined to assist applicant Ding with rehabilitation training. The ambient audio stream was processed by a speaker separation algorithm to extract the voiceprint fragments of caregivers Wu and Ji, which were compared with the scheduling registration database, and the identities of both were confirmed. The dialogue content ASR recognition showed that caregiver Wu said, "Cooperate with deep breathing, now I will pat your back," and caregiver Ji responded, "Okay, I'll hold your shoulders." The system detected keywords such as "cooperate with deep breathing" and "pat your back," and the confidence level of the cross-voiceprint dialogue verification was 0.92.

[0078] Step Four: Identification and Early Warning of Insurance Fraud in Other Locations (Anomaly Simulation). Assume applicant D is actually traveling in another location. To maintain his long-term care insurance benefits, the institution arranges for another person to impersonate caregiver E's account to clock in at the institution. Because the impersonator fails the voiceprint consistency comparison (voiceprint similarity score below 0.40) and voiceprint activity detection (failing to pass the dynamic verification of the natural spectrum of random number readings), the system marks it as "voiceprint-location coupling verification failed" on day 1, generating a high-risk warning and pushing it to the handling agency. Even if the institution attempts a replay attack using pre-recorded audio from caregiver E, the voiceprint activity detection process fails to match the spectral characteristics of the real-time random number string, triggering a "suspected audio replay attack" warning. Based on this, the handling agency adds applicant D to the key verification list, arranging a surprise visit or video verification, effectively breaking the chain of collusion to commit insurance fraud.

[0079] Example 5: Network Disconnection Retransmission and Authorization Revocation The home network of disabled person E was interrupted for 3 days due to a fault. During this period, the system performed the following operations: (1) Local encrypted caching: Each sensor unit encrypted and stored the collected data in the local device and recorded the continuity of the local sequence number; (2) Network outage marking: The system automatically marked the data during the network outage as "to be retransmitted"; (3) Retransmission after network recovery: After the network is restored, the system automatically detects the continuity of the sequence number of the local cached data and marks the data with missing sequence numbers as "potential risk of centralized retransmission"; (4) Integrity verification: After the retransmission is completed, the system calculates the hash value of the retransmitted data and compares it with the hash value of the local cache to verify the integrity of the data; (5) Anomaly warning: If the amount of data during the network outage is significantly higher than the normal level, or the retransmission time is concentrated in a short period of time after the network is restored, the system generates a "risk of centralized retransmission" warning.

[0080] The daughter of disabled person F, C, applied to revoke the authorization for location data collection for personal reasons. The system performs the following operations: (1) Immediately stop location data collection: After the authorization is revoked, the daughter C's mobile client stops uploading location data; (2) Historical data processing: The historical location data that has been collected is retained and frozen in accordance with audit requirements and will no longer be used for new assessment calculations; (3) Assessment adjustment: The system recalculates the disability continuity consistency score, removes the location data dimension, and conducts the assessment only based on the remaining data sources; (4) Notification mechanism: The system automatically sends an authorization change notification to the handling agency, explaining the adjustment of the assessment basis.

[0081] Example 6: Audit and Triage Output Based on the data within the 90-day observation window, the system automatically generates review and triage suggestions: Green (Low-risk path recommendation): All verification dimensions passed, voiceprint-location coupling verification continued to pass, no abnormal signals, high data integrity, high LSCS score, "fast review recommendation" path is recommended to reduce the frequency of on-site verification.

[0082] Yellow (Medium-risk path recommendation): A slight anomaly exists in a single dimension (such as a low density of acoustic events in a service but not completely missing, or the voiceprint verification score is at the critical value). Enter the "enhanced monitoring" mode. Subsequent services will have an increased frequency of random spot checks. The "routine review recommendation" or "remote certificate supplementation recommendation" path is recommended.

[0083] Red (High-risk path suggestion): Multiple abnormalities may be superimposed, or trigger rules such as voiceprint verification failure, suspected recording replay attack, excessive proportion of non-registered personnel's voiceprints, serious location verification failure, and a large number of MEUs being added in a concentrated manner. It is recommended to follow the "key verification suggestion" path, automatically initiate the manual review process, notify the insurance company / social security agency, and suggest suspending claims pending manual verification.

[0084] This tiered early warning system reduces the cost of self-verification in real nursing scenarios to near zero (nursing staff only need to provide normal service and speak naturally), while the cost of multi-dimensional forgery in fraudulent scenarios increases significantly with the number of verification dimensions: forgers must simultaneously deal with GPS location forgery, Wi-Fi fingerprint forgery, voiceprint feature forgery, natural dialogue content forgery, acoustic event sequence forgery, vital sign data forgery, and video posture forgery, and each forged data must maintain temporal consistency and logical self-consistency within a 90 to 180-day evaluation period. The technical difficulty and coordination cost are extremely high, thus significantly increasing the marginal cost of forgery without increasing the cost of real nursing care.

Claims

1. A long-term care disability status assessment system, characterized in that, The evaluation system includes: The application identity authorization module is used to generate authorization tokens and data collection permissions based on the identity relationships of the applicant, the person being cared for, the authorized caregiver, the caregiver, or the care institution. The mobile terminal sound acquisition module is used to collect the nursing staff's service statement voice, environmental acoustic events during the service process, and the voice response of the nursing staff through the mobile terminal during authorized service periods or preset observation periods. The data desensitization processing module is used to perform acoustic feature extraction, voice activity detection, service keyword extraction, acoustic event tag generation, and sensitive information masking on the sound data locally on the mobile terminal. Only feature vectors, event tags, and timestamps are uploaded, while the original audio is stored on the local edge device. The multi-source sensor enhancement module is used to selectively access one or more of the following: vital sign summary, desensitized video behavior tags, nursing equipment event summary, location or near-field verification data. When no data is accessed, the evaluation system operates independently based on the mobile terminal sound acquisition module. The minimum nursing evidence unit generation module is used to solidify the identity, authorization, time, location level, service statement summary, acoustic event summary, optional sensor summary, activities of daily living (ADL) dimension labels and verification information corresponding to a single nursing service or a single state observation into a minimum nursing evidence unit; The Disability Persistence Consistency Assessment Module is used to calculate the frequency of nursing dependence evidence, continuity of evidence, risk of missing evidence, consistency of activity distribution, and consistency of multi-source data in the ADL dimension based on multiple minimal nursing evidence units within an observation window of 90 to 180 days, and generate a Disability Persistence Consistency Score according to preset weights. The audit support decision-making module is used to output audit path suggestions based on the disability continuity consistency score, service authenticity risk marker, and data integrity. The hierarchical authorization output module is used to output anonymized evidence summaries and verifiable credentials that match the permissions of different authorized entities; The assessment report output module is used to extract key information from each data stream, generate a standard structured assessment report, and send it to the long-term care insurance management terminal.

2. The long-term care disability status assessment system according to claim 1, characterized in that, The application identity authorization module is associated with a server. The server responds to the authorized subject identity identifier, authorization scope label, token validity period and token issuance timestamp, and automatically generates a dynamic authorization token in combination with anti-replay random number. The application identity authorization module also includes a nursing staff voiceprint feature template registration unit, which is used to collect the voiceprint feature vector of registered nursing staff and generate a voiceprint identity identifier.

3. The long-term care disability status assessment system according to claim 1, characterized in that, The mobile terminal sound acquisition module includes: Service statement voice acquisition unit, used to collect service statement voices from nursing staff; An environmental acoustic event detection unit is used to detect at least two of the following based on an acoustic event database: water flow sound, tableware collision sound, fabric friction sound, bed and chair movement sound, regular tapping sound, and nursing voice prompts. The voice response acquisition unit for cared-for individuals is used to collect the voice responses of cared-for individuals during the service process.

4. The long-term care disability status assessment system according to claim 3, characterized in that, The evaluation system further includes an acoustic event and service item matching unit, which generates a list of expected acoustic events based on the service declaration speech and matches the detected acoustic event sequence with the list of expected acoustic events.

5. The long-term care disability status assessment system according to claim 1, characterized in that, The multi-source sensing enhancement module includes: The vital signs detection submodule is used to collect at least one statistical summary of heart rate, blood oxygen, sleep, body movement, and wearing status; The video behavior recognition submodule is used to generate human skeleton point coordinate sequences and action capability feature vectors at the edge, without uploading the original face image or complete video frame; The nursing equipment event submodule is used to receive event summaries output by intelligent nursing beds or other nursing equipment; The location verification submodule is used for near-field verification via at least one of GPS, base station, Wi-Fi, Bluetooth, QR code, or institutional virtual fence.

6. The long-term care disability status assessment system according to claim 1, characterized in that, The differentiated desensitization processing of the data desensitization processing module includes: extracting MFCC acoustic feature vectors and corresponding timestamps and event tags from the original sound signal locally on the mobile terminal; calculating statistical features from the original vital sign waveforms locally on the mobile terminal, and uploading the mean, standard deviation, and event frequency; performing skeletal point extraction, face blurring, and background replacement on the original video frames locally on the mobile terminal, and uploading only the skeletal point coordinate sequence and motion feature vectors; and generating Boolean values ​​and daily home time statistics summaries from the location coordinates using geofencing technology.

7. The long-term care disability status assessment system according to claim 1, characterized in that, The minimum unit of care evidence includes: applicant identity identifier, anonymous identifier of the person being cared for, data collection role identifier, authorization token hash, timestamp, location level, service statement summary, acoustic event summary, behavioral tag summary, vital signs summary, device event summary, service item code, ADL dimension tag, anomaly description, digital signature, hash checksum, and index of the previous minimum unit of care evidence.

8. The long-term care disability status assessment system according to claim 1, characterized in that, The disability persistence consistency assessment module performs continuity, missing, and abnormal data supplementation analysis on multiple minimal nursing evidence units within a 90- to 180-day observation window. It calculates the frequency of nursing dependence evidence, the number of days of missing nursing services, the distribution pattern of service time, the consistency of activity distribution, and the consistency of multi-source data for each ADL dimension, and generates a disability persistence consistency score according to preset weights. The preset weights are configured according to regional policies or insurance terms.

9. The long-term care disability status assessment system according to claim 1, characterized in that, The audit support decision-making module outputs quick audit suggestions, routine audit suggestions, remote certificate supplementation suggestions, sampling home visit suggestions, or key verification suggestions based on the disability continuity consistency score, nursing service authenticity score, insurance clause matching degree, and data integrity.

10. The long-term care disability status assessment system according to claim 1, characterized in that, The hierarchical authorization output module outputs a service completion summary and anomaly alert to the family member; outputs caregiver service records and abnormal work orders to the nursing institution; and outputs service item codes and the smallest nursing evidence unit hash to the medical insurance agency. Provide ADL evidence, waiting period satisfaction, and claim confidence to commercial insurance companies; and provide on-site key verification checklists and long-term trend reports to assessors.

11. The long-term care disability status assessment system according to claim 1, characterized in that, The evaluation system stops collecting new data after the user revokes authorization, and retains, freezes, anonymizes, or deletes existing evidence according to audit requirements; the evaluation system performs local encryption caching, sequence number continuity recording, and retransmission integrity verification on the data collected during the network outage.

12. The long-term care disability status assessment system according to claim 1, characterized in that, The assessment system also includes an abnormal forgery cost identification module, used to identify at least one of the following abnormal patterns: multiple caregiver accounts simultaneously serving the same applicant in different locations; long-term inconsistency between service items and vital signs or equipment events; a large number of minimal care evidence units being added before and after the application; and highly templated semantics of service statement voice and lack of acoustic events during the service process.

13. The long-term care disability status assessment system according to claim 1, characterized in that, The evaluation system also includes a voiceprint-location coupling verification unit, which is configured to: in the nursing scenario of a day care facility, compare the real-time voiceprint features of the caregiver with the pre-registered voiceprint template, and at the same time verify whether the caregiver's location data is within the virtual fence area of ​​the facility, and verify that the voiceprint activity detection during the service process is passed. When the three are coupled in a consistent manner, a voiceprint-location coupling verification pass identifier is generated.

14. The long-term care disability status assessment system according to claim 1, characterized in that, The assessment system also includes a multi-caregiver cross-voiceprint dialogue verification unit, which is configured to: when at least two registered caregivers are providing services together at the same assessment time point, extract the voiceprint identifiers and service dialogue audio streams of each caregiver, confirm the consistency between the actual caregivers present and their registered identities through dialogue voiceprint separation and cross-comparison, and detect whether the dialogue content contains preset service interaction keywords to generate cross-voiceprint dialogue verification confidence.

15. The long-term care disability status assessment system according to claim 1, characterized in that, The evaluation system enables different auditing entities to verify the existence and completeness of evidence without obtaining the complete original data by using verifiable digest codes, hash check codes, or authorization tokens.

16. A method for assessing disability status in long-term care, characterized in that, When applied to the evaluation system according to any one of claims 1-15, the evaluation method comprises the following steps: Authorization tokens and data collection permissions are generated based on the identity relationships of the applicant, the person receiving care, the authorized caregiver, the caregiver, or the care institution. During authorized service periods or preset observation periods, mobile terminals are used to collect nursing staff's service declaration voice, environmental acoustic events during the service process, and the voice responses of the patients. The sound data is processed locally on the mobile terminal, including acoustic feature extraction, voice activity detection, service keyword extraction, acoustic event tag generation, and sensitive information masking. Only feature vectors, event tags, and timestamps are uploaded, while the original audio is stored on the local edge device. Selectively access one or more of the following as enhanced data sources: vital sign summaries, desensitized video behavior tags, nursing equipment event summaries, location or near-field verification data; The identity, authorization, time, location level, service statement summary, acoustic event summary, optional sensor summary, ADL dimension label and verification information corresponding to a single nursing service or a single status observation are solidified into the smallest unit of nursing evidence; Within an observation window of 90 to 180 days, the frequency of nursing dependence evidence, continuity of evidence, risk of missing evidence, consistency of activity distribution, and consistency of multi-source data in the ADL dimension were calculated based on multiple minimal nursing evidence units, and a disability persistence consistency score was generated according to preset weights. Based on the aforementioned disability continuity consistency score, service authenticity risk marker, and data integrity output audit path suggestion; Output de-identified evidence summaries and verifiable credentials that match the permissions of different authorized entities; Key information from each data stream is extracted to generate a standard structured assessment report, which serves as the basis for review by the long-term care insurance management terminal.

17. A computer-readable storage medium, characterized in that, The system contains a computer program that, when executed by a processor, implements the steps of the evaluation method of claim 16.

18. An electronic device, characterized in that, It includes a processor, a memory, and a communication interface, wherein the memory stores a program that, when executed by the processor, implements the evaluation method of claim 16.

Images