AI-based timing-triggered risk screening for geriatric syndromes and intelligent push system for home care

Abstract

This invention relates to the fields of smart healthcare and AI-assisted decision-making technology, specifically to an AI-triggered intelligent system for risk screening and home care of geriatric syndromes. The system includes a data integration module for connecting to existing hospital information systems, collecting and standardizing multi-source clinical data from elderly hospitalized patients. Based on existing structured and semi-structured data in hospital information systems and electronic medical records, this invention continuously monitors changes in the patient's condition in the background. At key points such as admission, abnormal test results, and changes in condition, it automatically pushes risk screening reminders to the medical team. After professional assessment and confirmation by medical staff, personalized home care suggestions are generated upon patient discharge and pushed to family members' terminals through compliant channels. This improves the timeliness and coverage of screening while ensuring the professionalism of medical procedures and data security. It can be widely applied in geriatric departments or related departments of hospitals at all levels.

Description

Technical Field

[0001] This invention relates to the fields of smart healthcare and AI-assisted decision-making technology, specifically to an AI-based, time-triggered intelligent system for screening risks of geriatric syndromes and providing home care recommendations. Background Technology

[0002] Geriatric syndromes (falls, delirium, frailty, malnutrition, etc.) are common complications in elderly hospitalized patients, significantly increasing the risk of adverse events and readmission. Current clinical practice presents three prominent problems: First, screening relies on regular manual processes, making it difficult to dynamically trigger reassessment during critical "windows" of disease progression (such as abnormal test results or medication adjustments), leading to widespread missed or delayed screenings. Second, discharge instructions often use generic templates, making it difficult for families to obtain individualized care points related to in-hospital assessments, resulting in a disconnect between home care and clinical practice. Third, some existing assistive systems or attempts to replace medical decision-making with AI pose compliance risks or require additional hardware deployment, resulting in poor reusability.

[0003] A search revealed a system and method for generating individualized nursing plans based on comprehensive assessments of the elderly, with publication number CN120412881B. This patent collects various physiological time-series data of the elderly in specific life scenarios and extracts their periodic fluctuation characteristics to identify subtle trends in their daily rhythms. By comparing the degree and persistence of rhythm deviations in different time periods, it reveals the dynamic impact of chronic disease progression on physiological states. Combined with the rate and direction of trend changes, it characterizes the interaction pattern between physiological changes and chronic disease indicators through continuous observation, thereby measuring the degree of coupling adaptation between individual physiological responses and chronic disease conditions. Through normalized amplitude difference and matching interval screening, it accurately assigns individual physical fitness and chronic disease suitability labels, and then uses a multi-dimensional scoring method to comprehensively measure the suitability priority of various intervention elements in an individual, constructing a nursing intervention combination tailored to specific population conditions. However, this solution focuses on rhythm shift analysis based on long-term time series data. Its data collection relies on physiological time series data in specific life scenarios. It does not solve the problem of real-time dynamic triggering based on existing HIS / EMR data in hospital settings, nor does it involve a complete closed-loop mechanism from in-hospital assessment to compliant home care delivery.

[0004] While some AI-based medical assistance systems have emerged in recent years, most have excessively pursued AI's autonomous decision-making capabilities, attempting to replace medical staff in diagnosis and assessment. This poses risks to medical safety and compliance, and there is a lack of awareness regarding data compliance, potentially leading to patient privacy breaches. Furthermore, they fail to establish a complete closed loop from screening triggers and manual assessment to discharge care guidance, and most require additional hardware deployment, making it impossible to directly reuse data assets from existing hospital information systems, thus limiting their practicality. Therefore, we propose an AI-based, timing-triggered intelligent system for geriatric syndrome risk screening and home care recommendations. Summary of the Invention

[0005] The purpose of this invention is to provide an AI-based, time-triggered intelligent system for geriatric syndrome risk screening and home care, which features proactive reminders at key points, establishes clear boundaries of responsibilities, breaks down information barriers within hospitals, and complies with data security regulations. It addresses how to automatically trigger geriatric syndrome risk screening reminders at critical moments such as hospital admission, changes in patient condition, and abnormal laboratory indicators, while reusing existing HIS / EMR data. This represents a shift from passively waiting for manual initiation to proactively triggering reminders. The invention also addresses how to construct a complete closed loop of AI triggering, human assessment, and compliant delivery, clearly defining the boundaries of responsibility between machine assistance and human guidance to ensure the professionalism and compliance of medical practices. Furthermore, it addresses how to intelligently transform in-hospital assessment results into de-identified, personalized home care recommendations and deliver them to family members in compliance with regulations, thus breaking down information barriers between in-hospital assessment and home care.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an AI-based, time-triggered intelligent system for screening risks of geriatric syndromes and providing home care, comprising: The data integration module is used to connect with the hospital's existing information system, collect and standardize the multi-source clinical data of elderly inpatients. The multi-source clinical data includes patient basic information, admission assessment data, diagnostic information, laboratory indicators, key fields of nursing records, medication information and data on changes in the patient's condition. The AI ​​timing trigger module has a built-in risk rule base and risk prediction model to continuously monitor changes in patient data and automatically generate graded risk screening reminders at key moments such as admission, changes in condition, and abnormal test indicators. The AI ​​timing trigger module includes a monitoring engine and a trigger decision-maker. The monitoring engine traverses the data of elderly patients in the hospital at a preset period and matches it with the trigger conditions in the rule base. The trigger decision-maker filters the successfully matched trigger conditions according to a preset deduplication and suppression mechanism and generates risk screening reminder records. The manual assessment interaction module is used to push the risk screening reminder to the medical staff terminal, provide a visual interface that includes patient information, risk type, triggering basis and suggested assessment tools, and receive and save the professional assessment results of medical staff; The assessment result processing and care knowledge base module includes an assessment result integrator and a care knowledge base. The assessment result integrator is used to summarize and integrate the assessment results of multiple geriatric syndromes of the same patient and generate a comprehensive risk assessment file. The care knowledge base is used to store home care knowledge items corresponding to the risk types and risk levels of each geriatric syndrome, and to perform condition matching based on the assessment results to generate individualized care recommendations. The authorization management and home care push module is used to obtain electronic authorization confirmation from the patient or legally authorized representative before the patient is discharged, integrate the assessment results after the patient is discharged and generate individualized home care suggestions, which are then sent to the family members through a preset secure push channel after de-identification processing.

[0007] Preferably, the hospital information systems connected to the data integration module include HIS, EMR, laboratory, and nursing systems. The collected data includes Morse fall score, Braden pressure ulcer score, Barthel activities of daily living score, simplified nutritional assessment (MNA-SF) score, serum albumin, hemoglobin, C-reactive protein, and descriptions of consciousness and activity levels in nursing record texts. For unstructured nursing record data, a natural language processing model is used for entity extraction.

[0008] Preferably, in the AI ​​timing triggering module, the rule base covers geriatric syndrome risk types including fall risk, delirium risk, frailty risk, malnutrition risk, pressure ulcer risk, incontinence risk, and aspiration risk. The triggering conditions include age threshold conditions, assessment score threshold conditions, test indicator threshold conditions, nursing record keyword conditions, medication type conditions, and condition change event conditions. The conditions are connected by "AND" logic or "OR" logic.

[0009] Preferably, the deduplication and inhibition mechanism of the triggering decision-maker includes: triggering only once within a preset cooling-off period for the same patient and the same risk type; Suppress this trigger if the patient already has a manual assessment record for this risk type during the cooling-off period; When a patient experiences a pre-defined significant risk event during the cooling-off period, the cooling-off period restriction can be ignored for triggering.

[0010] Preferably, the AI ​​timing triggering module adopts a graded early warning mechanism, which divides the early warning into three levels: high, medium, and low, according to the degree of risk and sets different response time limits. The response time limit for high-risk early warning is shorter than that for medium-risk and low-risk early warning.

[0011] Preferably, the manual assessment interaction module is embedded in the hospital's existing EMR and nursing system as a plug-in. The visual interface provides operation options such as "Assess Now", "Remind Later", and "Confirm Known / No Assessment Required", as well as the function of one-click jump to the standardized electronic assessment form. The standardized electronic assessment form includes the Morse Fall Scale, the MNA-SF Nutritional Assessment Scale, the CAM Delirium Assessment Scale, the FRAIL Frailty Assessment Scale, and the Braden Pressure Ulcer Assessment Scale.

[0012] Preferably, the home care knowledge entries in the care knowledge base are organized and stored according to three dimensions: risk type, risk level, and care area. A "condition-template" matching mechanism is adopted, which selects the corresponding template based on the risk type and risk level in the assessment results as the matching conditions, and fills the variable fields in the template according to the score values ​​of each dimension in the assessment results to generate individualized care suggestion text. The care areas include home environment safety, daily activity guidance, nutritional support, medication management, cognitive care, pressure ulcer prevention, and excretion care.

[0013] Preferably, when the authorization management and home care push module generates the authorization confirmation page, it displays a description of the scope of the push content to the patient or family, clearly stating that it only includes care points and risk prevention measures, and does not include original medical records, diagnostic information and test data. The authorization is completed through electronic signature, and the system saves the authorizing person's name, authorization timestamp and electronic signature hash value as compliance records.

[0014] Preferably, the home care recommendation includes a cover, a comprehensive risk warning section, a section on specific care points, a section on emergency identification and medical treatment indications, and a section on follow-up / revisit recommendations. The content is transmitted using the TLS encryption protocol, and the push channels include the hospital's officially certified WeChat channel, SMS with a secure link, or the hospital's APP family terminal message push channel.

[0015] Preferably, the system further includes a model iteration module and a data security audit module. The model iteration module is used to collect labeled data of AI warnings from medical staff and support rule base adjustment and incremental model training. The data security audit module is used to de-identify and store all patient clinical data and record full-link audit logs. The system is deployed in the hospital's intranet environment, and all data processing and storage are completed within the hospital.

[0016] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention enables time-triggered dynamic screening based on existing hospital data, proactively reminding patients at key points, significantly reducing missed or delayed screenings, and establishing clear boundaries of responsibility between machine-assisted and human-led approaches to ensure compliance of medical practices.

[0017] 2. This invention breaks down the information barriers between in-hospital assessment and home care, and pushes personalized suggestions with authorization to alleviate the anxiety of family caregivers. Moreover, the data is processed in a closed loop throughout the hospital, which complies with data security regulations. Furthermore, the system is highly configurable, can reuse existing data assets, and has low deployment costs. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall system architecture of the present invention; Figure 2 This is a flowchart of the system workflow of the present invention; Figure 3 This is an example diagram of the user interface for the manual evaluation interaction module of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0020] Please see Figures 1-3 As shown, this invention provides a technical solution: an AI-based, time-triggered intelligent system for screening risks of geriatric syndromes and providing home care, deployed within a hospital's internal information network environment, connected to HIS and EMR via a standardized data interface, comprising the following modules: Data Integration Module: This module establishes data channels with HIS, EMR, laboratory systems, and nursing systems via the HL7 / FHIR standard interface. It acquires clinical data from hospitalized elderly patients (≥65 years old) periodically (configurable every 4 hours or every 8 hours) or in real-time. The acquired data includes, but is not limited to: basic patient information (age, gender, hospital number, admission time, discharge time, ward, bed number); admission assessment data (Morse fall score, Braden pressure ulcer score, Barthel activities of daily living score, Mini-Nutritional Assessment (MNA-SF) score, pain score, etc.); diagnostic information (admission diagnosis ICD code, newly added diagnosis ICD codes during hospitalization); and laboratory indicators (serum albumin AL). B. Prealbumin (PA), Hemoglobin (Hb), C-reactive protein (CRP), serum creatinine (Cr), blood urea nitrogen (BUN), electrolytes, etc.; key fields in nursing records (description of consciousness, history of falls, description of activity level, excretion status, nighttime sleep, medication adherence, etc., extracted from nursing record text using natural language processing rules); medication information (use of high-risk drugs such as benzodiazepines, antipsychotics, antihypertensives, diuretics, hypoglycemics, etc.); events of change in condition (ICU transfer records, surgical records, fall event records, restraint use records, etc.). The collected data are standardized, cleaned, marked for missing values, removed from outliers, and structured mapped to key fields before being stored in the system's built-in database.

[0021] AI Timing Trigger Module: This module is the core of the system and includes three sub-components: rule base, monitoring engine, and trigger decision-maker. Rule base: Pre-set risk trigger rules covering six types of geriatric syndromes (fall risk, delirium, frailty, malnutrition, pressure sores, and incontinence). Each rule includes fields such as risk type, trigger condition, logical relationship (AND / OR), risk level (high / medium / low), and suggested assessment tools.

[0022] Example rules are as follows: Fall risk triggering rules: If (age ≥ 80 years and Morse score missing upon admission) or (albumin < 35 g / L and Hb < 100 g / L) or (use of benzodiazepines and nursing record shows "unsteady gait") or (Morse score ≥ 45 upon admission and length of hospital stay ≥ 3 days and nursing record shows "getting out of bed" but no recent fall reassessment record), then a fall risk screening reminder is triggered. The recommended assessment tool is the Morse Fall Assessment Scale.

[0023] Malnutrition risk triggering rules: If (admission MNA-SF score is missing and age ≥70 years) or (ALB <35g / L) or (admission diagnosis includes ICD code E40-E46 and no nutritional assessment record) or (BMI <18.5 and length of hospital stay ≥5 days), a nutritional risk screening reminder will be triggered. The recommended assessment tool is MNA-SF or NRS-2002.

[0024] Delirium risk triggering rules: If (age ≥75 years and nursing records show "sleep inversion", "irritability" or "incoherent speech"), or (use of antipsychotic medication and nursing records show "inattention"), or (within 72 hours of ICU admission and no delirium assessment record), then a delirium risk screening reminder is triggered. The recommended assessment tool is the CAM scale.

[0025] Frailty risk triggering rules: If (age ≥75 years and Barthel score <60 on admission) or (ALB <35g / L and CRP >10mg / L) or (weight loss >5% during hospitalization and “fatigue” recorded in the nursing record), a frailty risk screening reminder is triggered. The recommended assessment tools are the FRAIL scale or the Clinical Frailty Scale (CFS).

[0026] Pressure ulcer risk triggering rules: If (the admission Braden score is missing and the patient is ≥70 years old) or (the admission Braden score is ≤18 points and the length of hospital stay is ≥7 days and there is no reassessment record) or (the nursing record shows "skin damage" and there is no pressure ulcer staging assessment record), then a pressure ulcer risk screening reminder will be triggered. The recommended assessment tool is the Braden score.

[0027] Incontinence risk triggering rules: If (age ≥ 80 years and nursing record shows "frequent urination" or "difficulty urinating") or (use of diuretics and nursing record shows "increased nocturia") or (admission diagnosis includes ICD code R32 / N39), then incontinence risk screening reminder will be triggered.

[0028] Aspiration risk triggering rules: If (age ≥ 65 years and admission diagnosis includes ICD code I60-I69 cerebrovascular disease / G20 Parkinson's disease / F00-F03 dementia / G70 myasthenia gravis / C00-C14 head and neck malignancy and no swallowing function assessment record) or (nursing record shows "choking cough" or "difficulty swallowing" or "change in voice after eating") or (within 6 hours after general anesthesia and no swallowing assessment record) or (new onset of consciousness disorder during hospitalization and nursing record shows "drowsiness" or "stunted sleep" or "coma"), then an aspiration risk screening reminder will be triggered. The assessment tools were selected according to the patient type: the modified Kubota water swallowing test was used for non-neurological patients who were conscious; the aspiration risk assessment scale (including 8 dimensions: age, comorbidities, artificial airway, consciousness, sputum characteristics, diet type, body position, and modified Kubota water swallowing test grading, with a total score of 10-12 indicating low risk, 13-18 indicating moderate risk, and 19-23 indicating high risk) was used for patients who were not neurological patients and were conscious.

[0029] Monitoring Engine: The monitoring engine iterates through all data snapshots of elderly patients in the hospital using a timed polling method (running once every 6 hours by default), matching the patient's current data against the trigger conditions in the rule base one by one. The matching logic uses forward reasoning (Rete algorithm or its simplified version). When at least one piece of data of a patient meets all the conditions of a rule in the rule base, the rule is marked as "to be triggered". The monitoring engine also supports risk prediction models built based on logistic regression algorithms. The training data of the model comes from the de-identified medical records of elderly inpatients in the target hospital over the past 3 years, which are labeled by clinical experts and used for model training and validation.

[0030] Trigger Decision Maker: To avoid alarm fatigue caused by repeated triggering, the trigger decision maker is equipped with the following deduplication and suppression mechanism: For the same patient and the same risk type, it will only be triggered once within the preset "cooling-off period" (e.g., 72 hours), unless new significant risk evidence (e.g., new abnormal test values ​​or new disease change events) appears. If a patient already has a manual assessment record for this risk type during the cooling-off period, the trigger is suppressed. The trigger decision-maker uses a tiered early warning mechanism, classifying warnings into three levels—high, medium, and low—based on the degree of risk, and employing different push methods and processing priorities: high-risk warnings are indicated by a red pop-up window and an audio alert, requiring an assessment to be completed within one hour; Medium-risk areas are indicated by a yellow pop-up notification, requiring an assessment to be completed within 4 hours; low-risk areas are indicated by a regular message notification, requiring an assessment to be completed within 24 hours.

[0031] Once a triggering event is confirmed by the triggering decision-maker, a risk screening reminder record is generated, which includes the patient's identity information, risk type, risk level, triggering basis (i.e., which data fields were matched and their current values), suggested assessment tools, trigger timestamp, and is then pushed to the reminder distribution module.

[0032] Human Assessment Interaction Module: This module is embedded into the hospital's existing EMR and nursing system as a plug-in. It pushes risk screening reminders to the responsible nurse or physician through the medical and nursing workstation terminal (PC and / or mobile nursing PDA). The reminder interface includes: patient basic information (name, bed number, hospital number, age). Trigger risk type and risk level (displayed by color, high risk is red, medium risk is yellow, and low risk is blue); Triggering Criteria Summary (Concisely list the key data that triggered this alert, such as "serum albumin: 32 g / L (reference range 35-55 g / L), MNA-SF score missing on admission"); Suggest the name of the assessment tool and a link that allows one-click access to the corresponding electronic assessment form; It also includes three operation options: "Evaluate Now", "Remind Me Later", and "Confirm Known / No Evaluation Required".

[0033] Healthcare workers can click "Assess Now," and the system will automatically redirect them to the corresponding standardized electronic assessment forms (such as the Morse Fall Scale, MNA-SF Nutrition Assessment, CAM Delirium Assessment, FRAIL Frailty Assessment, and Braden Pressure Ulcer Assessment). The assessment forms have pre-set scoring items for each dimension. After completing the assessment in a clinical setting, the nurse clicks submit, and the system automatically calculates the total assessment score and provides the corresponding risk level. If healthcare workers believe that the AI ​​warning is incorrect, they can mark it as "false alarm" and explain the reason. The system will then feed this data back to the model iteration module for optimization.

[0034] The assessment results are saved to the system database in a structured data format, with fields including: Patient ID, Risk Type, Assessment Tool Name, Scores for Each Dimension, Total Score, Risk Level, Assessor, and Assessment Timestamp. This assessment result is also fed back as an "Existing Assessment Record" to the inhibition logic that triggers the decision-making process.

[0035] Assessment Results Processing and Care Knowledge Base Module: This module includes two sub-components: an assessment results integrator and a care knowledge base. Among them, the assessment result integrator: summarizes and integrates the assessment results of multiple geriatric syndromes for the same patient to generate a comprehensive risk assessment file for the patient's geriatric syndromes. The integration logic includes: listing the assessment results and risk levels according to risk type. High-risk items are marked (such as Morse score ≥45, Braden score ≤12, MNA-SF score ≤7, CAM assessment positive, FRAIL score ≥3, etc.). Sort by risk level in descending order, with high-risk items displayed first; Generate a "Comprehensive Risk Alert" field, summarizing the key risk areas that the patient needs to pay attention to.

[0036] Secondly, the care knowledge base: a pre-built knowledge base of home care recommendations corresponding to various risk types of geriatric syndromes. The knowledge items are organized according to three dimensions: risk type, risk level, and care area. Taking fall risk as an example, the knowledge base includes: key points for home environment safety assessment for high-risk patients (Morse score ≥ 45) (lighting, anti-slip flooring, handrail installation, obstacle removal), assistive device usage guidance (selection and use of walkers / wheelchairs), posture transition guidance (three steps to getting up), medication safety management (warnings about sedative-hypnotic drugs), etc. Daily activity precautions, balance training suggestions, and footwear selection guidance for patients at medium risk (Morse score 25-44); General fall prevention recommendations for low-risk patients (Morse score < 25).

[0037] Taking aspiration risk as an example, the knowledge base includes: key points for airway management of high-risk patients (modified Kubota water swallowing test level V or aspiration risk assessment form score 19-23) (confirm artificial airway cuff pressure to be maintained at 25-30 cmH2O, effective coughing or suctioning before feeding), nutritional route management (suspend oral feeding, recommend postpyloric feeding, hang aspiration prevention reminder card), feeding parameter control (single feeding volume not exceeding 400ml, enteral nutrition pump initiation rate 20-50ml / h, nutrition solution temperature 37℃-40℃), and body position management (raise the head of the bed 30°-45° during continuous feeding, raise the head of the bed 60°-90° during intermittent feeding, maintain a semi-recumbent position for 30-60 minutes after feeding); Key points for the safety management of patients at medium risk (modified Kubota water swallowing test level IV or aspiration risk assessment form score of 13-18) (immediately stop oral feeding and water intake and report to the doctor, insert a gastric tube or nasoenteric tube if necessary, and prescribe anti-aspiration orders), and adjust the food state (add thickener to liquid food, and change solid food to puree). Feeding instructions for low-risk patients (modified Kubota water swallowing test grade III or aspiration risk assessment scale score of 10-12): food should be soft, finely chopped and cooked thoroughly; eating should be slow, taking 30-40 minutes; the patient should be sitting upright or in a 30°-60° semi-recumbent position with the neck tilted forward; a small, smooth-edged spoon with a capacity of 5-10 ml should be used; swallow one bite completely before eating the next; when feeding, place the food in the back of the mouth on the healthy side and gently press the tongue with the back of the spoon; maintain the patient's position for 30 minutes after eating and avoid turning over or patting the back.

[0038] The knowledge base entries employ a "condition-template" matching mechanism: the risk type and risk level in the assessment results serve as conditions, matching the corresponding template in the knowledge base. Then, based on the patient's specific assessment dimensions (e.g., a score of 10 on the "gait" dimension of the Morse scale indicates gait abnormality), the template is filled in to generate individualized care recommendations. The care knowledge base also covers multiple areas, including nutritional support (high-protein, small, frequent meals; recommendations for enteral nutrition supplementation), delirium management (quiet environment; light regulation; regular sleep patterns), pressure ulcer prevention (frequent turning; skin care; use of pressure-reducing devices), and frailty intervention (encouragement of safe activities; guidance on protein intake).

[0039] Authorization Management and Home Care Push Module: This module enables compliant push notifications from the hospital's internal system to family members. The core design follows the principles of "data not leaving the hospital, authorized push notifications, and unlabeled content." The authorization management process involves the responsible nurse initiating a "home care push authorization" process in the system 24-48 hours before the patient's discharge. The system generates an authorization confirmation page, showing the patient or their family a description of the scope of the push content (only including key care points and risk prevention measures, excluding original medical records, diagnoses, test data, and other specific treatment information). The patient or their legally authorized representative confirms the authorization with an electronic signature. The authorization record (including the authorizing person's name, the last four digits of their ID number, the authorization timestamp, and the electronic signature hash value) is stored in the system database as a basis for the compliance of the push.

[0040] Secondly, the care suggestion generation engine: After authorization is obtained and the patient is officially discharged (the "discharge" status in the HIS system is changed as the trigger signal), the system extracts the patient's comprehensive risk assessment file from the assessment result integrator, matches the corresponding care knowledge items from the care knowledge base, and synthesizes them through the template engine to generate a complete "Home Care Suggestion".

[0041] The recommendation letter should include: a cover page (with the patient's salutation, such as "family member of the elderly person") and the date of creation; The first part is a comprehensive risk warning (a brief description of the risk items and levels of geriatric syndromes found during the screening during this hospitalization, expressed in plain language, such as "There is a risk of falling, and attention should be paid to walking and transfer safety"). The second part is the itemized care points (specific care recommendations and precautions are listed one by one according to the risk type); The third part is emergency identification and medical treatment indications (such as "If the following situations occur, please seek medical attention immediately: inability to stand or walk, worsening confusion, weight loss of more than 1 kg within 3 days, etc."). Part Four provides recommendations for follow-up visits / re-visits.

[0042] Push Channels: After the care recommendation is generated by the hospital's internal system, it is sent to the family's mobile device in encrypted PDF format via the hospital's officially certified WeChat official account, SMS with a secure link, or push notifications to the family's mobile device through the hospital's app. The push content is encrypted using TLS 1.2 or higher during transmission. The push content does not contain any direct identifiers that could be associated with the patient (such as ID number, hospital number, or detailed address), retaining only necessary appellation information for identification.

[0043] Push status tracking: The system records push logs such as push time, push channel, and reception status (sent / read / unread), and sends follow-up reminders to family members via SMS or WeChat on the 7th day after the push, inquiring about the care situation and encouraging feedback.

[0044] Data storage and security audit module: This module is responsible for the full lifecycle management of the system's data, including: de-identified storage of all patient clinical data (patient direct identifiers are replaced by hash values ​​or internal numbers). Audit logs for all risk triggers, assessments, and push notifications (recording the operator, operation time, operation content, and IP address); Encryption of push content during transmission; Identity verification is required for family members to view care recommendations (via SMS verification code or WeChat real-name authentication). All data processing and storage are performed on the hospital's local server; no data leaves the hospital.

[0045] Model Iteration Module: The system has established a comprehensive model iteration mechanism. The system automatically collects labeled data (correct / false positive / irrelevant) from medical staff regarding AI alerts, and performs incremental training on the model monthly to continuously optimize its accuracy and recall. Simultaneously, when clinical guidelines are updated or hospitals introduce new assessment scales, the system supports timely updates to model parameters and trigger conditions. All trigger-assessment data, after anonymization, can be exported for the construction and validation of geriatric syndrome risk prediction models.

[0046] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0047] Example 1: This example uses a real-world application scenario in the geriatrics / general medicine department of a tertiary-level Class A hospital to describe the specific implementation process of the system of the present invention in terms of fall risk screening triggering and assessment.

[0048] Implementation Environment: The system is deployed within the hospital's intranet environment. The server is configured with an Intel Xeon 16-core processor, 64GB of RAM, 2TB of storage, and runs a Linux operating system. The system establishes data connections with the hospital's HIS and EMR systems via an HL7 V2 / FHIR R4 interface.

[0049] Step S1: Data Acquisition and Preprocessing After system deployment, the data acquisition and preprocessing module registers a data subscription service with the HIS / EMR system. The system filters inpatients aged ≥65 years from the HIS, obtaining basic information such as admission time, ward, and bed number for each patient, and also retrieves Morse fall score records (if available) from the admission assessment data. The laboratory system synchronizes all test results from the previous day every morning at midnight, and the nursing record system incrementally retrieves key fields from the latest nursing record text every 4 hours (extracted using natural language processing rules).

[0050] The following is a specific case to illustrate this: Patient Zhang, male, 82 years old, was admitted on October 8, 2025, due to "acute exacerbation of chronic obstructive pulmonary disease" (COPD), hospital number ZY20251008042. On the day of admission, the responsible nurse completed the admission assessment, but the Morse fall assessment scale was missed due to being busy, and there was no Morse score record for this patient in the system. Laboratory test results upon admission were: albumin 33.2 g / L (reference range 35-55 g / L), hemoglobin 105 g / L (reference range 130-175 g / L). The admission diagnosis ICD code was: J44.1 (acute exacerbation of COPD). The medication prescription included diazepam 5 mg orally every night (benzodiazepine).

[0051] Step S2: Risk Monitoring and Timing Triggering The system monitoring engine ran a scheduled polling task at 08:00 on October 9, 2025 (the morning after admission, approximately 20 hours after admission). Rule matching was performed on patient Zhang's data. The matching process is as follows: One of the fall risk triggering rules in the rule base is: "If (age ≥ 80 years and Morse score missing upon admission) or (albumin < 35 g / L and Hb < 100 g / L) or (using benzodiazepines and 'gait instability' recorded in the nursing record)". Matching revealed that patient Zhang met both branch conditions of this rule: ① Age = 82 years ≥ 80 years, and the Morse score field upon admission was "empty"; ②Albumin = 33.2g / L < 35g / L. Rule matching successful. Trigger decision maker check: This patient has no fall risk trigger records within 72 hours and no fall assessment records (0 Morse assessment form submission records). All trigger conditions are met. The system generates a risk screening reminder record: Risk type = fall risk, risk level = high, trigger basis = "Age 82 years old, Morse score missing upon admission; Albumin 33.2g / L < 35g / L”, recommended assessment tool = Morse Fall Assessment Scale, trigger time = 2025-10-09 08:00:23.

[0052] Step S3: Reminder Distribution and Manual Evaluation After the trigger event is generated, the reminder distribution module pushes the risk screening reminder to the to-do list of the responsible nurse (bed nurse) through the nursing workstation terminal. After logging into the nursing workstation, the nurse can see the reminder in the "Geriatric Syndrome Screening To-Do List". The interface displays "High Risk" with a red background, and lists the patient's name, bed number, summary of triggering basis, and "Assess Now" button.

[0053] The nurse clicks "Assess Now," and the system automatically redirects to the electronic form for the Morse Fall Assessment Scale. The Morse Scale includes six dimensions: fall history (0 or 25 points), secondary diagnosis (0 or 15 points), walking assistance (0 / 15 / 30 points), intravenous infusion / heparin lock (0 or 20 points), gait (0 / 10 / 20 points), and mental status (0 or 15 points). The nurse assesses the patient in a clinical setting: observing unsteady gait (score 10 points), normal mental status (score 0 points), presence of an indwelling intravenous catheter (score 20 points), secondary diagnosis (COPD, score 15 points), no fall history (0 points), and use of a walking aid (score 15 points). The total score = 10 + 20 + 15 + 15 = 60 points (≥45 points indicates high risk). The nurse completes the assessment and submits it; the assessment results are saved to the system database, and fall prevention nursing measures (bed rail use, call bell placement, toilet assistance, etc.) are simultaneously developed.

[0054] Step S4: Subsequent dynamic monitoring On the fifth day of hospitalization (October 13, 2025), the nursing record described the patient as having "unsteady gait when getting out of bed to use the toilet at night." The monitoring engine again matched the fall risk rule (use of benzodiazepines and "unsteady gait" in the nursing record). Because the 72-hour cooling-off period had expired and new risk keywords appeared in the nursing record, the decision-maker generated another alert. The nurse's Morse score remained at 60, so nighttime rounds and toilet assistance were increased. Compared to similar cases before the system went live (where only one assessment was completed upon admission), this case achieved screening at two critical moments: early admission and during a period of change in the patient's condition, avoiding potential missed screenings and risks that might result from relying solely on a single admission assessment.

[0055] Example 2 describes the complete closed-loop process from admission screening triggering, multi-risk comprehensive assessment to the generation and push of home care recommendations upon discharge.

[0056] Admission to Mid-Hospitalization: Multiple Risk Triggers and Assessment Patient Li, female, 86 years old, was admitted on November 3, 2025, due to "fatigue and decreased appetite for 2 weeks" (hospital number ZY20251103015). On admission assessment, her MNA-SF score was 8 (suggesting malnutrition risk), Morse score was 35 (medium risk), and Braden score was 16 (mild pressure ulcer risk). Admission laboratory results: albumin 30.5 g / L, hemoglobin 92 g / L, CRP 15.6 mg / L.

[0057] On the third day after admission (November 6, 2025), the system monitoring engine detected the following triggering conditions during rule matching: Malnutrition risk - Admission MNA-SF score of 8 (≤11 points is high risk), albumin 30.5g / L <35g / L. Meeting both conditions simultaneously triggered a nutritional risk screening alert; Frailty risk - Age 86 years ≥75 years, admission Barthel score 55 points <60 points, albumin 30.5g / L <35g / L and CRP 15.6mg / L >10mg / L, triggered a frailty risk screening alert.

[0058] After the reminder was distributed to the responsible nurse, the nurse completed the MNA-SF (reassessment score 7, malnutrition) and FRAIL scale (score 4, frailty) assessments. On the 8th day of hospitalization (November 11, 2025), the nursing record showed "the patient had difficulty falling asleep at night, was agitated, and repeatedly asked about the time." The monitoring engine matched the delirium risk trigger rules: age ≥75 years (86 years), nursing record containing "agitation," triggering the delirium risk screening reminder. The nurse used the CAM scale for assessment, and the result indicated a positive result for delirium. Therefore, intervention measures such as a quiet environment, nighttime light regulation, and avoiding frequent changes of caregivers were implemented.

[0059] Pre-discharge comprehensive risk assessment and authorization The patient's condition is stable, and discharge is scheduled for November 15, 2025. Twenty-four hours before discharge (November 14, 2025), the responsible nurse initiated a "Home Care Push Authorization" process in the system. The system generated an authorization page, displaying a description of the push content to the patient's daughter (the legally authorized representative): "This push will provide you with key points for the patient's home care after discharge and risk prevention suggestions, including guidance on fall prevention, nutritional support, and cognitive care. The push content does not include specific medical information such as medical records, diagnoses, or test results." The patient's daughter completed the electronic signature authorization on her PDA. The authorization record was saved to the system database.

[0060] Discharge care recommendations generation and delivery At 09:30 on November 15, 2025, the patient's status in the HIS system changed to "Discharged". This event triggered the operation of the assessment result processing and care knowledge base module.

[0061] The assessment results integrator extracted the patient's assessment records during hospitalization: Fall risk – Morse score 35 (medium risk), with the main risk dimensions being gait (10 points) and intravenous infusion (20 points); Malnutrition risk – MNA-SF score 7 (malnutrition), with albumin as low as 30.5 g / L; Weakness – FRAIL score 4 (weakness), manifested as fatigue, slow walking speed, and decreased activity level; Delirium risk – CAM assessment positive, with delirium episodes during hospitalization; Pressure ulcer risk – Braden score 16 (mild risk).

[0062] Care knowledge base matching and template synthesis process: Based on the risk type and risk level in the assessment results, the system sequentially matches corresponding care knowledge items from the knowledge base. Table 1 shows the correspondence between the assessment results and the knowledge base matching, as well as examples of generated care suggestions.

[0063] Table 1. Examples of matching assessment results with care knowledge base and generation of care recommendations. Risk type Evaluation results Risk level Knowledge base matching rules Generating care recommendations (excerpt) Fall risk Morse = 35 points Medium risk Morse 25-44 points template group It is recommended to use a walking aid at home; move slowly when getting up and turning; keep the floors dry and remove obstructions from walkways; turn on the lights or use a bedside lamp when using the toilet at night; wear non-slip shoes and avoid walking in slippers. malnutrition MNA-SF=7 points malnutrition MNA-SF≤7 points template group It is recommended to eat small, frequent meals, 5-6 meals a day; prioritize high-protein foods (eggs, fish, milk, soy products); enteral nutrition supplements can be taken under the guidance of a physician; weigh yourself at a fixed time each week and record weight changes. weak FRAIL = 4 points weak FRAIL ≥ 3-point template group Encourage gentle activities (indoor walking, seated leg raises) within safe limits; avoid prolonged bed rest; ensure adequate daily protein intake; if activity levels significantly decline, seek medical attention promptly. Delirium / Cognition CAM positive delirium positive Delirium Management Templates Keep your home environment quiet and well-lit; avoid frequently changing caregivers; establish a regular routine, with appropriate daytime light exposure and activity, and ensuring adequate sleep at night; if confusion worsens or incoherent speech occurs, seek medical attention immediately. Pressure ulcer risk Braden = 16 points Mild risk Braden's 15-18 point template group It is recommended to assist the patient to turn over every 2 hours; keep the skin clean and dry; check for redness in bony prominences (sacrum, coccyx, heels); use a pressure-relieving mat or air mattress. The care recommendation generation engine integrates the care recommendation texts corresponding to each risk type to create a complete "Home Care Recommendation Form." The recommendation form is generated in encrypted PDF format and pushed to the patient's daughter's WeChat account via the hospital's WeChat Work channel at 10:15 AM on November 15, 2025.

[0064] Push log record: Push time = 2025-11-15 10:15:32, Push channel = WeChat Work, Last four digits of recipient's mobile phone number = ×××5678, Receipt status = Sent, Read status = 2025-11-15 14:22:18 Read. Seven days after the push (November 22, 2025), the system automatically sent a follow-up reminder message to the family, inquiring about the care situation and whether there were any questions.

[0065] To verify the technical effectiveness of this invention, a prospective application observation was conducted for 3 months in the geriatrics / general medicine department of a tertiary-level Class A hospital. A total of 326 elderly inpatients aged ≥65 years were included. During the system's operation, a total of 187 risk screening alerts were generated, including 42 high-risk alerts, 76 medium-risk alerts, and 69 low-risk alerts. The response rate of medical staff to the alerts was 100%, with an average response time of 28 minutes. The system's operational performance indicators are shown in the table below.

[0066] Table 2 System Application Effectiveness Indicators index numerical values Geriatric syndrome screening coverage 100%（326 / 326） AI alert accuracy rate (confirmed by human evaluation) 89.3% Average time per evaluation 3.2 minutes Family care advice reading rate 92.7% Patient / Family Satisfaction Rating 96.5 points (out of 100) Regarding the timeliness of screening, the average time from admission to completion of the first assessment for fall risk assessment has been shortened from 22.5 hours before the system went online to 8.2 hours, malnutrition risk assessment from 46.3 hours to 12.5 hours, frailty assessment from 78.6 hours to 18.3 hours, and delirium assessment from 54.2 hours to 10.8 hours.

[0067] In terms of system stability, the system operated stably during the 3-month intervention period, with no interruption in the data interface. The trigger engine ran an average of 4 times per day, the response rate of medical staff (the proportion of those who clicked "Assess Now" to complete the assessment) was 89.1%, and the average system push delay (the time from the change of discharge status to the delivery of care advice to the family) was 48.3 ± 12.5 minutes.

[0068] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the essence and scope of the technical solutions of the present invention.

Claims

1. An AI-based, time-triggered intelligent system for screening geriatric syndrome risks and providing home care recommendations, characterized in that: include: The data integration module is used to connect with the hospital's existing information system, collect and standardize the multi-source clinical data of elderly inpatients. The multi-source clinical data includes patient basic information, admission assessment data, diagnostic information, laboratory indicators, key fields of nursing records, medication information and data on changes in the patient's condition. The AI ​​timing trigger module has a built-in risk rule base and risk prediction model to continuously monitor changes in patient data and automatically generate graded risk screening reminders at key moments such as admission, changes in condition, and abnormal test indicators. The AI ​​timing trigger module includes a monitoring engine and a trigger decision-maker. The monitoring engine traverses the data of elderly patients in the hospital at a preset period and matches it with the trigger conditions in the rule base. The trigger decision-maker filters the successfully matched trigger conditions according to a preset deduplication and suppression mechanism and generates risk screening reminder records. The manual assessment interaction module is used to push the risk screening reminder to the medical staff terminal, provide a visual interface that includes patient information, risk type, triggering basis and suggested assessment tools, and receive and save the professional assessment results of medical staff; The assessment result processing and care knowledge base module includes an assessment result integrator and a care knowledge base. The assessment result integrator is used to summarize and integrate the assessment results of multiple geriatric syndromes of the same patient and generate a comprehensive risk assessment file. The care knowledge base is used to store home care knowledge items corresponding to the risk types and risk levels of each geriatric syndrome, and to perform condition matching based on the assessment results to generate individualized care recommendations. The authorization management and home care push module is used to obtain electronic authorization confirmation from the patient or legally authorized representative before the patient is discharged, integrate the assessment results after the patient is discharged and generate individualized home care suggestions, which are then sent to the family members through a preset secure push channel after de-identification processing.

2. The AI-based, time-triggered intelligent push system for geriatric syndrome risk screening and home care as described in claim 1, characterized in that: The data integration module interfaces with hospital information systems including HIS, EMR, laboratory, and nursing systems. The collected data includes Morse fall score, Braden pressure ulcer score, Barthel activities of daily living score, simplified nutritional assessment (MNA-SF) score, serum albumin, hemoglobin, C-reactive protein, and descriptions of consciousness and activity levels in nursing record texts. For unstructured nursing record data, a natural language processing model is used for entity extraction.

3. The AI-based, time-triggered intelligent push system for geriatric syndrome risk screening and home care as described in claim 1, characterized in that: In the AI ​​timing triggering module, the rule base covers geriatric syndrome risk types including fall risk, delirium risk, frailty risk, malnutrition risk, pressure ulcer risk, incontinence risk, and aspiration risk. The triggering conditions include age threshold conditions, assessment score threshold conditions, test indicator threshold conditions, nursing record keyword conditions, medication type conditions, and condition change event conditions. The conditions are connected by "AND" logic and "OR" logic.

4. The AI-based, time-triggered intelligent push system for geriatric syndrome risk screening and home care as described in claim 1, characterized in that: The deduplication and inhibition mechanism of the trigger decision-maker includes: for the same patient and the same risk type, it is triggered only once within a preset cooling-off period; Suppress this trigger if the patient already has a manual assessment record for this risk type during the cooling-off period; When a patient experiences a pre-defined significant risk event during the cooling-off period, the cooling-off period restriction can be ignored for triggering.

5. The AI-based, time-triggered intelligent push system for geriatric syndrome risk screening and home care as described in claim 1, characterized in that: The AI ​​timing triggering module adopts a graded early warning mechanism, which divides early warnings into three levels: high, medium, and low, based on the degree of risk, and sets different response time limits. The response time limit for high-risk early warnings is shorter than that for medium-risk and low-risk early warnings.

6. The AI-based, time-triggered intelligent push system for geriatric syndrome risk screening and home care as described in claim 1, characterized in that: The manual assessment interaction module is embedded into the hospital's existing EMR and nursing system as a plug-in. The visual interface provides operation options such as "Assess Now", "Remind Later", and "Confirm Known / No Assessment Required" as well as the function of one-click jump to the standardized electronic assessment form. The standardized electronic assessment form includes the Morse Fall Scale, the MNA-SF Nutritional Assessment Scale, the CAM Delirium Assessment Scale, the FRAIL Frailty Assessment Scale, and the Braden Pressure Ulcer Assessment Scale.

7. The AI-based, time-triggered intelligent push system for geriatric syndrome risk screening and home care as described in claim 1, characterized in that: The home care knowledge entries in the care knowledge base are organized and stored according to three dimensions: risk type, risk level, and care area. A "condition-template" matching mechanism is used to select the corresponding template based on the risk type and risk level in the assessment results. The variable fields in the template are filled in according to the score values ​​of each dimension in the assessment results to generate personalized care suggestion text. The care areas include home environment safety, daily activity guidance, nutritional support, medication management, cognitive care, pressure ulcer prevention, and excretion care.

8. The AI-based, time-triggered intelligent push system for geriatric syndrome risk screening and home care as described in claim 1, characterized in that: When generating the authorization confirmation page, the authorization management and home care push module displays a description of the scope of the push content to the patient or their family, clearly stating that it only includes key care points and risk prevention measures, and does not include original medical records, diagnostic information, or test data. The authorization is completed through electronic signature, and the system saves the authorizing person's name, authorization timestamp, and electronic signature hash value as compliance records.

9. The AI-based, time-triggered intelligent push system for geriatric syndrome risk screening and home care as described in claim 1, characterized in that: The home care recommendations include a cover, a comprehensive risk warning section, a section on specific care points, a section on emergency identification and medical treatment indications, and a section on follow-up / revisit recommendations. The content is transmitted using the TLS encryption protocol, and the push channels include the hospital's officially certified WeChat channel, SMS messages with secure links, or push messages from the hospital's APP for family members.

10. The AI-based, time-triggered intelligent push system for geriatric syndrome risk screening and home care according to any one of claims 1-9, characterized in that: The system also includes a model iteration module and a data security audit module. The model iteration module is used to collect labeled data of AI warnings from medical staff and support rule base adjustment and incremental model training. The data security audit module is used to de-identify and store all patient clinical data and record full-link audit logs. The system is deployed in the hospital's intranet environment, and all data processing and storage are completed within the hospital.

Images