Intelligent reminding and management system for medication compliance of chronic diseases of old people

Abstract

The invention relates to the technical field of elderly chronic disease medication management, and discloses an elderly chronic disease medication compliance intelligent reminding and management system. The system comprises an individualized medication modeling unit for establishing an individual pharmacodynamic model containing medication taboo and dosage rules based on an electronic health record of a patient and pharmacokinetic parameters; the multi-modal behavior sensing unit is used for collecting physiological data and movement tracks through wearable equipment, and generating a medication behavior sequence in combination with the record of the intelligent medicine box; the compliance deviation analysis unit is used for comparing an expected behavior with an actual behavior to calculate indexes such as a time matching degree and dose deviation; the risk decision engine unit is used for outputting an intervention level and a reminding strategy in combination with the deviation index, the physiological state and drug interaction; and the cross-platform collaborative execution unit triggers multi-channel reminding, medicine box locking and emergency contact notification. According to the system, the whole medication management process is intelligent, and the medication compliance and safety of the elderly patients are improved.

Description

Technical Field

[0001] This invention relates to the field of medication management technology for chronic diseases in the elderly, specifically to an intelligent reminder and management system for medication adherence in chronic diseases in the elderly. Background Technology

[0002] Currently, the elderly population suffering from chronic diseases continues to expand, with most requiring long-term, regular use of multiple medications to control disease progression. However, this group commonly experiences memory decline, insufficient understanding of medication effects and dosage requirements, and irregular daily routines, resulting in consistently low medication adherence. Missed doses, incorrect doses, duplicate doses, or unauthorized dosage adjustments are frequent occurrences. These problems not only directly impact the effectiveness of drug treatment, leading to relapses or worsening of the condition, but may also trigger adverse drug reactions, increase the probability of hospitalization, and exacerbate the medical burden on patients' families and the consumption of social medical resources.

[0003] To improve this situation, some medication reminder solutions have emerged in the existing technology, such as traditional alarm clocks, mobile app time push notifications, and basic smart pillboxes. However, these solutions generally suffer from limitations such as limited functionality, lack of individualized adaptability, and dynamic adjustment capabilities: traditional alarm clocks and mobile apps can only issue simple reminder signals based on preset times, and cannot combine information such as the patient's specific medical history, drug metabolism characteristics, and real-time physiological status to determine the necessity and urgency of the reminder, nor can they identify implicit adherence issues such as "receiving a reminder but not actually taking the medication" or "accidentally taking other medications"; while basic smart pillboxes can record the time the pillbox is opened, they cannot link it to the patient's activity trajectory, heart rate, blood pressure, and other physiological indicators, making it difficult to distinguish whether "not opening the pillbox is due to forgetting" or "being unable to take the medication due to physical discomfort," resulting in inaccurate judgment of the reasons for adherence deviation and failing to provide an effective basis for subsequent intervention.

[0004] Existing solutions generally lack cross-scenario collaboration capabilities and risk emergency response mechanisms: alert signals are often limited to single terminals (such as mobile phone ringtones or medicine box indicator lights), easily overlooked in scenarios where patients are out or in noisy environments; when significant adherence deviations (such as overdosing or long-term missed doses) or abnormal patient physiological indicators are detected, it is impossible to quickly coordinate with family members, community healthcare workers, and other relevant parties to initiate intervention, nor can it prevent dangerous medication behaviors through methods such as locking the medicine box, leading to delayed risk management and difficulty in responding to emergencies. In summary, current technologies still cannot form a complete management closed loop of "individualized modeling - dynamic perception - precise analysis - intelligent decision-making - collaborative execution," failing to meet the actual needs of elderly patients with chronic diseases for medication safety, convenience, and accuracy. A solution that can integrate multi-dimensional data and achieve intelligent management throughout the entire process is needed. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent reminder and management system for medication adherence in elderly patients with chronic diseases, in order to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides an intelligent reminder and management system for medication adherence in elderly patients with chronic diseases, the system comprising:

[0007] Personalized medication modeling unit: Based on the patient's medical history and pharmacokinetic parameters in the electronic health record, establish a personal pharmacodynamic model that includes contraindications and dosage adjustment rules;

[0008] Multimodal behavior perception unit: Continuously collects patients' physiological index fluctuation data and daily activity trajectory through wearable devices, and combines the opening records of the smart pillbox to generate a spatiotemporal feature sequence of medication behavior;

[0009] Compliance Deviation Analysis Unit: Dynamically compares the expected medication patterns output by the personal pharmacodynamic model with the actual behavioral sequences obtained by the multimodal behavior perception unit, and calculates the time matching degree, dose deviation index and behavioral abnormality confidence.

[0010] Risk Decision Engine Unit: Based on the deviation indicators generated by the compliance deviation analysis unit, combined with the patient's current physiological state and drug interaction matrix, it outputs decision instructions that include emergency intervention levels and adaptive reminder strategies;

[0011] Cross-platform collaborative execution unit: Based on the instructions of the risk decision engine unit, it synchronously triggers multi-channel reminder signals, medicine box locking mechanism and emergency contact notification link of smart terminal.

[0012] Preferably, the personalized medication modeling unit specifically includes:

[0013] Contraindication Feature Extraction Module: Analyzes liver and kidney function indicators, allergy history, and past adverse reaction records in the patient's electronic health record to construct a drug contraindication feature vector;

[0014] Metabolic parameter mapping module: Based on the patient's age, weight, and gene testing data, it matches the clearance rate and volume of distribution parameters in the pharmacokinetic database to generate an individualized metabolic parameter table;

[0015] Dosage rule generation module: Input the contraindication feature vector and individualized metabolic parameter table into the rule engine, combine it with the dosage adjustment logic in the drug instructions, and output a set of medication rules including the maximum safe dose, dosing interval and incremental gradient.

[0016] Preferably, the multimodal behavior sensing unit specifically includes:

[0017] Physiological indicator acquisition module: Captures electromyographic signals and autonomic nerve function changes before and after medication through a heart rate variability sensor and skin conductance electrodes in a wearable device;

[0018] Behavioral trajectory reconstruction module: Integrates motion data from indoor positioning beacons and inertial measurement units to reconstruct the patient's activity path and posture changes before and after medication;

[0019] Pillbox event logging module: Monitors the number of times the smart pillbox is opened, the force applied when taking the medicine, and the closing timestamp, and generates a medication operation log with timestamps.

[0020] Preferably, the compliance deviation analysis unit specifically includes:

[0021] Timing alignment module: Employs a dynamic window sliding algorithm to align the timestamps of the expected medication time point with the actual behavior sequence, compensating for clock skew between devices;

[0022] Dosage Difference Calculation Module: Compares the percentage deviation between the expected dosage and the actual amount taken from the medicine box, and calculates the cumulative deviation by combining the drug blood concentration curve;

[0023] Behavioral anomaly detection module: Analyzes the magnitude of physiological index changes and abnormal activity trajectories before and after medication, and generates a behavioral credibility score.

[0024] Preferably, the risk decision engine unit specifically includes:

[0025] Status assessment module: Based on current heart rate variability and skin conductance data, determine whether the patient's stress level has reached the preset physiological threshold;

[0026] Drug interaction detection module: Queries the risk level of incompatible combinations between the currently taken drug and other prescription drugs in the drug interaction matrix;

[0027] Strategy generation module: Based on deviation amount, stress level and risk level, generate a graded response plan ranging from voice prompts to automatic call to the emergency center.

[0028] Preferably, the cross-platform collaborative execution unit specifically includes:

[0029] Multi-channel alert module: synchronously drives the smart bracelet's vibration motor, mobile phone voice broadcast, and indoor smart light color changes;

[0030] Pillbox control module: When an overdose is detected, the electromagnetic locking mechanism is activated to limit the number of times the pillbox can be opened;

[0031] Emergency Contact Module: Sends alert information containing the patient's location and deviation details to preset contacts via encrypted communication protocol.

[0032] Preferably, the process of constructing the personal pharmacodynamic model includes performing:

[0033] The drug metabolism rate parameters are dynamically updated based on the patient's recent liver function test reports;

[0034] When a new allergy record is added to the electronic health record, the dimension weights of the contraindication feature vector are recalculated.

[0035] Based on feedback data regarding the actual amount of medicine taken from the pillbox, the recommended dosing interval is adjusted.

[0036] Preferably, the behavior trajectory reconstruction module includes the following:

[0037] The cumulative error of the inertial measurement unit is compensated by trilateration data from Bluetooth beacons;

[0038] An activity intensity classification model was used to distinguish medication use from other daily activities.

[0039] The event of the medicine box being opened is spatially matched and verified with the location coordinates.

[0040] Preferably, the multi-channel alert module includes the following functions:

[0041] The smart light blinking frequency is adjusted based on data from the ambient light sensor.

[0042] Adjust the repetition interval of voice broadcast based on the patient's historical response delay;

[0043] When the system detects that the bracelet is not being worn, it automatically switches to the phone's high-frequency vibration mode.

[0044] Preferably, the emergency contact module includes the function of performing:

[0045] Prioritize selecting the communication base station with the highest signal strength in the patient's activity trajectory over the past 30 minutes;

[0046] Embed anomaly data summary generated by the medication deviation analysis unit into SMS alert messages;

[0047] If the initial contact fails to respond, a sequence of phone calls to backup contacts is initiated.

[0048] Compared with the prior art, the beneficial effects of the present invention are:

[0049] From a personalized adaptation perspective, the personalized medication modeling unit does not employ a universal medication model. Instead, it is based on the specific medical history records and pharmacokinetic parameters in the patient's electronic health record, integrating contraindications and dosage adjustment rules into the individual pharmacokinetic model. This design allows medication management plans to fully adapt to individual patient differences. For example, for patients with weak liver or kidney function or slow drug metabolism, the model can pre-incorporate dosage adjustment rules to avoid the risk of drug accumulation caused by universal dosage regimens. For patients with a history of drug allergies or multiple underlying diseases, the model can clearly identify contraindications, reducing the risk of incorrect administration from the outset and making medication management more closely aligned with the patient's actual health condition.

[0050] In terms of the comprehensiveness and accuracy of behavioral perception, the multimodal behavioral perception unit breaks through the limitations of existing solutions that rely solely on single data (such as time or pillbox opening records). It continuously collects physiological indicator fluctuation data and daily activity trajectories through wearable devices, and then combines this with smart pillbox opening records to generate a spatiotemporal feature sequence of medication behavior. This multi-dimensional data integration can more accurately reconstruct the patient's medication scenario. For example, when the system detects that the smart pillbox is not opened but the patient's activity trajectory shows they are in a hospital and their physiological indicators are normal, it can determine that the patient may need to adjust their medication due to medical treatment without triggering a reminder. When the system detects that the pillbox is open but the patient's heart rate suddenly increases or their blood pressure is abnormal, it can further analyze whether there is a possibility of incorrect medication administration, avoiding misjudgments caused by relying solely on single data points and providing a more comprehensive basis for subsequent adherence analysis.

[0051] In terms of the depth of adherence analysis, the adherence deviation analysis unit does not simply compare "whether medication is taken on time," but dynamically compares the expected medication pattern output by the individual pharmacodynamic model with the actual behavioral sequence obtained by the multimodal behavioral perception unit. By calculating three dimensions of indicators—time matching degree, dose deviation index, and behavioral abnormality confidence level—it achieves refined identification of adherence problems. For example, when the time matching degree is high but the dose deviation index is abnormal, the system can identify the hidden problem of "the patient opens the pillbox on time but does not take the prescribed dose"; when the behavioral abnormality confidence level is high and combined with physiological indicator data, the system can distinguish whether "missed dose is due to forgetting" or "unable to take medication due to physical discomfort," allowing adherence analysis to go beyond surface behavioral judgments and delve into the essence of the problem, providing precise direction for subsequent interventions.

[0052] Regarding the rationality and timeliness of risk decision-making, the risk decision-making engine unit combines adherence deviation indicators, the patient's current physiological state, and the drug interaction matrix to output decision instructions, avoiding the "one-size-fits-all" intervention approach of existing solutions. For example, when the system detects a slight time deviation and the patient's physiological indicators are normal, it only outputs an adaptive strategy to adjust the reminder frequency; when it detects a severe dose deviation index and the patient experiences physiological abnormalities such as dizziness and nausea, it immediately outputs a high-level emergency intervention instruction, ensuring that the decision-making neither over-intervenes and increases the patient's burden nor overlooks dangerous situations. Simultaneously, the integration of the drug interaction matrix can avoid the risk of adverse reactions caused by patients taking multiple medications simultaneously, making the decision-making safer.

[0053] In terms of coordination and coverage during execution, the cross-platform collaborative execution unit synchronously triggers multi-channel reminder signals, a pillbox locking mechanism, and an emergency contact notification link based on decision commands, solving the problems of existing solutions having a single reminder method and insufficient emergency response capabilities. Multi-channel reminder signals (such as sound, vibration, and light from smart terminals, and vibration from wearable devices) can adapt to different scenarios such as patients at home, out and about, and noisy environments, reducing the probability of reminders being ignored; the pillbox locking mechanism can be automatically activated when the risk of overdose is detected, avoiding dangerous medication behavior; and the emergency contact notification link can quickly connect with family members and medical staff when patients experience serious adherence deviations or physiological abnormalities, ensuring that emergencies are handled promptly. Attached Figure Description

[0054] Figure 1 This is a schematic diagram illustrating the working principle of the intelligent reminder and management system for medication adherence in elderly patients with chronic diseases as described in this invention.

[0055] Figure 2 Flowchart for a personalized medication modeling unit;

[0056] Figure 3 A flowchart for a multimodal behavior sensing unit;

[0057] Figure 4 This is a timing analysis diagram of the operation behavior of the smart pillbox. Detailed Implementation

[0058] 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.

[0059] Please see Figure 1This invention provides an intelligent reminder and management system for medication adherence in elderly patients with chronic diseases. The system includes: a personalized medication modeling unit, a multimodal behavior perception unit, an adherence deviation analysis unit, a risk decision engine unit, and a cross-platform collaborative execution unit. These units are connected through a data interaction protocol to form a closed-loop control process. The personalized medication modeling unit imports patients' electronic health records from the hospital information system, analyzes medical history records and pharmacokinetic parameters, and uses modeling algorithms to construct a personal pharmacokinetic model. This model dynamically stores medication contraindications and dosage adjustment rules, providing a personalized baseline for the system. The multimodal behavior perception unit deploys wearable devices such as smart bracelets and smart pillboxes to continuously monitor physiological indicators such as heart rate variability and pillbox opening events. It combines positioning technology to generate spatiotemporal feature sequences, capturing patients' daily behavioral patterns. The adherence deviation analysis unit receives expected medication patterns and actual behavioral data, uses a time-series alignment algorithm for real-time comparison, and outputs indicators such as time matching degree and dosage deviation index to quantify the degree of behavioral deviation. The risk decision engine unit integrates a drug interaction database and a physiological state monitoring module, calculates risk levels based on deviation indicators, and generates tiered intervention strategies. The cross-platform collaborative execution unit links with the smart terminal through wireless communication protocols to perform multi-channel reminders, medicine box control, and emergency notification functions, ensuring that intervention measures are implemented in a timely manner.

[0060] Example 1: See Figure 2 The implementation of the personalized medication modeling unit is a systematic project involving multi-source data integration and intelligent processing. Its core objective is to construct a medication model that can truly reflect the individual characteristics of patients. This unit achieves its function through the cascading operation of a contraindication feature extraction module, a metabolic parameter mapping module, and a dosage rule generation module. The contraindication feature extraction module establishes a secure data interface with the electronic health record system of medical institutions, and regularly and automatically acquires or receives updated health information from patients in real time. This information includes structured laboratory test reports such as alanine aminotransferase (ALT) and aspartate aminotransferase (AST) levels in liver function indicators, and creatinine clearance rate and blood urea nitrogen values ​​in kidney function indicators. It also includes unstructured text records such as descriptions of allergy history and previous adverse drug reaction reports entered by clinicians. The module internally deploys a natural language processing engine to parse text records, identify entities and relationships related to drug contraindications, such as extracting "penicillin" as an allergen from "patient is allergic to penicillin" and converting it into standardized medical coding system terminology. At the same time, it extracts numerical indicators from structured data and compares them with preset clinical thresholds. When liver function indicators exceed the upper limit of the normal range, it is marked as having a risk of metabolic disorder. Finally, all features are vectorized, with each dimension corresponding to a contraindication type or risk factor and assigned a weighted value based on a medical knowledge base, thereby constructing a dynamically updated drug contraindication feature vector.

[0061] The metabolic parameter mapping module relies on a pre-built local or cloud-based pharmacokinetic database. This database contains typical metabolic parameters of various drugs validated in clinical studies across different populations. The module receives individual attribute data imported from a patient basic information database, including age, weight, height, sex, and optional genetic testing results such as CYP450 enzyme family polymorphism information. The mapping process is not a simple table lookup operation, but rather employs parameter estimation algorithms. For example, based on a population pharmacokinetic model, the patient's individual characteristics are used as covariates to individually adjust parameters such as typical clearance rate, volume of distribution, and half-life in the database. For an older, lighter patient, the algorithm will correspondingly lower the clearance rate estimate for certain drugs excreted by the kidneys. If genetic testing data is available, parameters related to the activity of specific metabolic enzymes will be further adjusted. Finally, an individualized metabolic parameter table for the patient is generated, which details the key parameters of absorption, distribution, metabolism, and excretion of various drugs currently being taken in a machine-readable format.

[0062] The dosage rule generation module, serving as the decision-making center of this unit, integrates a rule engine logically constructed based on clinical guidelines and drug instructions. The engine receives the outputs from the first two modules—the contraindication feature vector and the individualized metabolic parameter table—as input. Internally, the rule engine maintains a large rule base, with rules existing in a "condition-action" format. For example, a rule might stipulate that if the weight value of the "renal insufficiency" dimension in the contraindication feature vector exceeds a threshold X, and the clearance rate of a certain drug in the individualized metabolic parameter table is lower than a threshold Y, then a dosage adjustment action is triggered, reducing the recommended dose of that drug to a certain percentage of the standard dose and extending the dosing interval. The rule engine executes the reasoning process, traversing all relevant rules, comprehensively evaluating various contraindication risks and metabolic constraints, and ultimately outputting a structured set of medication rules. This rule set explicitly specifies the maximum safe dose for each drug, the recommended dosing interval, the gradient steps for dose escalation or descent, and a list of other drugs that should be avoided when using them together. The data flow of the entire personalized medication modeling unit adopts a highly reliable message queue mechanism to ensure the asynchronous and reliable data transmission between modules. The modeling process can be executed automatically according to a predetermined plan, or it can be immediately triggered to remodel when a major update to the electronic health record is detected. This ensures that the generated personal pharmacodynamic model is always synchronized with the patient's latest health status, providing an accurate and personalized benchmark for subsequent adherence analysis.

[0063] At the technical implementation level, the contraindication feature extraction module needs to address the heterogeneity of data. It may employ OWL ontology technology to construct a medical knowledge graph to unify medical terminology from different sources and achieve semantic interoperability. The weighting process for feature vectors may incorporate machine learning techniques, using historical medical record data to train the model and learn the relative importance of different contraindication features on medication safety. The metabolic parameter mapping module may use Bayesian estimation methods, treating individual patient data as prior information and combining it with population data from a pharmacokinetic database to obtain posterior parameter estimates. This method can better handle inter-individual variability and uncertainty. The rule engine for the dosage rule generation module may be implemented using commercial rule management systems such as Drools, allowing medical experts to maintain and update complex clinical rules through a graphical interface without modifying program code, enhancing the system's maintainability and adaptability. The output of the entire unit, the individual pharmacodynamic model, is typically stored in structured data formats such as XML or JSON for easy parsing and use by other units in the system. A model version management mechanism records the content and time of each update for tracking and auditing. Specifically, the XML format supports the definition of nested data elements through a tag-based structure, facilitating the accurate representation of multidimensional weights in contraindication features and key-value pairs in metabolic parameter tables. The JSON format, on the other hand, enables rapid serialization and deserialization using lightweight key-value pairs, allowing other system units (such as the compliance deviation analysis unit) to directly read and call model data through standard parsing libraries (such as DOM parsers or JSON parsers), integrating it into the dynamic comparison process without additional conversion. The model version management mechanism achieves tracking and auditing through integrated timestamp recording and change log functions. Whenever the model is recalculated due to updates to electronic health records or medication feedback data, the system automatically generates a new version identifier and records the specific content of the update (such as adjustments to the dimensions of the contraindication feature vector or corrections to metabolic rate parameters) along with a precise timestamp in the version database. This database employs an incremental storage strategy, saving only the differences rather than complete model copies, saving storage space and allowing for quick backtracking of any historical version of the model during auditing, ensuring the transparency and traceability of each model iteration. The entire process relies on a highly reliable message queue mechanism to ensure the integrity of data flow, enabling model storage and version management to be seamlessly embedded into the overall workflow of the unit.

[0064] The implementation of this unit also fully considers the practical constraints of the clinical environment. For example, the interface with the hospital information system needs to comply with medical information exchange standards such as HL7 and FHIR to ensure interoperability. Data access strictly adheres to privacy protection regulations, and all transmission and storage of patient health information are encrypted. The calculation process may be completed on the hospital's internal servers or through cloud computing resources that comply with standards such as HIPAA, ensuring both computational efficiency and data security. As the starting point and foundation of the entire intelligent medication management system, the accuracy and reliability of the personalized medication modeling unit directly affect the effectiveness of subsequent behavioral perception, deviation analysis, and intervention decisions. Therefore, its design emphasizes data integrity, intelligent processing, and the dynamic adaptability of the model. By continuously integrating the latest patient information, it strives to ensure that medication recommendations are as closely aligned as possible with the patient's real-time physiological state and pathological changes.

[0065] Example 2: See Figure 3 The implementation of the multimodal behavior perception unit relies on the collaborative deployment of multiple sensing devices and the refined processing of data fusion algorithms. Its core task is to continuously capture physiological, activity, and medication box operation data related to medication use and generate behavioral feature sequences with spatiotemporal context. The physiological indicator acquisition module is implemented through a customized wearable device. This device integrates a medical-grade photoplethysmography sensor for monitoring heart rate variability and is equipped with a high-precision skin conductance electrode to measure changes in skin conductivity. The device firmware is set to continuously acquire electromyographic signals at a specific frequency (e.g., once per second) and performs preliminary signal preprocessing on the device, including motion artifact filtering and baseline drift correction. The preprocessed data is transmitted to a smart terminal or home gateway in the form of data packets via Bluetooth Low Energy at regular intervals. A circular buffering mechanism is used during transmission to prevent data loss. After receiving the data, the agent program on the smart terminal further performs time-domain and frequency-domain analysis to extract feature values ​​reflecting the state of autonomic nervous function, such as the low-frequency to high-frequency power ratio, forming a physiological indicator time series associated with the medication time point.

[0066] The operation of the behavior trajectory reconstruction module is based on the integration of multi-source positioning and inertial navigation technologies. The patient's activities in the home environment are located by Bluetooth Low Energy beacons deployed in key locations in the room. Each beacon continuously broadcasts a signal containing its own location information. The smart bracelet or special tag worn by the patient receives these signals and calculates the received signal strength indication value. The approximate position coordinates of the patient relative to the beacons are estimated by the trilateration algorithm. At the same time, the inertial measurement unit built into the smart bracelet collects raw data from the three-axis accelerometer, three-axis gyroscope and three-axis magnetometer at a higher frequency. These data are used to calculate the device's attitude angle and motion trajectory by sensor fusion algorithms (such as complementary filtering or Kalman filtering). The absolute position information provided by the Bluetooth beacon is used to periodically correct the trajectory drift caused by the integral error of the inertial measurement unit. For example, the cumulative error is reset every 30 seconds using the beacon positioning result, thereby reconstructing the continuous path of the patient's movement between different rooms. The module runs a pre-trained activity recognition and classification model, which analyzes the time-frequency characteristics of the inertial measurement unit data and can distinguish typical actions such as walking, sitting, standing, and reaching for objects. It pays special attention to specific arm movement patterns related to medication behavior. When the smart pillbox generates an opening event, the behavior trajectory reconstruction module will perform spatial matching verification between the timestamp of the event and the currently estimated patient position coordinates to determine whether the patient is near the pillbox, thereby increasing the credibility of the behavior record.

[0067] The pillbox event logging module is embedded within the smart pillbox hardware. Each pill compartment is equipped with a touch switch or photoelectric sensor to detect the opening and closing status of the compartment door. Simultaneously, a high-precision weighing sensor mounted on the base of the pillbox monitors the weight change before and after pill removal to estimate the amount of medication dispensed. The pillbox's main microcontroller continuously polls the status of these sensors; any change in status (such as door opening or significant weight reduction) is recorded as an event with a high-precision timestamp. Event data is temporarily stored in the pillbox's local non-volatile memory. The smart pillbox maintains a connection to a cloud server via Wi-Fi or cellular network, periodically uploading event records in batches to form a time-sorted medication operation log. The log includes information such as event type, occurrence time, involved compartment identifier, and estimated number of pills. The data synchronization mechanism of the multimodal behavior perception unit adopts a time synchronization scheme based on the network time protocol to ensure that the data from wearable devices, positioning beacons and smart pillboxes have a unified time reference. The data fusion processor inside the unit aligns and correlates the time series of physiological indicators, the reconstructed activity path and posture information and the pillbox operation log to generate a structured spatiotemporal feature sequence of medication behavior. This sequence not only records the simple event of "whether to take the medicine", but also includes the physiological reactions before and after taking the medicine, the context of physical activity and the details of operating the pillbox, providing a rich data foundation for subsequent in-depth analysis of medication adherence.

[0068] In terms of hardware selection and deployment, Bluetooth beacons are typically installed in areas frequently used by patients, such as living rooms, bedrooms, and kitchens, as well as near medicine box storage locations. The installation location and signal coverage of the beacons need to be optimized through on-site surveys to reduce blind spots. Wearable devices need to consider the comfort and battery life of the elderly, and may adopt a wristband design with wireless charging support. The design of smart medicine boxes emphasizes user-friendly human-computer interaction, featuring a large-font display and simple buttons, while ensuring the accuracy and stability of the weighing sensors. On the software algorithm front, activity recognition and classification models are typically trained on large labeled activity datasets using machine learning techniques, enabling them to identify typical medication-related actions with high accuracy. Sensor fusion algorithms are continuously optimized to balance computational complexity and trajectory accuracy, adapting to magnetic field interference and signal reflection issues in different home environments. Data privacy and security are maintained throughout the implementation process. All sensor data is encrypted on the device before transmission, data stored in the cloud is anonymized, and access control is strictly limited, complying with medical data protection regulations. The multimodal behavior perception unit integrates multi-dimensional information such as vision, motion, and object interaction to construct a system that comprehensively digitally maps medication behavior. The spatiotemporal feature sequences it generates surpass the single-point information of traditional medication records, revealing the potential correlation between behavioral patterns and physiological states, and providing indispensable perception layer support for intelligent adherence management.

[0069] See Figure 4In the visualization analysis of the medication adherence management system for elderly patients with chronic diseases, the dynamic patterns of smart pillbox operation behavior are displayed through multi-layer time-series data fusion technology. Specifically, a dual-axis coordinate system is used to present multi-dimensional indicators in parallel: the pillbox operation frequency statistics chart focuses on frequency analysis, with the black line (number of openings) recording the daily successful opening frequency of the pillbox, fluctuating between 1.0 and 4.0 times, reflecting the basic frequency of daily medication use. The gray cross line (number of erroneous operations) serves as a key risk indicator, used to capture non-routine operations (such as accidental touches, multiple invalid attempts, etc.), with its value fluctuating between 0.0 and 2.0 times. Any non-zero peak may indicate fluctuations in the patient's cognitive state or obstacles in the human-computer interaction interface. The comparison of the two lines can preliminarily determine the regularity and stability of medication behavior. The pillbox operation quality indicator chart provides a deeper portrayal from the dimensions of accuracy and efficiency. The diamond-shaped line graph (medication accuracy) quantifies the proportion of correctly dispensed medication in each operation, with its fluctuation range of 70% to 100% directly correlated with medication adherence. The circular line graph (average opening time), on the other hand, indirectly reflects the patient's proficiency, cognitive load, or physical condition through operation time ranging from 10 to 30 seconds. Abnormally prolonged opening times may be highly correlated with patient mobility impairments or inattention. By comparing time series data, the macroscopic operational events are correlated with microscopic operational quality, transforming isolated medication events into continuous behavioral trajectories. This provides a data foundation for optimizing personalized reminder strategies and providing early warnings of adherence deviations.

[0070] Example 3: The joint implementation of the compliance deviation analysis unit, risk decision engine unit, and cross-platform collaborative execution unit is aimed at the continuous operation of home and institutional medication scenarios. Using the expected medication pattern given by the personal pharmacodynamic model as a reference, a closed-loop process of alignment, deviation summarization, anomaly identification, level determination, and graded execution is carried out around the spatiotemporal characteristic sequence formed by actual behavior and physiological observation. Data is transferred between units with time slices as the main line and tracked and recorded at the event level, which is used as the basis for subsequent investigation and audit review.

[0071] In the timing alignment module, the system constructs a sliding window with the expected medication time point as the anchor point. It merges the smart pillbox opening record, medication taking and taking action markers, heart rate variability and skin conductance response change points of wearable devices, and attitude switching times reconstructed by indoor positioning and inertial measurement unit into candidate event sequences. For each anchor point, it searches for the alignment position with the minimum comprehensive cost within the window range. The comprehensive cost consists of three components: the absolute value of the event time difference, the penalty term for event type matching, and the compensation term for device time drift. The window length is adaptively adjusted according to the event density and intraday medication pattern of the past week. The output is a structured result consisting of the aligned behavior subsequence and its weight, with confidence annotations. The confidence annotations are derived from the signal quality assessment rules so that subsequent steps can perform noise reduction processing during weight fusion.

[0072] In the dose difference calculation module, the system aggregates and aligns the drug dispensing readings from the pillboxes, removes minor opening logs caused by mechanical vibration and accidental touches, and determines a valid drug dispensing event by the opening duration, force variation, and interval between two openings. Then, it evaluates the relative deviation between the valid drug dispensing amount and the expected dosage given by the individual pharmacodynamic model, and uses the influence range of the blood drug concentration curve on the time axis as a weighting function to accumulate and summarize multiple deviations. If there are cases of missed doses or premature over-dispensing, the module adds the weighting strategy to the interval between two adjacent dispensings and the coverage of the concentration rise and fall phases to reflect the real impact across time periods. Finally, it generates two outputs: dose deviation index and cumulative deviation amount, and retains the one-to-one correspondence between events and values ​​for risk assessment.

[0073] In the behavioral anomaly detection module, the system makes judgments based on the physiological and kinematic characteristics before and after medication. Heart rate variability is extracted using short-window non-overlapping statistics to extract time-domain indicators and rhythm fluctuation amplitude. Skin conductance response is correlated with the strength of autonomic nervous system response using step detection and recovery time estimation. The path reconstructed by the indoor positioning and inertial measurement unit identifies anomalies by curvature, dwell time, and attitude switching frequency. After the above features are processed by removing extrema and scaling, they are mapped to a behavioral credibility score. A high score means that the observed physiological and behavioral changes are more consistent with a real medication process, while a low score means that the probability of suspicious behavior or alternative actions is higher. The module outputs the behavioral credibility and its confidence interval and writes it into the time slice record.

[0074] In the state assessment module, the system reads heart rate variability and skin conductance observations within the current time slice, calculates stress level relative to the individual baseline, and performs contextual correction by combining recent sleep quality and activity level. When the stress level reaches the physiological threshold, it is marked as an acute state. If it does not reach the threshold, it is mapped to continuous intensity values ​​in a segmented manner for parameter tuning of the grading strategy. There is an interrelationship between the state assessment results and the dose deviation index. The module jointly labels the two through a rule table to avoid over-intervention in cases of short-term high stress but small dose deviation.

[0075] In the drug conflict detection module, the system reads the current prescription list and retrieves the combination risk in the drug interaction matrix. The matrix includes contraindication levels, precautions for combined use, and key points for clinical monitoring. The retrieval output is converted into a conflict intensity value between zero and one. Zero represents no known conflict detected, and one represents a combination that is highly unsuitable for combined use. If multiple combinations are effective at the same time, the module adopts the upper bound priority synthesis rule and records the triggering entry and source version number for medical staff to trace and verify.

[0076] In the strategy generation module, the system inputs alignment indicators, dose deviation index, behavioral credibility, stress level, and conflict intensity into the grading mapping. The core of the grading mapping is a judgment process that combines a linear summary score with a threshold table. An adaptive reminder parameter is attached to the output to reflect the impact of historical response delay and recent deviation frequency. The calculation of the linear summary is implemented in a simplified weight form in this embodiment for easy deployment and maintenance. The calculation of the linear part is expressed as follows:

[0077] L=pT+qG+rH+sM+tI

[0078] The meanings of all characters in this expression are as follows: L represents the comprehensive score used to determine the intervention level; p represents the time matching degree weight coefficient configured by the strategy table and its value is between zero and one; T represents the normalized dimension-consistent value of the time matching difference, which is between zero and one; q represents the dose deviation weight coefficient configured by the strategy table and its value is between zero and one; G represents the standardized value of the dose deviation index, which is between zero and one; r represents the behavior credibility weight coefficient configured by the strategy table and its value is between zero and one; H represents the inverse vector of behavior credibility, i.e., the standardized value of one minus credibility, which is between zero and one; s represents the physiological state weight coefficient configured by the strategy table and its value is between zero and one; M represents the stress level intensity value after segmented mapping, which is between zero and one; t represents the drug conflict weight coefficient configured by the strategy table and its value is between zero and one; I represents the conflict intensity value obtained by the interaction matrix retrieval, which is between zero and one. The weight coefficients are set by the operation and maintenance side according to the population and drug category and the source and effective time are marked in the version record for auditing and rollback requirements.

[0079] After the linear score calculation is completed, the strategy generation module maps the emergency intervention level according to the threshold table and generates an adaptive reminder strategy. The threshold table contains multiple levels of intervals and corresponding execution templates. The execution templates list the reminder channel combination, the upper limit of the repetition interval, the triggering order of light and vibration, the action type and duration of the pillbox control, and the waiting time and escalation order of communication triggers. The adaptive reminder strategy reads the individual's historical response delay and automatically shortens or extends the repetition interval within the allowable range to balance the relationship between disturbance and achieving the action. If multiple high score intervals occur in a recent period, the adaptive logic will also appropriately increase the intensity of the first reminder and gradually drop it back to the normal level after the medication is taken.

[0080] In the multi-channel reminder module, the system simultaneously coordinates the smart bracelet, mobile terminal, and indoor smart lights. The bracelet triggers the vibration motor to execute according to the pulses and intervals given by the strategy template. The mobile terminal broadcasts a voice message and displays a reminder card when the screen is locked. The indoor lights switch colors and flashing rhythms according to the strategy template and dynamically adjust the flashing frequency based on ambient light sensor data. If the system detects that the bracelet is not being worn, the multi-channel reminder module automatically switches to a combination of high-frequency vibration from the mobile terminal and external voice message, and increases the screen lighting time. The reminder execution results are written to the log, including the start and end times and user interaction actions such as button confirmation or voice response, so that the learning module can update the individual response delay statistics later.

[0081] In the pillbox control module, the system schedules the electromagnetic locking mechanism based on the output of the strategy generation module and the real-time trend of the dosage deviation index. When there is a risk of overdose or a trend of multiple openings in a short period of time, the locking action is triggered. The locking duration and unlocking conditions are given by the corresponding level template. Common unlocking conditions include reaching the next recommended dosing time, completing risk confirmation through a mobile terminal, and receiving a remote unlocking command from the medical staff through the human-machine interface. Each locking and unlocking action records a timestamp, trigger source, and result status. If mechanical pulling or impact occurs during the locking period, the module will record the abnormal event and the force sensor amplitude and increase the weight of the risk score in the next time slice to prompt the strategy generation module to select a stronger reminder and communication solution.

[0082] In the emergency contact module, the system initiates an early warning communication when no valid confirmation is received from the user or no medication is detected, based on the waiting time set by the strategy template. The mobile terminal registers with the communication base station with the strongest signal strength based on the activity trajectory and mobile signal observation of the past 30 minutes. The early warning information is sent to the preset contact through an encrypted channel and includes a standardized data digest. The digest includes the anchor time of the most recent alignment, the linear score of the current time slice, the dose deviation index, the discrete description of behavioral credibility and stress level, the interaction status and location information. If no response is received within the waiting time, the module dials the phone number one by one according to the contact sequence and marks the connection status and duration in the log. If the loop is not closed after one round of contact, the module enters a delayed retry and moves the threshold selection in the strategy generation module to a higher level range.

[0083] In terms of cross-unit data governance, the system maintains structured records for each time slice. The items include event lists, signal quality, alignment weights, intermediate values ​​for dose deviation calculation, summary hashes of behavioral credibility feature vectors, intervals for linear scores and threshold hits, execution template version numbers, results and time consumption of reminders and control actions. Records adopt a segmented retention strategy to balance traceability and storage costs. Data involving personal privacy is encrypted and stored end-to-end. Cross-platform transmission completes consistency confirmation through short-term credentials and double-end verification. The operation and maintenance side can perform offline simulation of weight tables and threshold tables in a security sandbox and distribute new versions in a gray-scale ratio. During runtime, each module performs behavioral consistency verification according to version number and rolls back to the previous stable version in case of anomalies.

[0084] Regarding parameter updates and adaptation, the system summarizes the time distribution between reminder execution and medication achievement at fixed intervals, updates individual response delay estimates and writes them back to the strategy generation module. The adaptive logic adjusts the repetition interval of voice broadcasts and the intensity of vibration pulses accordingly. When normal achievement is achieved for several consecutive days, the system fine-tunes the lowest range of the linear score threshold to reduce unnecessary channel occupation. In the case of frequent high scores, the system outputs prompts to the medical staff to suggest reviewing prescriptions and contraindications. Alarms and event statistics of each module are uploaded after generating summaries locally. The backend only retains necessary version and range information to avoid the risk of long-term leakage of original physiological data. Thus, this embodiment completes the description of the feasible implementation of the compliance deviation analysis unit, risk decision engine unit and cross-platform collaborative execution unit in a real scenario and realizes a closed-loop management process from observation to execution with a simple and readable linear score as the core connection point.

[0085] Example 4: The dynamic update mechanism in the personal pharmacodynamic model construction process is the core of ensuring that medication recommendations are always synchronized with the patient's latest health status. This process triggers model recalculation and optimization in response to specific events, including regular updates to physiological indicators, new clinical diagnostic records, and actual medication feedback from the medication kit. Taking a virtual elderly patient, Mr. Wang, as an example, he suffers from hypertension and type 2 diabetes and has been taking antihypertensive drug A and hypoglycemic drug B for a long time. The initial personal pharmacodynamic model built for him by the system is based on his physical examination data from three months ago. When the hospital's laboratory information system generates Mr. Wang's latest liver function test report, the system automatically retrieves this report through a secure data interface. The report shows that his alanine aminotransferase (ALT) level has significantly increased compared to the previous test, suggesting potential hepatocellular damage. This change triggered the model update process, invoking the metabolic parameter mapping module. It input Mr. Wang's new ALT value along with other liver function indicators into an internal algorithm. Based on the known relationship between liver metabolic capacity and drug clearance rate, the algorithm re-estimated the metabolic rate parameters of drug A, which is mainly metabolized by the liver. Specifically, the algorithm lowered the expected clearance rate of drug A and correspondingly extended its estimated half-life. This means that, at the same dose, the duration of action of drug A in Mr. Wang's body may be prolonged, increasing the cumulative risk.

[0086] During the model update cycle, Mr. Wang's electronic health record was updated by his attending physician, adding a new record indicating that the patient recently developed a rash after taking a new joint pain reliever, C, which the doctor recorded as a suspected drug allergy. This new allergy record was captured by the contraindication feature extraction module. The module initiated a natural language processing flow to parse the text record, identifying the two key entities "drug C" and "rash," and mapping them to allergy reaction types in the knowledge base, confirming it as a drug allergy. Subsequently, the module recalculated the contraindication feature vector, assigning a higher weight value to the dimension representing "drug C allergy" in the vector. At the same time, the system also assesses whether there is a risk of cross-allergy between drug C and drugs A and B in Mr. Wang's existing medication regimen. Even if A and B do not currently cause an allergy, if their chemical structures are similar to C, the system may add a lower warning weight to A and B in the feature vector. This update ensures that the model remains highly sensitive to Mr. Wang's allergy history.

[0087] The dosage rule generation module integrates the updated metabolic parameters and contraindication feature vectors, along with feedback data from the pillbox event recording module. The system detected a pattern in Mr. Wang's actual medication B dosage: on numerous occasions, the amount taken from the pillbox was only half the prescribed dose. This persistent dosage deviation was considered a significant behavioral feedback. When regenerating medication rules, the rule engine considers this behavioral pattern. It doesn't simply label it as poor adherence but initiates an analysis process to explore possible causes behind this deviation, such as whether the standard dose caused Mr. Wang discomfort (e.g., hypoglycemia), leading him to reduce the dosage himself. Based on this, the engine might output a revised recommended dosing interval, for example, suggesting that the doctor consider adjusting the twice-daily, one-tablet regimen of medication B to a once-daily, two-tablet regimen. If pharmacokinetic characteristics allow, this adjustment might better suit the patient's actual tolerance and behavioral habits, thereby improving adherence while achieving therapeutic effects. Table 1 shows the dynamic changes in key parameters of Mr. Wang's personal pharmacokinetic model after receiving the above three types of information.

[0088] Table 1: Updated parameters of Mr. Wang's personal pharmacokinetic model

[0089]

[0090] The entire update process takes place in a controlled environment. The system retains historical versions of the model and records the reason for each update, timestamp, and specific data of the triggering event. This version management helps doctors trace the trajectory of model parameter changes and understand the origin of adjustments to the system's recommended treatment. After the model update is complete, the new personal pharmacodynamic model takes effect immediately, replacing the old model and providing a new, more accurate baseline for the expected medication pattern in the adherence deviation analysis unit. For example, the system will adjust the expected blood drug concentration curve based on the reduced clearance rate of drug A, thereby more accurately determining the optimal time window for the next dose. Or, when the model detects that Mr. Wang is attempting to remove drug C, it will immediately trigger the highest level of contraindication alert. Through this continuous, cyclical perception-update-optimization mechanism, the personal pharmacodynamic model transforms from a static initial setting into a dynamic entity that evolves continuously with changes in the patient's health status and behavioral feedback, greatly improving the individualization and adaptability of the entire medication management system.

[0091] Example 5: The implementation of the multi-channel alert module and emergency communication module focuses on the final execution stage of the system intervention strategy. Its core design aims to achieve intelligent adaptability of alerts and robust reliability of communication. The operation of the multi-channel alert module is not a simple signal broadcast, but an intelligent process that dynamically adjusts based on environmental context and user status. The module receives decision instructions from the risk decision engine unit, which include the urgency level and type of the alert, but the specific execution parameters are adaptively generated by the module. Ambient light sensor data is read in real time. When the ambient light level is below a certain threshold, it is determined to be in a dim or nighttime environment. At this time, the intelligent lighting strategy will adjust from color cues (such as changing to blue) to a brighter white flash, and the flashing frequency will be appropriately reduced to avoid overstimulation. Conversely, in well-lit environments, information is mainly conveyed through color changes. The repetition interval adjustment mechanism for voice broadcasts relies on a simple learning process. The module records the average delay time from the first broadcast to the patient's confirmation response via the medicine box or mobile terminal button in historical reminder events. Based on this historical delay data, the module sets an initial repetition interval, for example, 80% of the average delay time. If no response is received within this interval after the reminder is issued, the system gradually shortens the interval of subsequent repetition reminders, creating a gradual urging effect. The device status detection function ensures the availability of the reminder channel. The module periodically checks the Bluetooth connection signal strength with the smart bracelet and the data from the bracelet's built-in accelerometer to determine whether it is being worn. If the signal strength is found to be extremely low and the accelerometer data remains unchanged for several consecutive detection cycles, it is inferred that the bracelet is not being worn. At this time, the reminder strategy automatically removes the tactile vibration channel from the execution list and transfers the reminder intensity originally intended for the bracelet to the high-frequency vibration mode of the mobile phone. The mobile phone vibration mode simulates a rapid rhythm to attract attention.

[0092] The emergency contact module focuses on ensuring that warning information is accurately and promptly delivered to pre-defined contacts in critical situations. Upon activation, the module executes a communication channel optimization process. This process queries locally stored patient activity trajectory data, provided by the behavior trajectory reconstruction module. This data includes timestamps of the patient's stays at different locations over a period of time, along with the cellular network base station identifiers and signal strengths detected at those times. The module extracts trajectory points from the past 30 minutes, analyzes the signal strength of the base stations associated with each point, and prioritizes the base station with the highest signal strength as the primary communication base station for connection. This approach aims to improve the initial connection success rate and data transmission stability. The construction of the warning information includes structured data filling. Key information fields are reserved in the SMS or instant message templates. The emergency contact module extracts a summary of core abnormal data from the real-time results generated by the compliance deviation analysis unit. For example, "A heart rate drop to 45 beats / min was detected, accompanied by a suspected record of repeated β-blocker use; the most recent pillbox opening was 3 minutes ago." This automatically generated, non-human-written descriptive text is embedded in the warning information, forming a complete warning message along with the patient identifier and real-time geographic coordinates. The communication protocol employs industry-standard encryption methods for end-to-end information protection. The contact sequence management logic handles potential initial contact failures. After sending an alert to the primary contact, the module starts a configurable waiting timer (e.g., 120 seconds). If no confirmation (e.g., SMS reply, in-app confirmation) is received before the timer expires, the module determines the contact attempt has failed and automatically initiates a call sequence for backup contacts. The sequence attempts to call the next contact in a pre-defined priority order, with each contact's call attempt having its own independent timeout control. The entire contact process will only stop when a contact answers the phone and confirms receipt of the alert; otherwise, it will continue looping until all backup options have been tried.

[0093] Multi-channel alert modules require deep integration with different smart home platforms and mobile operating systems. They typically control hardware such as lights and vibration motors through their respective open application programming interfaces (APIs). This necessitates an internal device abstraction layer to manage different types of terminal devices in a unified manner. The base station selection algorithm for the emergency contact module may involve limited data interaction with mobile network operators or decision-making based on historical field strength measurement reports at the device end. The execution of its call sequence may rely on a reliable voice call interface. Both modules share a high degree of scenario-driven characteristics. Their behavioral logic is entirely determined by the output of the upstream analysis unit and the real-time environment and device status. The aim is to maximize the effectiveness of interventions through intelligent execution strategies and ensure patient safety at critical moments through redundant communication methods.

[0094] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0095] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent reminding and management system for medication adherence of the elderly with chronic diseases, characterized in that, The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine.

2. The intelligent reminding and management system for medication adherence of the elderly with chronic diseases according to claim 1, characterized in that, The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. 3.The intelligent reminding and management system for medication adherence of the elderly with chronic diseases according to claim 2, characterized in that, The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine.

4. The medicine adherence intelligent reminding and management system for the elderly chronic diseases according to claim 3, characterized in that, The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine.

5. The medicine adherence intelligent reminding and management system for the elderly chronic diseases according to claim 4, characterized in that, The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application relates to a medication compliance monitoring system based on individualized pharmacokinetics modeling, multi-modal behavior sensing, compliance deviation analysis and risk decision engine. The application Strategy generation module: integrate deviation amount, stress level and risk level to generate a hierarchical response plan from voice reminder to automatic call to emergency center.

6. The medicine adherence intelligent reminding and management system for the elderly chronic diseases according to claim 5, characterized in that, The cross-platform collaborative execution unit specifically includes: Multi-channel reminder module: synchronously drive the vibration motor of the smart bracelet, the voice broadcast of the mobile phone, and the color change of the indoor intelligent light; Medicine box control module: activate the electromagnetic locking mechanism to limit the opening times of the medicine box when detecting overdosage; Emergency contact module: send early warning information containing patient location and deviation details to the preset contact through an encrypted communication protocol.

7. The medicine adherence intelligent reminding and management system for the elderly chronic diseases according to claim 1, characterized in that, During the construction of the personal pharmacokinetic model, the following steps are included: Based on the patient's recent liver function test report, dynamically update the drug metabolism rate parameter; When the electronic health record adds an allergy record, recalculate the dimension weight of the contraindication feature vector; According to the actual feedback data of the medicine box, correct the recommended value of the dosing interval. 8.The intelligent reminding and management system for medication adherence of the elderly with chronic diseases according to claim 3, characterized in that, In the behavior trajectory reconstruction module, the following steps are included: Compensate for the cumulative error of the inertial measurement unit through the three-edge positioning data of the Bluetooth beacon; Use the activity intensity classification model to distinguish between medication behavior and other daily actions; Spatially match the medicine box opening event with the positioning coordinates for verification. 9.The intelligent reminding and management system for medication adherence of the elderly with chronic diseases according to claim 6, characterized in that, In the multi-channel reminder module, the following steps are included: Adjust the flicker frequency of the intelligent light according to the ambient light sensor data; Adjust the repetition interval of the voice broadcast based on the patient's historical response delay; When the bracelet is not detected, automatically switch to the mobile phone high-frequency vibration mode. 10.The intelligent reminding and management system for medication adherence of the elderly with chronic diseases according to claim 1, characterized in that, In the emergency contact module, the following steps are included: Prioritize the communication base station with the highest signal strength in the patient's recent 30-minute activity trajectory; Embed the abnormal data summary generated by the medication deviation analysis unit in the SMS early warning information; When the first contact does not respond, sequentially start the phone call sequence of the backup contact.

Images