A smart medication management method and system based on medication behavior

Abstract

This invention belongs to the field of medical informatics and relates to an intelligent medication management method and system based on medication behavior, comprising: Step 1, acquiring multimodal medication data and matching the medication data to a drug knowledge base to obtain multimodal feature representations of candidate drugs; Step 2, performing cross-modal consistency verification on candidate drugs based on multimodal feature representations to obtain a target drug set and target drug features; Step 3, constructing a behavior status determination model and an adherence risk prediction model based on the pressure changes in the target drug inventory to obtain medication behavior status and adherence risk level; Step 4, constructing intervention records based on medication behavior status, adherence risk level, and actual medication behavior; Step 5, processing the intervention records and generating prescription optimization suggestions based on the data processing results; thereby achieving closed-loop management of medication adherence in elderly patients.

Description

Technical Field

[0001] This invention relates to the field of medical informatics, and specifically discloses an intelligent medication management method and system based on medication behavior. Background Technology

[0002] Elderly patients often have multiple coexisting diseases, require long-term medication, take various medications, and have complex medication schedules. Medication management has become a crucial aspect of chronic disease management and home-based health management. In cases of chronic diseases such as hypertension, diabetes, and coronary heart disease, missed doses, incorrect doses, duplicate doses, or unauthorized discontinuation of medication can not only affect treatment outcomes but may also induce fluctuations in the patient's condition, adverse drug reactions, and even increase the risk of readmission.

[0003] Existing medication reminder products for home use mainly include alarm clocks, mobile app reminders, and smart pillboxes. However, they have significant shortcomings in actual use: First, the interaction methods are not suitable for the elderly, relying on mobile phone operation and standard Mandarin input, which is not suitable for elderly people with declining vision, hearing loss, or who speak dialects. Second, the drug recognition ability is insufficient, relying solely on manual input and unable to automatically complete prescription mapping, making it difficult to identify drugs with similar packaging. Third, the action of taking medication is disconnected from the actual medication behavior; it can only determine whether the lid has been opened, but cannot distinguish between normal medication taking, accidental touch, and repeated medication taking. Fourth, the reminder mechanism is static and rigid, failing to dynamically adjust according to individual patient characteristics, which can easily cause annoyance or fail to correct missed doses. Fifth, the information is isolated, failing to provide decision support for clinical practice. Sixth, adherence management is disconnected from clinical prescription optimization, merely recording behavior without optimizing treatment plans.

[0004] Therefore, this invention proposes an intelligent medication management method and system based on medication behavior, which realizes closed-loop management of medication adherence in elderly patients through multimodal age-friendly interaction, cross-modal consistency verification, stress-sensing behavior recognition, progressive adaptive intervention, and prescription optimization feedback. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent medication management method and system based on medication behavior, addressing the following problems: how to overcome the shortcomings in medication adherence management for middle-aged and elderly patients, such as unfriendly interaction, inaccurate identification, insufficient behavior judgment, rigid reminder strategies, inadequate doctor-patient collaboration, and lack of closed-loop optimization; the specific solution is as follows: A smart medication management method based on medication behavior includes: Step 1, acquiring multimodal medication data and matching the data to a drug knowledge base to obtain multimodal feature representations of candidate drugs; Step 2, performing cross-modal consistency verification on candidate drugs based on the multimodal feature representations to obtain a target drug set and target drug features; Step 3, constructing a behavior status determination model and a compliance risk prediction model based on the pressure changes in the target drug inventory to obtain medication behavior status and compliance risk level; Step 4, constructing intervention records based on medication behavior status, compliance risk level, and actual medication behavior; Step 5, processing the intervention records and generating prescription optimization suggestions based on the data processing results.

[0006] Further, the process includes: Step 1.1, where the patient obtains multimodal medication data through a terminal device; the medication data includes drug image data, voice description data, and prescription data; Step 1.2, matching the drug image data with drugs in a drug knowledge base to obtain a set of image candidate drugs; Step 1.3, matching the voice description data with drugs in a drug knowledge base to obtain a set of voice candidate drugs; Step 1.4, performing structured processing on the patient's prescription data to obtain a set of prescription candidate drugs; Step 1.5, identifying the image candidate drug set, the voice candidate drug set, and the prescription candidate drug set as candidate drugs, and processing the candidate drugs to obtain a multimodal feature representation of the candidate drugs.

[0007] Further, the method includes: Step 2.1, calculating the single-modal similarity of candidate drugs to obtain the multimodal matching degree of candidate drugs; the multimodal matching degree includes image matching degree, semantic matching degree, and prescription matching degree; Step 2.2, based on the multimodal matching degree, performing a multimodal consistency comprehensive score on candidate drugs to obtain the consistency score of candidate drugs; Step 2.3, based on the multimodal matching degree, performing conflict detection on candidate drugs to obtain conflict markers; Step 2.4, based on the consistency score of candidate drugs, conflict markers, and prescription data, obtaining the target drug set and target drug features.

[0008] Further steps include: Step 3.1, collecting pressure changes in the target drug compartment of the medicine box to obtain a pressure sensing sequence, and extracting features from the pressure sensing sequence to construct pressure-sensing behavioral features; Step 3.2, recording the actual drug dispensing time, and calculating the medication time deviation based on the actual drug dispensing time and the medication time in the prescription data; Step 3.3, extracting features from the pressure-sensing behavioral features and the medication time deviation to obtain a set of medication behavior states; Step 3.4, constructing a behavior state determination model based on the set of medication behavior states to obtain the medication behavior state of the current time window; Step 3.5, constructing a compliance risk prediction model based on the medication behavior state to obtain the compliance risk level of the next time window.

[0009] Furthermore, the behavior state determination model is as follows: ; in, This refers to the state of medication use behavior; Stress-perceiving behavioral characteristics; This is due to a deviation in medication timing; Frequency of administration for prescription drugs; Let be the function for determining medication use behavior.

[0010] Furthermore, the compliance risk prediction model is as follows: ; in, The predicted compliance risk level for the t+1 period; For risk prediction function; This represents the current medication behavior status within the current time window. For historical compliance statistics; This represents the trend of time deviation changes.

[0011] Further, step 4 includes: step 4.1, constructing a set of intervention actions and an intervention strategy selection function to obtain intervention actions; step 4.2, constructing intervention records based on intervention actions and the patient's actual medication behavior after intervention.

[0012] Furthermore, the set of intervention actions is as follows: ; in, The medicine box will vibrate locally as a notification. Provide voice prompts; For mobile message push; Notify the patient's family or healthcare provider; The intervention action is the optimal intervention action: based on medication behavior status, adherence risk level, and user historical behavior characteristics, an intervention strategy selection function is constructed to obtain the optimal intervention action: ; in, This is the optimal intervention action at present; The current behavioral state; Compliance risk level; For user's historical behavior characteristics; For the intervention strategy function; The intervention is a progressive intervention: based on medication time deviation, a progressive intervention control mechanism is used to construct an intervention strategy selection function to obtain the progressive intervention actions. ; in, This is a progressive intervention action; This refers to the timeout period during which medication has not been collected. , and These are the preset first, second, and third trigger time nodes; The intervention record is as follows: ; Wherein, Log represents the intervention record; t represents the type of intervention; t represents the intervention time. This indicates whether the patient has completed taking the medication.

[0013] Furthermore, step 5 includes: step 5.1, processing the patient's intervention records during the assessment period to obtain adherence indicators; step 5.2, judging adherence based on missed doses and medication time deviation trends; step 5.3, generating prescription optimization suggestions based on adherence indicators and adherence; and step 5.4, pushing the prescription optimization suggestions hierarchically to the medical staff, patient's family, and patient's end to form a collaborative feedback mechanism between medical staff, patients, and family.

[0014] This invention also provides an intelligent medication management system based on medication behavior, including a candidate drug acquisition module, a target drug acquisition module, a medication behavior acquisition module, an intervention record module, and a prescription optimization module. The candidate drug acquisition module acquires multimodal medication data and matches it to a drug knowledge base to obtain multimodal feature representations of candidate drugs. The target drug acquisition module performs cross-modal consistency verification on candidate drugs based on the multimodal feature representations to obtain a set of target drugs and target drug features. The medication behavior acquisition module constructs a behavior state determination model and a compliance risk prediction model based on changes in the target drug inventory pressure to obtain medication behavior status and compliance risk level. The intervention record module constructs intervention records based on medication behavior status, compliance risk level, and actual medication behavior. The prescription optimization module processes the intervention records and generates prescription optimization suggestions based on the data processing results.

[0015] The present invention has the following advantages and beneficial effects: This invention eliminates the need for complex mobile phone interface operations, solving the core pain points of existing medication management systems that are unfriendly to users and difficult for elderly users to use, and significantly improving the operability and convenience of the system among the elderly.

[0016] This invention significantly improves the accuracy of drug matching and medication guidance through multi-source information cross-validation and conflict risk labeling. Furthermore, by analyzing the pressure on the medicine box, this invention identifies various medication behaviors, overcoming the limitation of existing technologies that can only determine whether the box has been opened, thus ensuring the authenticity and reliability of adherence analysis results.

[0017] This invention constructs an intervention strategy based on individual behavior. It dynamically selects different levels of intervention intensity based on the patient's historical compliance, current medication behavior, and risk level, and supports automatic adjustment of intervention intensity according to the patient's response. This avoids the problem of insufficient reminders failing to correct missed doses, and solves the patient's aversion and resistance caused by excessive reminders, significantly improving the pertinence and actual effectiveness of the intervention.

[0018] This invention analyzes patients' long-term medication behavior, missed dose patterns, and time deviation trends to provide doctors with prescription optimization suggestions. All suggestions are supported by behavioral data, allowing doctors to directly refer to and adjust treatment plans. This breaks down the data barriers between home medication monitoring and clinical treatment optimization, enabling medication behavior records for adherence management to assist clinical decision-making.

[0019] This invention establishes a multi-party collaborative linkage mechanism, enabling real-time sharing and tiered push of medication behavior. It also supports automatic system updates and recognition after doctors adjust prescriptions, which not only improves the medication safety and long-term adherence of elderly patients, but also reduces the care burden on family members and the follow-up pressure on medical staff, providing a feasible and scalable solution for home management of patients with chronic diseases. Attached Figure Description

[0020] Figure 1 An exemplary flowchart of an intelligent medication management method based on medication behavior provided by the present invention; Figure 2 An exemplary module diagram of an intelligent medication management method based on medication behavior provided by the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] Step 1: Obtain multimodal medication data and match the medication data to the drug knowledge base to obtain multimodal feature representations of candidate drugs.

[0023] Because elderly patients encounter diverse sources of drug information and inconsistent formats during medication management, including drug packaging images, patient voice descriptions, and doctor's prescriptions, and given the differences in their expressive, recognition, and operational abilities, data from a single source is often incomplete or biased. Therefore, a unified multimodal input channel was established to lower the barrier to entry for elderly patients. Heterogeneous data such as images, voice descriptions, and prescriptions were standardized to construct a unified multimodal feature representation for candidate drugs, providing a reliable input foundation for subsequent consistency verification and behavior recognition.

[0024] Step 1.1: The patient obtains multimodal medication data through a terminal device. The terminal device can be used to acquire image data, voice data, and prescription information, including mobile terminals (e.g., smartphones) and smart pillbox terminals. Medication data refers to data related to the patient's medication use, including drug image data, voice description data, and prescription data. Drug image data is obtained by photographing drug packaging or blister packs. Voice description data is obtained by voice input describing the drug name, purpose, or time of administration. Prescription data is obtained by retrieving currently valid prescriptions from the hospital system or historical records.

[0025] The drug use data are analyzed and processed separately to generate a multimodal candidate drug set, and the candidate drug set is then fused to form a multimodal candidate space.

[0026] Step 1.2: Match the drug image data with the drugs in the drug knowledge base to obtain a set of candidate drugs in the images.

[0027] Drug image data is preprocessed to obtain drug image features, which include text features and appearance features. Preprocessing includes denoising, cropping, and enhancement of the image data. Text features include text-related features such as drug name and specifications, which can be obtained through OCR of the drug image data. Appearance features include appearance-related features such as the drug's color, shape, and packaging structure.

[0028] Based on textual and appearance features, drug matching is performed in a drug knowledge base to obtain a set of candidate drugs from the images. ; in, For image-based candidate drug sets; The first one obtained by matching drug image data There are 1 candidate drugs; i is the candidate drug variable in the image; This represents the total number of candidate drugs in the image.

[0029] Step 1.3: Match the voice description data with the drugs in the drug knowledge base to obtain a set of voice candidate drugs.

[0030] Speech recognition and semantic parsing are performed on the speech description data to obtain the semantic features of the drug. These features include the drug name, its intended use, and the time of administration. The drug name includes related names, similar names, and alternative names. The intended use is related to the drug's function, such as lowering blood pressure or blood sugar. The time of administration includes whether the medication is taken in the morning.

[0031] The semantic features of drugs are mapped to a drug knowledge base for drug matching through semantic matching, resulting in a set of candidate drugs for speech. ; in, For voice-based candidate drug sets; Let j be the j-th candidate drug obtained by matching voice description data; j is the voice candidate drug variable; This represents the total number of candidate drugs for speech therapy.

[0032] Step 1.4: Perform structured processing on the patient's prescription data to obtain the set of prescription drugs and drug prescription characteristics: ; ; in, This is a structured collection of prescription data, specifically drug prescription characteristics; k represents the prescription drug variable, where... This refers to a simple set of drug identifiers extracted from prescription data. It is a structured collection containing complete medication attributes. (Only drug identifiers are extracted). For clarity, all subsequent determinations regarding the existence of prescription matching will be based on... ; For the k-th prescription drug label, A collection of prescription drugs; This represents the dosage for the k-th prescription drug. The dosing frequency of the k-th prescription drug; Let K be the time of administration for the k-th prescription drug; K is the total number of drugs in the prescription.

[0033] Step 1.5: Identify candidate drugs and process them to obtain multimodal feature representations of the candidate drugs.

[0034] By merging the image-based candidate drug set, the voice-based candidate drug set, and the structured set of prescription data, a unified candidate drug set is constructed that integrates candidate drugs from these three sources, thus creating a drug candidate space. ; in, This provides a pool of potential drug candidates.

[0035] For each candidate drug in the drug candidate space Constructing multimodal feature representations of candidate drugs: ; in, The candidate drug characteristics are those of candidate drug d. The drug image features of candidate drug d include text features and appearance features; The semantic features of candidate drug d include the results of drug semantic recognition. The prescription features of candidate drug d include prescription matching tags and their attribute information (whether it is in the prescription and the corresponding dosage, etc.). Multimodal feature representation provides a unified, computable, structured expression for each candidate drug.

[0036] Step 2: Based on multimodal feature representation, perform cross-modal consistency verification on candidate drugs to obtain the target drug set and target drug features.

[0037] In the candidate drug set and its multimodal feature representation Based on this, the system addresses the potential information conflicts and uncertainties among different modal data (drug image data, voice description data, and prescription data), enabling mutual verification and error correction when there are deviations in multi-source information. It quantifies the matching confidence of candidate drugs and prescriptions and determines the final target drug set, providing a reliable basis for subsequent behavior recognition and medication plan generation.

[0038] Step 2.1: Calculate the single-modal similarity of the candidate drugs to obtain their multimodal matching degree. The multimodal matching degree includes image matching degree, semantic matching degree, and prescription matching degree. For any candidate drug... : The image features of candidate drugs are matched with the standard drug image features in the drug knowledge base to obtain the image matching degree. Standard drug image features include textual images and visual images of drug standards.

[0039] The semantic features of candidate drugs are matched with their standard drug semantic features in the drug knowledge base to obtain the semantic matching degree. The semantic features of standard drugs include the drug name, drug purpose, and administration time in the drug standard.

[0040] Determining whether a candidate drug exists in the prescription drug set yields the prescription match degree: ; in, denoted as the prescription matching degree of candidate drug d, where 1 represents a match and 0 represents a non-match.

[0041] Step 2.2: Based on the multimodal matching degree, perform a multimodal consistency comprehensive score on the candidate drugs to obtain the consistency score of the candidate drugs.

[0042] Construct a comprehensive consistency scoring function: ; in, The consistency score for candidate drug d; , and These are the image matching weight coefficients, semantic matching weight coefficients, and prescription matching weight coefficients, respectively, satisfying... Prescription matching weight coefficient The highest priority (ensuring medical safety) is given to image matching weight coefficients and semantic matching weight coefficients, which are dynamically adjusted based on data quality. For example, the semantic matching weight coefficient is reduced when the voice confidence is low.

[0043] Step 2.3: Based on multimodal matching degree, conflict detection is performed on candidate drugs to obtain conflict labels; to avoid false matching, conflict detection rules are introduced, and candidate drugs are labeled with high-risk conflicts when one of the following conditions is met: ; ; in, and These are the preset image conflict threshold and semantic conflict threshold, such as 0.7 to 0.8. This formula indicates that if there is a strong match between the image and speech but the content is not in the prescription, there may be a risk of incorrect medication. 1 indicates the presence of a conflict marker, and 0 indicates the absence of a conflict marker.

[0044] Step 2.4: Based on the consistency scores of candidate drugs, obtain the target drug set and target drug features; target drug features include consistency scores, conflict marker values, and drug prescription data. Drug prescription data includes drug dosage, frequency of use, and time of use. Define the target drug set: ; in, A collection of target drugs; The preset consistency score threshold is, for example, 0.6 to 0.8.

[0045] Step 3: Based on the pressure changes in the target drug warehouse, construct a behavioral status determination model and a compliance risk prediction model to obtain the medication behavior status and compliance risk level.

[0046] In the identified target drug set Based on existing methods, relying solely on coarse-grained information such as whether the pillbox has been opened or whether a reminder has been given cannot accurately reflect actual medication use behavior. Therefore, this invention proposes using pressure sensing and time-series data from a smart pillbox to perform fine-grained identification of medication use behavior, categorizing the behavior into discrete states such as "normal medication use, delayed medication use, missed dose, repeated medication use, and accidental activation." While identifying the current state, it also provides interpretable predictions of adherence risk in the next time window, thus providing a quantitative basis for selecting subsequent intervention strategies.

[0047] The system revolves around each target drug Constructing the corresponding medication behavior data stream mainly includes: Step 3.1: Collect the pressure changes of the target drug compartment in the medicine box to obtain the pressure sensing sequence. The system extracts features from the pressure sensing sequence to construct pressure-sensing behavioral characteristics. These characteristics include peak pressure at drug dispensing, duration of dispensing, amplitude of variation, and number of dispensing cycles. The system analyzes the continuous pressure sensing sequence signals acquired by the pressure sensor. Extract key behavioral features: ; in, Stress-perceiving behavioral characteristics; This represents the magnitude of pressure change (difference between maximum value and baseline). Duration of pressure change; The number of pressure changes per unit time (reflecting repeated drug dispensing); Pressure fluctuation stability (reflecting whether the operation is standardized).

[0048] Step 3.2, record the actual medication pickup time. And based on the actual medication collection time and the medication time in the prescription data. Calculate the medication time deviation: ; in, This is due to a deviation in medication timing; This refers to the actual time spent picking up the medication. This refers to the planned medication time in the prescription data.

[0049] Step 3.3: Extract features from stress-perceived behavioral characteristics and medication time deviations to obtain a set of medication behavior states: ; Where S is the set of medication behavior states; To take medication as prescribed; To delay taking medication; For missed doses; To obtain medication repeatedly; The medication was accidentally dispensed.

[0050] Step 3.4: Based on the set of medication behavior states, construct a behavior state determination model to obtain the medication behavior states. The behavior state determination model is as follows: ; in, To determine the medication use behavior status, each medication collection behavior is meticulously categorized. That is, the output result must be to one; The dosing frequency of the k-th prescription drug; Let this be the function for determining medication use behavior. For example, the determination rule is: if... and If within the normal range, it is considered normal medication; if If no valid medication refill event occurs within the dosing cycle, it is considered a delayed dose; if no valid medication refill event occurs within the dosing cycle, it is considered a missed dose; if If so, it is determined to be duplicate medication; if If the value is below the effective threshold and the duration is extremely short, it is determined to be a false touch. The time tolerance threshold (e.g., 30 minutes); This is the threshold for determining repeated medication collection.

[0051] Step 3.5: Based on historical medication behavior, construct an adherence risk prediction model to obtain the adherence risk level for the next time window, thereby predicting the risk of missed doses or abnormalities within future time windows based on historical behavior sequences. The adherence risk prediction model is as follows: ; in, The predicted compliance risk level for the t+1 period; For risk prediction function; This represents the medication behavior status in the current time period t. For historical compliance statistics (such as missed dose rate, delayed dose rate); This represents the trend of time deviation changes.

[0052] For example, engineering implementation can use sliding window statistics (last 7 days / 14 days), state transition frequency estimation, and risk classification rules (low / medium / high risk) to obtain output. This indicates the level of risk associated with medication use within the next time window.

[0053] Step 4: Based on medication behavior status, adherence risk level, and actual medication behavior, construct intervention records. This addresses the problems of fixed reminder strategies, singular intervention methods, and lack of individualized adaptation in existing technologies. For example, for each target drug... The system obtains the current behavior state. Risk level and time deviation In conjunction with the patient's historical behavioral characteristics (such as frequency of missed doses, reminder response time, etc.), the current intervention context is constructed, appropriate intervention strategies are selected from the preset set of intervention actions, and the intervention is executed according to the progressive logic, and the execution results are recorded.

[0054] Step 4.1: Construct a set of intervention actions and, based on medication behavior status, adherence risk level, and user's historical behavioral characteristics, construct an intervention strategy selection function to obtain the optimal intervention action. The optimal intervention method is dynamically selected based on the patient's current behavioral status and risk level. The set of intervention actions is as follows: ; in, The medicine box will vibrate locally as a notification. Provide voice prompts (age-friendly voice prompts); For mobile message push (APP / SMS); Notify the patient's family or healthcare provider.

[0055] The intervention strategy selection function is: ; in, This is the optimal intervention action at present; The current behavioral state; Compliance risk level; For users' historical behavioral characteristics (such as reminder response rate and missed service rate), the historical behavioral characteristics include historical reminder response rate, average response delay time, number of missed services in the past N days, and intervention action preference weight (such as response frequency to vibration / voice / push notifications), etc. This is the intervention strategy function.

[0056] For example, if and If you choose to delay taking medication, then... or ;like Then choose Combination; if If you have missed a dose, then choose .

[0057] In some embodiments, based on medication timing deviations, a progressive intervention control mechanism can be used to upgrade the initial intervention (e.g., the optimal intervention) to obtain progressive intervention actions, thereby reducing interference with patients while ensuring the effectiveness of the intervention. To avoid user aversion caused by a one-time strong intervention, a progressive intervention strategy system from weak to strong is constructed: ; in, This is a progressive intervention action; This refers to the delay time that exceeds the prescribed medication time, i.e., the timeout period when the medication has not been picked up; , and These are the preset first, second, and third trigger time points. The rule for progressive intervention is that if the patient does not respond after the initial reminder, the intensity of the intervention will be escalated step by step.

[0058] In some embodiments, an individualized parameter adaptation mechanism is also included. For example, a time threshold is dynamically adjusted based on the patient's historical performance. Progression interval And the intervention intensity weight. Specifically, for patients who frequently miss doses, the interval between doses is shortened; for highly sensitive patients, the frequency of intervention is reduced, and voice reminders are given priority.

[0059] The system can first determine the initial optimal intervention action based on medication behavior status, adherence risk level, and user's historical behavior characteristics through an intervention strategy selection function. If the patient fails to complete medication within the preset time window after the initial intervention, a progressive intervention control mechanism is activated, automatically escalating the intervention intensity according to the current delay duration until the patient responds or the intervention level is escalated to the highest level. These two mechanisms work together to ensure personalized initial intervention while avoiding intervention gaps due to continued missed doses.

[0060] Step 4.2: Based on the intervention actions and the patient's medication behavior after the intervention, construct a traceable intervention record for subsequent adherence improvement and model optimization: ; Wherein, Log represents the intervention record; t represents the type of intervention; t represents the intervention time. This indicates whether the patient has completed taking the medication (whether they have picked it up). 1 indicates picking up the medication, and 0 indicates not picking up the medication.

[0061] Step 5: Process the intervention records and generate prescription optimization suggestions based on the processing results.

[0062] After intervention implementation, simply focusing on reminders and records is insufficient to continuously improve medication adherence. This invention constructs quantitative adherence assessment indicators based on historical behavior and intervention response data, analyzes patients' long-term medication behavior patterns and risk trends, and can provide interpretable medication optimization suggestions to healthcare professionals, forming a closed-loop optimization mechanism of behavior-assessment-adjustment-re-behavior. This is tailored to each target drug. Within the set evaluation period (e.g., 7 days, 14 days, or 30 days), the system performs the following processing: Obtain the behavioral state sequence of the target drug Intervention records and response results .

[0063] Step 5.1: Process the patient's intervention records during the assessment period to obtain adherence indicators. Adherence indicators include adherence rate, miss rate, and delay rate.

[0064] The on-time medication rate was: ; in, To ensure timely medication adherence; The number of times medication is taken within the allowed time window; This refers to the total number of times medication is planned to be taken.

[0065] The missed service rate was: ; in, This refers to the missed dose rate; This indicates the number of missed doses.

[0066] The latency rate is: ; in, For latency; The number of times medication is delayed.

[0067] Step 5.2: Based on missed doses and the trend of medication time deviation, assess adherence. When there are multiple consecutive missed doses (e.g., 3 consecutive missed doses)... Furthermore, the trend of medication time deviation indicates a gradual increase in delay; and when the intervention is ineffective, patients are marked as high-risk due to decreased adherence. Ineffective intervention is defined as… ; Step 5.3: Generate prescription optimization suggestions based on compliance indicators and compliance status. Construct rule-based decision-making logic based on compliance indicators and behavioral patterns. For example, if... (If the missed dose rate is high), it is recommended to reduce the frequency of medication (e.g., from 3 times a day to once a day); if (If there is a significant delay) it is recommended to adjust the medication time (avoiding periods of low patient activity); if frequent refills are required, it is recommended to strengthen medication education or adjust the pillbox structure; if long-term adherence is poor, it is recommended to switch to a long-acting formulation or simplify the medication regimen. Output Recommendation Set: ; in, For the first The prescription optimization suggestion text includes phrases such as "It is recommended to reduce the frequency of medication" or "It is recommended to adjust the medication time to XX time period".

[0068] Step 5.4 involves pushing prescription optimization suggestions in a tiered manner to the healthcare provider, patient's family member, and patient's end, forming a collaborative feedback mechanism between healthcare providers, patients, and family members. For example, for healthcare providers, adherence indicators and optimization suggestions are pushed; for patient's family member, abnormal behavior reminders (such as consecutive missed doses) are pushed; and for patients, simplified feedback (such as "I have taken my medication on time for 3 consecutive days") is provided. Simultaneously, the system automatically updates when the doctor adjusts the prescription. The system includes reminder plans and behavior judgment baselines to achieve synchronized updates of model parameters and data.

[0069] This invention also provides an intelligent medication management method based on medication behavior, including a candidate drug acquisition module, a target drug acquisition module, a medication behavior acquisition module, an intervention record module, and a prescription optimization module. The candidate drug acquisition module acquires multimodal medication data and matches it to a drug knowledge base to obtain multimodal feature representations of candidate drugs. The target drug acquisition module performs cross-modal consistency verification on candidate drugs based on the multimodal feature representations to obtain a target drug set and target drug features. The medication behavior acquisition module constructs a behavior status determination model and a compliance risk prediction model based on changes in the target drug inventory pressure to obtain medication behavior status and compliance risk level. The intervention record module constructs intervention records based on medication behavior status, compliance risk level, and actual medication behavior. The prescription optimization module processes the intervention records and generates prescription optimization suggestions based on the data processing results.

[0070] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart medication management method based on medication behavior, characterized in that, include: Step 1: Obtain multimodal medication data and match the medication data to the drug knowledge base to obtain multimodal feature representations of candidate drugs; Step 2: Based on multimodal feature representation, perform cross-modal consistency verification on candidate drugs to obtain the target drug set and target drug features; Step 3: Based on the pressure changes in the target drug warehouse, construct a behavioral status determination model and a compliance risk prediction model to obtain the medication behavior status and compliance risk level; Step 4: Construct an intervention record based on medication behavior status, adherence risk level, and actual medication behavior; Step 5: Process the intervention records and generate prescription optimization suggestions based on the data processing results.

2. The intelligent medication management method based on medication behavior according to claim 1, characterized in that, Step 1 includes: Step 1.1: The patient obtains multimodal medication data through a terminal device; the medication data includes drug image data, voice description data, and prescription data. Step 1.2: Match the drug image data with the drugs in the drug knowledge base to obtain a set of candidate drugs in the images; Step 1.3: Match the voice description data with the drugs in the drug knowledge base to obtain a set of voice candidate drugs; Step 1.4: Perform structured processing on the patient's prescription data to obtain a set of candidate prescription drugs; Step 1.5: The image candidate drug set, the voice candidate drug set, and the prescription candidate drug set are identified as candidate drugs, and the candidate drugs are processed to obtain the multimodal feature representation of the candidate drugs.

3. The intelligent medication management method based on medication behavior according to claim 1, characterized in that, Step 2 includes: Step 2.1: Calculate the single-modal similarity of the candidate drugs to obtain the multimodal matching degree of the candidate drugs; the multimodal matching degree includes image matching degree, semantic matching degree, and prescription matching degree; Step 2.2: Based on the multimodal matching degree, perform a comprehensive multimodal consistency score on the candidate drugs to obtain the consistency score of the candidate drugs; Step 2.3: Based on the multimodal matching degree, conflict detection is performed on the candidate drugs to obtain conflict markers; Step 2.4: Based on the consistency scores, conflict markers, and prescription data of candidate drugs, the target drug set and target drug characteristics are obtained.

4. The intelligent medication management method based on medication behavior according to claim 1, characterized in that, Step 3 includes: Step 3.1: Collect the pressure changes of the target drug compartment in the medicine box to obtain the pressure sensing sequence, and extract features from the pressure sensing sequence to construct pressure sensing behavior features; Step 3.2: Record the actual medication pickup time, and calculate the medication time deviation based on the actual medication pickup time and the medication time in the prescription data; Step 3.3: Extract features from stress-perceived behavioral characteristics and medication time deviation to obtain a set of medication behavior states; Step 3.4: Based on the set of medication behavior states, construct a behavior state determination model to obtain the medication behavior state of the current time window; Step 3.5: Based on medication behavior status, construct an adherence risk prediction model to obtain the adherence risk level for the next time window.

5. The intelligent medication management method based on medication behavior according to claim 4, characterized in that, The behavior state determination model is as follows: ； in, This refers to the state of medication use behavior; Stress-perceiving behavioral characteristics; This is due to a deviation in medication timing; Frequency of administration for prescription drugs; Let be the function for determining medication use behavior.

6. The intelligent medication management method based on medication behavior according to claim 4, characterized in that, The compliance risk prediction model is as follows: ； in, The predicted compliance risk level for the t+1 period; For risk prediction function; This represents the current medication behavior status within the current time window. For historical compliance statistics; This represents the trend of time deviation changes.

7. The intelligent medication management method based on medication behavior according to claim 1, characterized in that, Step 4 includes: Step 4.1: Construct the set of intervention actions and the intervention strategy selection function to obtain the intervention actions; Step 4.2: Based on the intervention actions and the patient's actual medication behavior after the intervention, construct the intervention record.

8. The intelligent medication management method based on medication behavior according to claim 7, characterized in that, The set of intervention actions is as follows: ； in, The medicine box will vibrate locally as a notification. Provide voice prompts; For mobile message push; Notify the patient's family or healthcare provider; The intervention action is the optimal intervention action: based on medication behavior status, adherence risk level, and user historical behavior characteristics, an intervention strategy selection function is constructed to obtain the optimal intervention action: ； in, This is the optimal intervention action at present; The current behavioral state; Compliance risk level; For user's historical behavior characteristics; For the intervention strategy function; The intervention is a progressive intervention: based on medication time deviation, a progressive intervention control mechanism is used to construct an intervention strategy selection function to obtain the progressive intervention actions. ； in, This is a progressive intervention action; This refers to the timeout period during which medication has not been collected. , and These are the preset first, second, and third trigger time nodes; The intervention record is as follows: ； Wherein, Log represents the intervention record; t represents the type of intervention; t represents the intervention time. This indicates whether the patient has completed taking the medication.

9. The intelligent medication management method based on medication behavior according to claim 1, characterized in that, Step 5 includes: Step 5.1: Process the patient's intervention records during the assessment period to obtain compliance indicators; Step 5.2: Based on missed doses and the trend of medication time deviation, assess adherence. Step 5.3: Generate prescription optimization suggestions based on compliance indicators and compliance status; Step 5.4: Push prescription optimization suggestions to the medical staff, patient's family, and patient's end in a tiered manner to form a collaborative feedback mechanism between medical staff, patients, and family.

10. An intelligent medication management system based on medication behavior, characterized in that, It includes a candidate drug acquisition module, a target drug acquisition module, a medication behavior acquisition module, an intervention record module, and a prescription optimization module; The candidate drug acquisition module is used to acquire multimodal medication data and match the medication data to the drug knowledge base to obtain the multimodal feature representation of the candidate drugs; The target drug acquisition module is used to perform cross-modal consistency verification on candidate drugs based on multimodal feature representation, and obtain the target drug set and target drug features; The medication behavior acquisition module is used to construct a behavior status determination model and a compliance risk prediction model based on the pressure changes of the target drug warehouse, and obtain the medication behavior status and compliance risk level. The intervention record module is used to construct intervention records based on medication behavior status, adherence risk level, and actual medication behavior; The prescription optimization module is used to process intervention records and generate prescription optimization suggestions based on the data processing results.

Images