Apparatus, system and method for monitoring a behavior of a patient

Abstract

There is provided an apparatus, system and method for monitoring a behavior of a patient in an environment. Behavioral recommendations for the patient are obtained by applying input data comprising clinical conditions of the patient and contextual data stored in data sources related to health and medicine, to a tuned large language model, LLM. The behavioral recommendations are provided to a user through a user interface. Sensor data (SD) are obtained from a plurality of sensors (C1-C5) placed in the environment. A global adherence score (GSC) of the patient to the behavioral recommendations (BRC) is determined based on an analysis of the sensor data. The global adherence score (GSC) is evaluated based on reference thresholds and one or more historic levels of the patient and a result of evaluation (RS) is reported to the user.

Description

[0001] Apparatus, System and Method for monitoring a behavior of a patient

[0002] Field of Invention

[0003] The present invention relates to monitoring a behavior of a patient in an environment.

[0004] Background

[0005] An emerging model of healthcare focuses on prediction and prevention. We cannot think about health only when a patient shows up at the emergency room at the hospital or in the doctor’s office. We need to think about how we can promote and support that person's health and wellbeing upstream, when they're at work and at home.

[0006] Advancing technology, an aging population, and rising health care costs motivate the management of chronic disease management in the community. There is growing evidence that social and behavioral factors play a key role in determining health status among the chronically ill (e.g. hypertension, type 2 diabetes, congestive heart failure, chronic respiratory illness, etc). Empowering family caregivers is important to achieve this.

[0007] Another issue is adherence to medicines. Patients only take their prescribed medicines correctly around 50% of the time. Poor adherence to medicines leads to an estimated $100 Billion spent on avoidable hospitalisations each year in the USA (Cutler and Everett, 2010).

[0008] Self-care is also impacted by mental health status. About one in five Americans experience a mental illness in a given year. Mental health impacts activities of daily living such as sleep patterns and diet. Mental health may also impact adherence to medication.

[0009] Zeidan et al. proposed in a document entitled "Smart Medicine Box System", published in 2018 by IEEE in the proceedings of 2018 IEEE International Multidisciplinary Conference on Engineering Technology (IMCET), Beirut, Lebanon, 2018, pp. 1-5, doi: 10.1109 / IMCET.2018.8603031 , an advanced medicine box monitoring, analysis and control system. The system assists patients in taking their pills treatment on time. The pillbox helps keep the medication out of the reach of children by automatically locking the medical box whenever the patient takes their pills. This system can also be monitored via a phone application. This application calculates the weight of each pill, sets the schedule of medical intake, notifies the user of the number of remaining pills, generating alarms whenever the patient does not take the required number of pills or does not take them at all.

[0010] However, it is not enough for a patient to take their medication correctly to take care of their health. The quality of the living environment, appropriate physical activity and the prevention of risky behavior(s) are all factors that impact a patient's well-being and health at home. There is thus a need for improved monitoring systems to help patients better care for their health and thereby improve their well-being and / or family members care for their ill relatives.

[0011] An objective of the present invention is to overcome the aforementioned challenges, amongst others.

[0012] Summary of Invention

[0013] In a first aspect, there is provided a method for monitoring a behavior of a patient in an environment, comprising the steps of: obtaining behavioral recommendations for the patient by applying input data comprising one or more clinical conditions of the patient and contextual data stored in one or more data sources related to health and medicine, to a tuned large language model, LLM, said behavioral recommendations being provided as output, providing said behavioral recommendations to a user through a user interface, obtaining sensor data from a plurality of sensors placed in the environment, said sensor data comprising electrical data, environmental data and data representative of medication intake by the patient, determining a global adherence score of the patient to the behavioral recommendations based on an analysis of said sensor data, evaluating the global adherence score based on one or more reference thresholds and one or more historic levels of the patient, reporting a result of evaluation to the user.

[0014] The proposed method provides a novel and inventive solution for monitoring a behavior of a patient in a living environment, that combines a prediction of behavioral recommendations specifically adapted to the clinical conditions of the patient, a collection of sensor data representative of a daily activity of the patient in the living environment, an evaluation of a global adherence of the patient to these behavioral recommendations from an analysis of said sensor data and a reporting of the evaluation results to the patient and / or a caregiver.

[0015] The user may be the patient and / or a caregiver, typically a lay person, i.e. someone who has not gone through the training to become a healthcare professional.

[0016] With this method, the patient's behavior is monitored on a regular time basis, for instance daily and feedback is provided to help them follow the activity and health guidelines specifically made for them. This method is thus intended to help the patient better care for their health and thereby improve their well-being.

[0017] According to one or more embodiments, said obtaining of behavioral recommendations comprises: obtaining said one or more clinical conditions of the patient, retrieving documents from said one or more data source, by conducting a semantic search based on at least said one or more clinical conditions of the patient, selecting a subset of the retrieved documents based on an estimated relevance, the one or more clinical conditions being applied along with the subset of retrieved documents to the LLM.

[0018] In this way, the most relevant retrieved documents are provided as contextual data to the LLM, helping to improve the accuracy of the output.

[0019] According to one or more embodiments, the determining of a global adherence score of the patient to the behavioral recommendations comprises: determining a medication adherence score based on the data representative of medical intake by the patient, and determining an activity adherence score based on the electrical and environmental data, and the global adherence score is derived at least from the medication adherence score an activity adherence score.

[0020] Thus, the evaluation takes into account the two criteria that have the greatest impact on improving a patient's well-being and prospects for recovery at home.

[0021] According to one or more embodiments, the determining of a global adherence score further comprises applying the sensor data as input to a machine learning model, said machine learning model having been trained to output a baseline daily living behavior from the sensor data collected over a period of time when the patient is in a stable condition applied as input, and said global adherence score is determined based on said medicine adherence score, said activity adherence score, said baseline daily living behavior and said behavior recommendations for the patient.

[0022] Once the baseline daily living behavior is established, the machine learning model, for instance a trained deep neural network, can continuously compare current data against this baseline daily living behavior to detect anomalies and deviations. This helps in early detection of potential health issues and ensures timely intervention.

[0023] According to one or more embodiments, the evaluating of the global adherence score comprises: comparing (the global adherence score against one or more given reference thresholds and one or more specific historic levels for the patient, detecting an anomaly based on a result of evaluation and, when it has been decided that the anomaly detected is significant, the reporting comprises sending an alert to a caregiver of the patient.

[0024] According to a second aspect, an apparatus for monitoring a behavior of a patient in a physical environment is configured to implement modules of: obtaining behavioral recommendations for the patient by applying input data comprising one or more clinical conditions of the patient and contextual data stored in one or more data sources related to health and medicine, to a tuned large language model, LLM, said behavioral recommendations being provided as output, providing them to a user through a user interface, obtaining sensor data from a plurality of sensors placed in the environment, said sensor data comprising electrical data, environmental data and data representative of medication intake by the patient, determining a global adherence score of the patient to the behavioral recommendations based on an analysis of said sensor data, evaluating the global adherence score based on one or more reference thresholds and one or more historic levels of the patient, and reporting a result of evaluation to the user.

[0025] According to one or more embodiments, the apparatus may comprise means for performing one or more or all steps of the method according to the first aspect. The means may include circuitry configured to perform one or more or all steps of a method according to the first aspect. The means may include at least one processor and at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to perform one or more or all steps of a method according to the first aspect.

[0026] According to a third aspect, a user equipment is configured to receive sensor data from a plurality of sensors arranged in an environment of a patient and said user equipment comprising an apparatus according to the second aspect.

[0027] According to a fourth aspect, a system comprises a user equipment according to the third aspect, a plurality of sensors arranged in the environment of the patient, and a LLM configured to output said behavioral recommendations for the patient.

[0028] According to one or more embodiments, the system further comprises a machine learning model, said machine learning model having been trained to output a baseline daily living behavior from the sensor data applied as input, said baseline daily living behavior being used to determine a global adherence score of the patient to behavioral recommendations output by the LLM.

[0029] According to a fifth aspect, a computer program comprises instructions for causing a computer comprising one or more processors, when said instructions are executed by said one or more processors, to perform the method for monitoring a behavior of a patient according to the first aspect.

[0030] According to a sixth aspect, a non-transitory computer-readable medium storing instructions thereon that, when executed by one or more processors of an apparatus for monitoring a behavior of a patient in an environment, cause the one or more processors to perform the method according the first aspect.

[0031] Brief Description of Drawings Embodiments of the invention are now described, by way of example, with reference to the drawings, in which:

[0032] Figure 1 schematically illustrates an example of a system and apparatus for monitoring a behavior of a patient in an environment;

[0033] Figure 2 is a flow diagram of a method for monitoring a behavior of a patient in an environment;

[0034] Figure 3 a flow diagram illustrating in more detail an exemplary embodiment of the method for monitoring a behavior of a patient in an environment;

[0035] Figure 4 schematically illustrates an example of a structural architecture of an apparatus for monitoring a behavior of a patient in an environment, according to one or more exemplary embodiments.

[0036] Detailed Description

[0037] Detailed example embodiments are disclosed herein. However, specific structural and / or functional details disclosed herein are merely representative for purposes of describing example embodiments and providing a clear understanding of the underlying principles. However, these example embodiments may be practiced without these specific details. These example embodiments may be embodied in many alternate forms, with various modifications, and should not be construed as limited to only the embodiments set forth herein. In addition, the figures and descriptions may have been simplified to illustrate elements and I or aspects that are relevant for a clear understanding of the present invention, while eliminating, for purposes of clarity, many other elements that may be well known in the art or not relevant for the understanding of the invention.

[0038] In the following, a system, an apparatus and a method are proposed to monitor a behavior of a patient in an environment, typically a living environment, such as at home.

[0039] Figure 1 illustrates an example of a system S for monitoring a behavior of a patient PT in an environment ENV, for example at home. The system S comprises a plurality of sensors C1-C5 arranged in the environment ENV to collect a variety of physical data. According to one or more exemplary embodiments, the physical data may comprise electrical data, which includes current, voltage, power data. This electrical data can be monitored by one or more electrical sensors C1 at different sampling rates, e.g., 1 kHz, or 1 MHz and collected from a dedicated device, namely a collector CLR, such as an Energy Management Circuit Breaker commercialized by Eaton.

[0040] The physical data may also comprise environmental data provided by home sensors C2-C4, typically configured to measure humidity, temperature, ventilation, CO2 level, concentration of particulate matter in air, etc. These can be monitored at different intervals e.g., once every second, every minute, or every hour. For example, an air quality sensor C2 can detect tobacco smoke by detecting particles in the air known as PM2.5, i.e. particulate matter of 2.5 micrometers and smaller. This is not specific to tobacco however, other types of smoke and cooking emissions can produce PM2.5, but the impact on health status is similar.

[0041] According to one or more examples, some physical data may also be provided by a connected or “smart” pill box C5. This is a container for prescribed medication that is to be taken at specified intervals. For example, some medicines may be taken once daily (e.g., perhaps the statin class of medication for elevated cholesterol, or antihypertensives such as diuretics (for high blood pressure)), whereas other medication (e.g., nifedipine antihypertensive) may be taken more frequently. The smart pill box C5 is configured to collect data when medicines are removed from the container. Thus, this is a proxy for adherence to treatment regime, i.e., gives some insight into the likelihood the patient PT is taking medication as prescribed. Physical data outputted by the “smart” pill box C5 is timestamp and a name and dosage of pharmaceutical agent.

[0042] In the example of Figure 1 , the system S further comprises a user equipment UE, typically a user terminal, such as a laptop, a smartphone, a tablet or any other user device configured to connect to a telecommunication network and to communicate with distant devices through this telecommunication network and comprising a user interface III. According to one or more examples, the user equipment UE presents a structure of a computer comprising one or more processors operably connected to a non-transitory storage medium MEM. For example, the physical data is received from the plurality of sensors through communication means E / R and stored in memory MEM. In an alternative, the physical data is stored in a cloud storage and accessed by the user equipment UE when needed.

[0043] In the example of Figure 1 , the system S further comprises an apparatus 100 for monitoring a behavior of a patient, for instance the patient PT. In one or more examples the apparatus 100 is configured to obtain behavioral recommendations for the patient PT by applying input data comprising one or more clinical conditions of the patient and contextual data stored in one or more data sources SRC related to health and medicine to a tuned large language model, LLM, said behavioral recommendations being provided as output. The apparatus 100 is further configured to provide said behavioral recommendations to a user through a user interface Ul, obtain physical data from the plurality of sensors C1-C5 placed in the environment ENV, determine a global adherence score of the patient PT to the behavioral recommendations based on an analysis of said sensor data, evaluate the global adherence score based on one or more reference criteria and one or more historic levels of the patient and report a result of evaluation to the user. The user may be the patient PT and / or another user, typically a lay person who takes care of the patient at home. It should be noted that this lay person is someone who has not gone through a training to become a healthcare professional. In one or more examples, the apparatus 100 is configured to implement a method for monitoring a behavior of a patient in an environment, that will be described hereinafter in relation with Figure. 2.

[0044] In one or more examples and as will be described hereinafter in relation to Figure 4, the apparatus 100 comprises one or more processors operably connected to a storage medium storing a computer program comprising instructions that when executed may cause the apparatus to perform the method.

[0045] In the following, clinical condition(s) are meant to refer to a current state of a patient's health as assessed by healthcare professionals, encompassing one or more identified pathologies (diagnosed diseases or disorders), symptoms, signs, medical history, vital signs, laboratory and imaging results, functional status, treatment plans, and prognosis, or multiple clinical conditions.

[0046] In the following, a large language model, LLM, refers to a type of artificial intelligence model that is trained on massive amounts of text data to understand and generate human-like language. These models are typically based on deep learning techniques. Large language models can be used for various natural language processing (NLP) tasks, such as text generation, translation, summarization, question answering, and more. They learn to generate coherent and contextually relevant text based on the patterns and information present in the training data. Language models have been pre-trained on massive datasets and can be fine-tuned for specific applications. They showcase the capabilities of large-scale language models in understanding and generating human-like text across a wide range of tasks. Some examples of LLMs with a known ability to perform a wide range of natural language processing tasks are GPT- 4 (Generative Pre-trained Transformer 3 or 4), developed by OpenAI®, BERT (Bidirectional Encoder Representations from Transformers), developed by Google and T5 (Text-to-Text Transfer Transformer), developed by Google.

[0047] According to the present disclosure, the LLM used is for example GPT-4, as this model is known for its advanced natural language understanding and generation capabilities.

[0048] The LLM is fine tuned to output behavioral recommendations from input data comprising clinical conditions of a patient along with contextual data accessible in health and medicine related data sources..According to one or more examples, this fine-tuning phase comprises the following steps: data collection: a specific, diverse and extensive dataset is gathered that is relevant to the tasks the model will perform (e.g medical data, home activities data, any useful data), preprocessing, including cleaning and organizing the data to ensure it is in a suitable format for training, training of the pre-trained model on the specific dataset. This involves adjusting the model’s parameters to improve its performance on the desired tasks. This is performed iteratively so as to refine the model based on feedback and performance metrics, continuous evaluation of the model’s performance using various metrics to ensure it meets the desired standards.

[0049] In the example of Figure 1 , the LLM is comprised within the user equipment UE. Of course, the proposed solution is not limited to this example of implementation. In an alternative, the LLM may be stored in a distant equipment, for instance a server, and accessed by the apparatus 100 when needed though the communication network CN.

[0050] It is noteworthy that the input data provided to the LLM may or not include a user prompt. A user prompt is the input or instruction provided by the human user to the LLM. It is the initial text or query that the user submits to the system, specifying the desired information or action. The user prompt conveys the user's intent and what they want from the LLM. It allows the user to express a wide range of requests and intents and enables dynamic interaction between the user and the LLM.

[0051] When input data includes a user prompt, the LLM is given specific instructions or context to guide its response. This is often referred to as prompt engineering. For example, if you want the LLM to generate a tutorial for well- being at home, you might provide a prompt like, “Write a short tutorial about a well-being at home when I usually have some headaches." The prompt helps the model LLM understand the context and the type of response expected.

[0052] In scenarii where there is no explicit user prompt, the LLM relies on the context provided by the surrounding text or previous interactions. This is common in tasks like text continuation, where the model generates the next part of a text based on what has already been written. In the present disclosure, without user prompt, the LLM is fine-tuned to automatically analyze the situation of patient at home on a regular basis, for example every week, to check how things go and if there is any progress in their behavior.

[0053] In this regard, fine-tuning is related to both scenarios (with or without a user prompt) as it involves training the LLM on specific datasets to improve its performance on particular tasks. During fine- tuning, the model can be trained with various types of prompts and contexts to enhance its ability to generate relevant and accurate responses. This fine tuning process helps the LLM learn how to handle different types of inputs effectively, whether they include explicit prompts or not.

[0054] In the example of Figure 1 , the data sources SRC are external sources of health and medicine related data that are stored in one or more distant storage media and are accessible through a telecommunication network CN, for instance Internet. The data sources may typically be. peer- reviewed databases such as the National Library of Medicine.

[0055] According to one or more embodiments, the user equipment UE further comprises an Artificial Intelligence machine learning model MLM, that has been trained to output a baseline daily behavior from the sensor data applied as input, said baseline daily behavior being used to determine the global adherence score GSC of the patient to behavioral recommendations BRO output by the LLM.

[0056] Figure. 2 shows a flowchart of a method for monitoring a behavior of a patient in an environment, for example at home. For example, this method is implemented by the apparatus 100 of Figure 1 . While the steps of the method are described in a sequential manner, the man skilled in the art will appreciate that some steps may be omitted, combined, performed in different order and I or in parallel.

[0057] At step S200, clinical conditions CLC of the patient PT are obtained.

[0058] According to one embodiment, a user can enter data representative of the clinical condition(s) of the patient PT into the system S, for instance through the user interface III of the user equipment UE. This data can comprise a clinical condition (e.g. rheumatoid arthritis or congestive heart failure) or multiple clinical conditions. The user can be a family member or other lay person UT who is caring for or monitoring the patient PT or the PT themselves. According to one or more embodiments, the clinical conditions of the patient PT also include a medicine schedule, e.g. the prescribed medication with name and dosage of the pharmaceutical agent and frequency of intake.

[0059] Alternatively, according to another embodiment, the clinical conditions can be automatically entered to the system S, e.g. based on text analysis of clinical notes and summaries received from a healthcare professional. For instance, these clinical notes are accessible via a patient portal to an electronic health record.

[0060] According to one or more embodiments, the clinical note(s) may be compliant to a common format which is known as a SOAP note (SOAP standing for “Subjective, Objective, Assessment, and Plan).

[0061] As an illustrative example, such a clinical note may comprise information related to:

[0062] Patient Information : Name: John XYZ, Date of Birth: 01 / 01 / 1980, Date of Visit: 12 / 07 / 2024, Medical Record Number: 123456

[0063] Subjective (S) :

[0064] • Chief Complaint: “I’ve been experiencing severe headaches for the past two weeks.” • History of Present Illness: The patient reports a gradual onset of headaches, primarily located in the frontal region. The pain is described as throbbing and is rated 7 / 10 in intensity. The headaches are accompanied by nausea and sensitivity to light. No significant relief from over-the-counter pain medications.

[0065] • Past Medical History: Hypertension, Type 2 Diabetes

[0066] • Medications: Metformin, Lisinopril

[0067] • Allergies: None

[0068] • Social History: Non-smoker, occasional alcohol use

[0069] • Family History: Mother has a history of migraines

[0070] Objective :

[0071] • Vital Signs: Blood Pressure: 140 / 90 mmHg, Heart Rate: 80 bpm, Respiratory Rate: 16 breaths / min, Temperature: 98.6°F

[0072] • Physical Examination: General: Alert and oriented, no acute distress, Head: Normocephalic, no tenderness, Eyes: PERRLA (Pupils Equal, Round, Reactive to Light and Accommodation), no papilledema, Neurological: Cranial nerves ll-XII intact, no focal deficits

[0073] Assessment (A) :

[0074] • Primary Diagnosis: Migraine without aura (ICD-10: G43.0),

[0075] • Differential Diagnoses: Tension-type headache (ICD-10: G44.2), Hypertension-related headache (ICD-10: R51.9)

[0076] Plan (P)

[0077] • Diagnostic Tests: MRI of the brain to rule out secondary causes

[0078] • Medications: Prescribe sumatriptan 50 mg as needed for headache

[0079] • Lifestyle Modifications: Advise on regular sleep patterns, hydration, and avoidance of known headache triggers

[0080] • Follow-Up: Re-evaluate in 2 weeks or sooner if symptoms worsen

[0081] Description Elements

[0082] • Patient Information: Basic details about the patient and the visit. • Subjective: Patient’s reported symptoms, medical history, and other relevant personal information.

[0083] • Objective: Clinician’s observations, including vital signs and physical examination findings.

[0084] • Assessment: Clinician’s diagnosis and differential diagnoses.

[0085] • Plan: Proposed treatment plan, including medications, tests, and follow-up instructions.

[0086] Once the clinical notes are obtained, their textual content is analyzed, for instance based on keyword identification, by matching with diagnoses specified in the International Classification of Disease, ICD. The ICD is a globally recognized system developed by the World Health Organization, WHO, for categorizing and coding all known diseases and health conditions. Each clinical condition is assigned a unique code, which includes letters and numbers to indicate its category and specifics The ICD is periodically updated to reflect advances in medical knowledge and healthcare practices. It is used internationally by healthcare providers, researchers, health information managers, and policy makers for epidemiological, clinical, and health management purposes and facilitates effective communication, data collection, and healthcare management.

[0087] At step S210, behavioral recommendations BRC to the patient PT are obtained based on the clinical conditions CLC. This is achieved by applying input data comprising the clinical conditions along with contextual data to a fine-tuned large language model, LLM, configured to output said behavioral recommendations BRC. According to one or more embodiments, the contextual data is sourced from one or more data sources SRC related to health and medicine, for example peer- reviewed databases such as the National Library of Medicine.

[0088] For example, the contextual data is stored in a storage medium, for instance in the local storage medium MEM of the user equipment or in a distant storage medium accessible to the LLM.

[0089] Behavioral recommendations BRC may comprise relevant guidelines for the patient PT regarding activities and self-care of daily living. According to one or more embodiments, each recommendation is linked to its source (e.g., a guidelines document from a given clinic, etc.).

[0090] An illustrative example of behavorial recommendations BRC is given below. It is adapted to Chronic Obstructive Pulmonary Disease (COPD):

[0091] “To effectively manage COPD and improve the quality of life, it’s important to adopt certain lifestyle changes and limit specific activities: avoid smoking and exposure to second-hand smoke to prevent further lung damage, maintain good hygiene by taking regular showers, but avoid excessively hot water which can cause breathing difficulties, eat smaller, frequent meals to avoid bloating and ensure proper nutrition without overexerting your respiratory system, engage in light physical exercises like walking or stretching to keep your muscles strong, but avoid strenuous activities that can lead to breathlessness, stay hydrated and use a humidifier to keep your airways moist. Lastly, ensure your living space is well-ventilated and free from dust and pollutants to create a healthier environment."

[0092] For example, for congestive heart failure, exercise sessions should be of mild intensity and should be brief, multiple short sessions are better than one longer session.

[0093] According to one or more embodiments, the strength of each recommendation is graded. Indeed, ability to exercise diminishes if patients do not reliably take medication to maintain the level of cardiovascular ejection fraction (i.e. , cardiac output). As a result, taking medication regularly as prescribed is a highly graded recommendation.

[0094] At step S220, the behavioral recommendations BRC are presented to the user through the user interface III of the user equipment UE. The presentation may take any form. For instance, an audiovisual presentation may be displayed on a screen of the user equipment UE. It is intended for the patient PT or caregiver UT. It is noteworthy that this step S220 is generally performed beforehand.

[0095] At step S230, sensor data SD are collected from a plurality of sensors C1-C5 arranged in the environment ENV or from one or more intermediate collectors CLR, as illustrated by Figure 1. The sensor data were for instance measured over a time period, for instance the day before. Ideally, the sensor data would be collected continuously and analyzed in real time to monitor the patient’s behavior and well-being without delay.

[0096] Concretely, frequency of monitoring patient activities can vary based on the specific conditions and the patient’s needs. For chronic conditions like COPD or congestive heart failure, daily monitoring is often recommended to ensure timely detection of any deviations from the behavorial recommendations BRC. However, in other cases, hourly monitoring might be necessary, especially if the patient is in a critical phase or recently discharged from the hospital.

[0097] At step S240, the collected sensor data SD are analysed to determine a global adherence score GSC of the patient to the behavioral recommendations BRC.

[0098] At step S250, the global adherence score GSC is evaluated based on one or more reference thresholds and one or more historic levels of the global adherence score stored for the patient.

[0099] According to one embodiment, the determining step S240 comprises sub-steps of determining S242 a medication adherence score MSC based on the data representative of medical intake by the patient, and determining S243 an activity adherence score ASC based on the electrical and environmental data, and the global adherence score is derived at least from the medication adherence score MSC an activity adherence score ASC.

[0100] For example, the GSC is derived from the MSC and ASC by using a simple average: GSC = MSC + ASC

[0101] 2

[0102] According to another example, a weighted average GSC =W1MSC+W2-ASCwjth i , W2 weights ranging between 0 and 1 to reflect that one score among the MSC and the ASC is considered more important than the other.

[0103] According to one or more embodiments, the GSC is evaluated based on the following reference thresholds: a high adherence corresponds to GSC > 0.8, a moderate adherence corresponds to : 0.5 < GSC 0.8 and a low adherence corresponds to GSC s= 0.5.

[0104] Historic levels refer to the baseline data collected over a period of time and stored in memory, representing the patient’s typical activity patterns and adherence to recommendations. This data is used to establish a moving average, which smooths out short-term fluctuations and highlights longer-term trends. By comparing current activity levels against these historic baselines, significant deviations can be identified that might indicate a problem.

[0105] For example, if a patient with COPD typically walks 5,000 steps per day but suddenly drops to 2,000 steps per day, detection of such a deviation from the historic level could trigger an alert for further investigation.

[0106] Reference thresholds are set to determine acceptable ranges of medical adherence score, activity adherence score and global adherence score . According to one or more embodiments, these reference thresholds can be based on a statistical analysis of the historic baseline data.

[0107] Some illustrative examples of reference thresholds are given below:

[0108] For a Medication Adherence score MSC (in %), the following reference thresholds may be considered: High Adherence: 90-100%, Moderate Adherence: 70-89%, Low Adherence: Below 70%

[0109] For an activity adherence score ASC related to physical activity:

[0110] Related to a number of steps per day: Normal Range: 4,500 - 5,500 steps, Alert Range: Below 4,000 steps or above 6,000 steps,

[0111] Related to a sleep duration: Normal Range: 7-9 hours per night, Alert Range: Below 6 hours or above 10 hours As an illustrative example, a patient with congestive heart failure has a historic average of taking their medication 95% of the time and walking 4,800 steps per day. If their medication adherence drops to 60% and their daily steps fall to 2,500, these deviations would exceed the established reference thresholds. An alert would be sent to the caregiver, prompting an investigation into potential causes, such as side effects from medication or worsening symptoms.

[0112] By continuously monitoring and analyzing these patterns, caregivers can intervene early and adjust treatment plans to better support the patient’s health.

[0113] At step S260, a result of evaluation RS is reported to the user. For example, reporting may comprise displaying audio and / or visual presentation displayed on the screen of the user equipment UE. According to one or more examples, the report RS includes at least a graphical representation of the Global Adherence Score GSC along with detailed insights and recommendations that are displayed on the screen of the user equipment or read to the patient. The report RS may also include a graphical representation of the medication adherence score MSG and the activity adherence score ASC. For instance, the ASC may be represented as a bar chart or pie chart showing the breakdown of steps, exercise, sleep, hydration, and diet adherence. Finally, the report RS may also comprise a brief summary of the evaluation results, highlighting key findings and any significant deviations from the baseline.

[0114] Some of the above steps will now be described in more detail in relation to Figure 3. Once the clinical conditions are obtained in S200, the step S210 of determining behavioral recommendations BRC to the patient PT is executed.

[0115] According to the exemplary embodiment of Figure 3; step S210 comprises the following substeps:

[0116] In a sub-step S211 , document retrieval from the data sources SRC is implemented, based on the clinical conditions CLC, by conducting a semantic search into the data sources SRC. According to one or more embodiments, a hybrid approach is conducted involving i) keyword search, ii) embeddings and use of a semantic similarity score.

[0117] Keyword search is a known technique involving a search engine configured to look for exact matches of the query terms within the documents.

[0118] Embeddings are numerical representations of text that capture the semantic meaning of words, sentences, or documents. By converting text into high-dimensional vectors, embeddings allow for more nuanced comparisons between texts. For example, the words “doctor” and “physician” would have similar embeddings, even though they are different words. The document by Neelakantan et al., entitled “Text and Code Embeddings by Contrastive Pre-Training”, published by ARXiv arXiv:2201.10005v1 [cs.CL] 24 Jan 2022, explains how embeddings are used for tasks like semantic search and provides examples of how to implement them.

[0119] A Semantic Similarity Score is designed to measure how similar two pieces of text are in meaning, rather than just in wording. This is often calculated using techniques like cosine similarity, which compares an angle between two vectors in a high-dimensional space. A higher similarity score indicates that the texts are more semantically alike.

[0120] For instance, the semantic search, including the use of semantic similarity scores, is conducted as described in the document by Steen et al, entitled “Semantic ranking in Azure Al Search”, published on12 / 06 / 2024 by Microsoft and available at the URL: https: / / learn.microsoft.com / en- us / azure / search / semantic-search-overview.

[0121] According to one or more embodiments, retrieved documents are ordered based on relevance to the clinical conditions CLC based on term and concept frequency. A subset of the retrieved documents is determined based on estimated relevance, for example as a ranked list of the most relevant documents. For example, relevance of the retrieved documents can be calculated based on a strength of correlation between semantic concepts related to the clinical condition and also to lifestyle / behavioral changes.

[0122] Semantic concepts related to clinical conditions refer to medical terms, symptoms, treatments, and other relevant information associated with a particular health issue. For example, in the context of Alzheimer’s disease, semantic concepts might include terms like “cognitive decline,” “memory loss,” and “beta-amyloid plaques.”

[0123] In the present disclosure, lifestyle and behavioral changes refer to modifications in daily habits and behaviors that can impact health. These might include diet, exercise, sleep patterns, etc.

[0124] For instance, regular physical activity and a healthy diet have been shown to influence cognitive health and can be crucial in managing conditions like Alzheimer’s disease, as explained in document by Ornish et al., entitled “Effects of intensive lifestyle changes on the progression of mild cognitive impairment or early dementia due to Alzheimer’s disease: a randomized, controlled clinical trial”, published by BMC in 2024, at URL https: / / doi.org / 10.1186 / s13195-024-01482-z.

[0125] According to one or more embodiments, additional documents can be integrated to the contextual data at sub-step S213. They may consist of medical guidance or wellness leaflets that are provided to the patient by clinicians. They can be scanned by a scanner application of the user equipment UE and stored into the local storage medium MEM. Then, they are applied optical character recognition, keyword recognition, and then are incorporated into the contextual data to be passed through the LLM. In a sub-step S212, the ranked list of most relevant documents is applied as contextual data with the clinical conditions CLC to the LLM, that has been fine tuned to output adapted behavioral recommendations BRC to the patient PT.

[0126] Several architectures may be considered for the LLM. According to one embodiment, an encoderdecoder architecture takes textual information as input data, typically comprising the ranked list of most relevant documents along with the clinical conditions CLC of the patient PT. The input data is converted (by an encoder) into vectors that are used to directly produce (by a decoder) a textual output, typically the behavioral recommendations BRC to the patient PT.

[0127] According to another embodiment, an encoder-only architecture is implemented. In this case, the ranked list of most relevant documents is provided to the LLM as a subset of vectors. In other words, decoding is performed upstream, at sub-step S212. Such an encoder-only architecture is that it is very quick and provides behavioral recommendations in real-time.

[0128] According to this embodiment, sub-step S212 further comprises transforming the text of ranked list of most relevant documents into a global vector that will be fed to the LLM.

[0129] The textual data comprised within the ranked list of most relevant documents and possibly the additional documents undergo several stages of transformation. First, the raw text is chunked into smaller units called segments. These can be words, sub-words, or even characters, depending on the tokenizer used. Second, the segments are embedded and positional encoded. Finally, the token embeddings and positional encodings are summed to form the final input vectors that are fed into the encoder. These combined vectors (token embeddings + positional encodings) are the input to the encoder layers, where they are processed in parallel to capture the contextual relationships between tokens.

[0130] Considering the additional documents of sub-step S213, in case of an encoder-only architecture, they may undergo the same transformation process as described above for the ranked list of most relevant documents.

[0131] According to the embodiment of Figure 3, step S230 of sensor data collection is divided into the following sub-steps:

[0132] Electrical data is collected from the smart meter sensors (C1 in Figure 1) in sub-step S231. Environmental data is collected from the home sensors (C2-C4 in Figure 1) in sub-step S232. Data related to medicine intake are collected from the “smart” pillbox (C5 in Figure 1). All the sensor data are stored in a storage medium in sub-step S234, for instance memory MEM of the user equipment UE. Step S240 of determining a global adherence score GSC will now be detailed. In a sub-step S241 , a baseline daily living behavior of the patient PT is learnt from the collected data stored in the storage medium.

[0133] In the present disclosure, baseline daily living behavior refers to the typical patterns and routines of a patient’s daily activities, such as physical activity, sleep, medication adherence, and other health-related behaviors. This baseline is established by collecting data over a period of time when the patient is in a stable condition.

[0134] For example, the baseline daily living behavior data is used in the monitoring process for:

[0135] Establishing normal ranges:the baseline data helps define what is “normal” for the patient. For example, if a patient typically walks 5,000 steps per day, this becomes their normal range for physical activity.

[0136] Detecting deviations: by comparing current activity levels to the baseline, we can identify significant deviations. For instance, if the patient’s daily steps suddenly drop to 2,000, this deviation from the baseline indicates a potential issue.

[0137] Setting thresholds: thresholds are set based on the baseline daily living behavior data to determine acceptable ranges of behavior. If the patient’s activity falls outside these thresholds, it triggers an alert. For example, if the baseline for medication adherence is 95%, a drop below 70% might trigger an alert.

[0138] Generating alerts: when deviations from the baseline exceed the set thresholds, alerts are generated to notify caregivers. This allows for timely intervention to address potential health issues.

[0139] Correlation analysis:the baseline data is also used to analyze correlations between different behaviors. For example, if poor medication adherence is correlated with reduced physical activity, this insight can help identify root causes of health issues.

[0140] The time period used to establish the baseline daily living behavior of a patient can vary depending on the specific application and the patient’s condition. However, for a robust and accurate baseline daily living behavior, data is typically collected over a longer period rather than just a single day. Here are some common time periods used for baseline daily living behavior data collection: weekly data: collecting data over a week helps capture variations in daily routines and provides a more comprehensive view of the patient’s typical behavior. monthly data:A month-long period can account for longer-term patterns and fluctuations, such as changes in activity levels due to weekends, work schedules, or monthly health cycles. multi-month data: for chronic disease conditions, collecting data over several months can help identify seasonal variations and more stable long-term patterns.

[0141] As an illustrative example, for a patient with COPD, the deep neural network might be trained using sensor data collected over the past three months. This period would include daily activity levels, medication adherence, sleep patterns, and other relevant behaviors. By using a longer time frame, the model can learn the patient’s baseline daily living behavior more accurately and account for any natural variations.

[0142] Longer periods for collecting data are preferred, because they provide more data points, leading to a more accurate and reliable baseline. They also allow to account for variability of the data by capturing different scenarios and variations in the patient’s routine which helps the model generalize better. Moreover, longer periods help identify trends and patterns that might not be apparent in shorter time frames.

[0143] Once the baseline daily living behavior is established, the model can continuously compare current data against this baseline daily living behavior to detect anomalies and deviations. This helps in early detection of potential health issues and ensures timely intervention.

[0144] As an illustrative example, it is assumed that a patient with congestive heart failure has a baselin daily living behavior of: Medication Adherence Score: 95% and Daily Steps: 5,000

[0145] If the patient’s current data shows a Medication Adherence Score of 60% and a number of Daily Steps of 2,500, these deviations from the baseline would trigger an alert. The caregiver can then investigate potential causes, such as side effects from medication or worsening symptoms, and take appropriate action.

[0146] According to an embodiment, the collected sensor data are applied to a trained machine learning model MLM, typically a deep neural network with multiple layers, that produces the baseline daily living behavior of the patient PT.

[0147] This baseline daily living behavior output by the trained machine learning model MLM typically includes detailed information about the patient’s regular activities and health metrics.

[0148] As an illustrative example of a structure of this output, the information contained in the baseline daily living behavior may comprise information about: physical activity comprising steps taken: Daily step count, exercise duration: Time spent on physical activities like walking, running, or specific exercises, activity intensity: levels of activity intensity (e.g., light, moderate, vigorous), vital signs comprising heart rate: average, minimum, and maximum heart rate, blood pressure: systolic and diastolic readings, respiratory rate: breaths per minute, sleep patterns comprising sleep duration: total hours of sleep per night, sleep quality: metrics like deep sleep, REM (Rapid Eye Movement) sleep, and sleep interruptions,. medication adherence comprising dosage compliance: percentage of prescribed doses taken, timing compliance: adherence to prescribed times for medication intake, diet and nutrition: meal frequency: number of meals and snacks per day, caloric intake: estimated daily calorie consumption, nutrient balance: Intake of key nutrients like proteins, fats, and carbohydrates, environmental factors comprising air quality: levels of pollutants or allergens in the living environment, temperature and humidity: indoor climate conditions.

[0149] The output from the deep neural network MLM is typically structured in a way that allows for easy comparison and analysis.

[0150] For example, the deep neural network MLM has been trained with an unsupervised approach.

[0151] Unsupervised learning involves training a neural network on unlabeled data, allowing the model to discover patterns and structures within the data without explicit guidance. The unlabeled data used for training does not have predefined labels. For example, sensor data from a patient’s daily activities are a number of steps taken, heart rate, and sleep patterns. Typically, data is collected over an extended period (e.g., weeks or months) to capture a comprehensive baseline of the patient’s behavior.

[0152] As an illustrative example, the structure of the deep neural network may include: an Input Layer that receives the raw sensor data (e.g., steps, heart rate). multiple hidden layers that transform the input data into higher-level features. These layers can include one or more autoencoders that compress the data into a lower-dimensional representation and then reconstruct it, convolutional Layers that extract spatial features from the data, recurrent layers that capture temporal dependencies in sequential data, fully- cinnected layers that combine features from previous layers to orm a comprehensive representation, and an output Layer that Produces the baseline daily living behavior, which can be a representation of typical activity patterns.

[0153] Typically, an unsupervised Training Process comprises the following steps of: randomly initializing the weights and biases of the deep neural network,. applying a forward Pass by passing the input data through the network to generate an output, compute a loss, which measures the difference between the input data and the reconstructed data (in the case of autoencoders) or between the predicted output and a specific target output (in case the model is not an autoencoder and there is no reconstructed data), backpropagating the loss so as to adjust the weights and biases to minimize the loss using for instance algorithms like Stochastic Gradient Descent (SGD) or Adam. iterating the forward pass and backpropagation steps for multiple epochs until the model converges.

[0154] In the case of an autoencoder, the encoder transforms the input data ( x ) into a lower-dimensional representation ( z ), the decoder reconstructs the input data from ( z ) to produce ({%}), the loss function used and to be minimized is for instance a Mean Squared Error (MSE) between ( x ) and ( {*})■

[0155] This baseline daily living behavior output from the neural network is typically structured in a way that allows for easy comparison and analysis, and may be applied to: anomaly Detection performed by comparing current data against the baseline to detect deviations.

[0156] Alert Generation comprising triggering alerts if deviations exceed predefined thresholds.

[0157] Correlation Analysis which is achieved between medication adherence and activity patterns to identify potential causes of deviations.

[0158] As an illustrative example, for a patient with COPD, the deep neural network MLM is trained using sensor data collected over the past three months. The network learns the typical patterns of daily activities, such as a number of steps taken, heart rate, and sleep duration and provides a baseline daily behavior for the patient which is then exploited by the system according to the present disclosure to continuously monitor and compare current data to this baseline, so as to detect anomalies and generate alerts for caregivers.

[0159] Using a deep neural network with multiple layers allows to extract and explore the hidden structures and relationships within the collected sensor data.

[0160] In a sub-step S242, a medication adherence score MSG is determined from the data collected by the smart pill box and based on the part of the behavioral recommendations related to medicine schedule. Taking the medicines on time is essential and can affect the health of a patient. In this sub-step the timestamps and portions of pharmaceutical agents removed from the smart pill box are compared with the medicine schedule provided in the behavioral recommendations BRC and used to generate the medication adherence score MSG. For example, the MSG can range from 0 to 1. A MSC above a given threshold can indicate a positive response (adherent) and a lower score can indicate the need for support and conversations from a caregiver.

[0161] An illustrative example of how the Medication Adherence Score (MSC) can be calculated is given below.

[0162] It is assumed the following medicine schedule: Morning Dose: 8:00 AM, Afternoon Dose: 2:00 PM, Evening Dose: 8:00 PM. Data Collected by the Smart Pill Box are: Morning Dose Taken: 8:05 AM, Afternoon Dose Taken: 2:10 PM, Evening Dose Taken: 8:00 PM

[0163] Calculation of MSC:

[0164] Definition of an Adherence Criteria: for example, a dose is considered adherent if taken within a 15-minute window before or after the scheduled time.

[0165] Evaluation of each Dose:

[0166] • Morning Dose: scheduled: 8:00 AM, taken: 8:05 AM, adherent: Yes (within the 15-minute window)

[0167] • Afternoon Dose: scheduled: 2:00 PM, taken: 2:10 PM, adherent: Yes (within the 15- minute window)

[0168] • Evening Dose: scheduled: 8:00 PM, taken: 8:00 PM, adherent: Yes (exactly on time) Calculation of the MSC: Total Doses: 3, Adherent Doses: 3, MSC = (Adherent Doses I Total Doses), MSC = 3 / 3 = 1.0

[0169] In this example, the calculated MSC is 1.0, indicating perfect adherence.

[0170] According to another example the patient misses a dose:

[0171] Evaluation of each dose:

[0172] • Morning Dose: scheduled: 8:00 AM, taken: 8:05 AM, adherent: Yes (within the 15-minute window)

[0173] • Afternoon Dose: scheduled: 2:00 PM, taken: 2:30 PM, adherent: No (outside the 15- minute window)

[0174] • Evening Dose: scheduled: 8:00 PM, not taken: 8:00 PM, adherent: No

[0175] • Calculation of the MSC:Total Doses: 3, Adherent Doses: 1 , MSC = (Adherent Doses I Total Doses), MSC = 1 / 3 « 0.33

[0176] For example, the reference thresholds for MSC are High Adherence: MSC > 0.8, Moderate Adherence: 0.5 < MSC s= 0.8 and Low Adherence: MSC s= 0.5. In this latter case, the MSC is approximately 0.33, indicating poor adherence and potentially triggering an alert for caregiver intervention. These reference thresholds help caregivers determine when to intervene and provide additional support to the patient. In a sub-step S243, an activity adherence score ASC is determined from the sensor data collected by the other sensors and by comparison with the recommended daily and healthy activities listed in the behavioral recommendations BRC.

[0177] An illustrative example of how the Activity Adherence Score (ASC) can be calculated based on sensor data and behavioral recommendations (BRC). is given below. extract a List of daily and healthy activities from the BRC. For example, the BRC defines: Steps: 10,000 steps per day, Exercise: 30 minutes of moderate exercise, Sleep: 7-9 hours per night, Hydration: 8 glasses of water, Diet: 3 balanced meals collect Sensor Data based on various sensors comprising for example:Wearable Devices: Track steps, exercise duration, and heart rate, Smart Home Devices: Monitor sleep patterns and hydration levels, Diet Tracking Apps: Record meal frequency and nutritional intake. compare Collected Data with recommended values of the list of daily and healthy activities: for each activity, calculate an individual adherence score for each activity from the collected data and the recommended values:

[0178] • seps Adherence for a Recommended value of: 10,000 steps and an Actual value of: 8,000 steps is SAS = 8000 / 10000 = 0.8

[0179] • exercise time Adherence for a Recommended value of 30 minutes and an actual value of 20 minutes, is TAS = 20 / 30 — 0.67

[0180] • sleep Adherence score for a recommended value between 7-9 hours and an actual value of 6 hours, is SAC=6 / 7 — 0.86 (if using the lower bound of the recommended range)

[0181] • hydration Adherence score for a Recommended value of 8 glasses and an actual value of 6 glasses, is HAS = 6 / 8=0.75

[0182] • diet Adherence score for a recommended value of 3 balanced meals and an actual value of 3 balanced meals is DAS = 3 / 3 = 1

[0183] Aggregate the above individual adherence cores to calculate the overall ASC.

[0184] For example: ASC = (SAC+TAC+SAC+HAC+DAC) / 5

[0185] According one or more embodiments, a weighted average is performed to enhance some activities that are more important than others.

[0186] Considering the example of the previous patient with the above individual adherence scores, the overall Activity adherence score ASC would be ASC=(0.8 + 0.67 + 0.86 + 0.75 + 1 ,0) / 5 = 0.816 ]

[0187] For example, the reference thresholds for the activity adherence score ASC may be: High Adherence: ASC > 0.8, Moderate Adherence: 0.5 < ASC < 0.8, Low Adherence: ASC < 0.5. In this case, the ASC of the patient is high.

[0188] Based on the clinical conditions of patient, the recommended daily and healthy activities can be different than before medical attention. This activity adherence scoring is meant to help patient to follow their healthy activity and meanwhile provide high level well-being. For example, adhering to healthy lifestyle factors, including avoiding tobacco use (using PM2.5 sensor), maintaining a healthy diet, engaging in sufficient physical activity, etc., is a cost-effective strategy for preventing noncommunicable diseases suggested by the World Health Organization, for example in the document entitled “Tackling NCDs: “best buys” and other recommended interventions for the prevention and control of noncommunicable diseases, published in 2017 at website https: / / apps.who.int / iris / handle / 10665 / 259232.

[0189] At sub-step S244, a global adherence score GSC is derived from the medicine adherence score and activity adherence score. The GSC is derived by combining the MSC and ASC. This can be done using a simple average or a weighted average if one score is considered more important than the other.

[0190] Considering the previous example, the patient would get a global adherence score GSC of:

[0191] GSC = (1 +0.816) / 2=0.908

[0192] For example, GSC is interpreted with the following reference thresholds of High Adherence: GSC > 0.8, Moderate Adherence: 0.5 < GSC < 0.8 andLow Adherence: GSC < 0.5. The GSC of 0.908 indicates high adherence, suggesting that the patient is following both their medication schedule and activity recommendations well.

[0193] At S250, the global adherence score GSC is evaluated for instance based on one of the formula exposed above. In a sub step S251 , a comparison of the GSC is made with one or more reference global adherence thresholds. Historic levels of the global adherence score GSC of the patient PT may also be used for comparison in order to evaluate an evolution of the global adherence score over time.

[0194] A sub-step S252, it is decided based on the crossed thresholds and the evolution determined at previous sub step if a significant anomaly is detected.

[0195] At step S260, a report RS is made to the user of the user equipment, typically the patient PT and / or the caregiver through the user interface. For example, a graphical representation of at least the global adherence score is. According to one or more examples, the report includes at least a graphical representation of the Global Adherence Score GSC along with detailed insights and recommendations that is displayed on the screen of the user equipment. It may also include a graphical representation of the medication adherence score and the activity adherence score. For instance, the ASC may be represented as a bar chart or pie chart showing the breakdown of steps, exercise, sleep, hydration, and diet adherence.

[0196] For instance the recommendations accompanying the graphical representations of the global adherence score, may comprise textual or audio comments with encouragement to continue taking medications on time, to keep hydrated (“Increase your water intake to 8 glasses per day, keep a water bottle with you to remind yourself to drink regularly.”), incentive to increase physical activity (“Try to increase your daily steps to reach the 10,000 goal », « Consider adding an extra 10 minutes to your exercise routine »), to have a good night’s sleep (“aim for at least 7 hours of sleep per night, establish a consistent bedtime routine to improve sleep quality.”) and so on.

[0197] When a significant anomaly has been decided at S252, an alert is sent at S261 to the user UT, for example the caregiver. For example, such a decision is made based on predefined rules.

[0198] For example, these predefined rules for detecting an anomaly comprise: regarding the number of steps per day: “If steps fall below 8,000 or exceed 12,000 for more than 2 consecutive days, trigger an alert.” regarding medication adherence:“lf adherence falls below 80% on any day, trigger an alert.” regarding sleep duration:“lf sleep duration falls outside the 6-10 hour range for more than 2 consecutive days, trigger an alert.”

[0199] Based on the rules of decision, the following alerts would be generated: steps per day alert: Message: “Alert: Significant decrease in daily steps detected. Steps have fallen below 8,000 for 3 days this week. Please check on the patient’s activity levels.” medication adherence alert: Message: “Alert: Medication adherence below 80% detected on Friday and Sunday. Please ensure the patient is taking their medications as prescribed. ” sleep Duration Alert: Message: “Alert: Sleep duration below 6 hours detected on Wednesday and Friday. Please check on the patient’s sleep patterns and consider possible interventions."

[0200] Figure 4 depicts a high-level block diagram of an apparatus 100 suitable for implementing various aspects of the disclosure. Although illustrated in a single block, in other embodiments the apparatus 100 may also be implemented using parallel and distributed architectures. Thus, for example, various steps such as those illustrated in the methods described above by reference to Figures 2 and 3 may be executed using apparatus 100 sequentially, in parallel, or in a different order based on particular implementations.

[0201] According to an exemplary embodiment, depicted in Figure 4, apparatus 100 comprises a printed circuit board 101 on which a communication bus 102 connects a processor 103 (e.g., a central processing unit "CPU"), a random access memory 104, a storage medium 111 , an interface 105 for connecting a display 106, a series of connectors 107 for connecting one or more user interface devices or modules 108, 109, a wireless network interface 110 and / or a wired network interface 112. Depending on the functionality required, the apparatus may implement only part of the above. Certain modules of Figure 4 may be internal or connected externally, in which case they do not necessarily form integral part of the apparatus itself. Memory 111 contains software code which, when executed by processor 103, causes the apparatus to perform the method described herein. In an exemplary embodiment, a detachable storage medium 113 such as a USB stick may also be connected. For example, the detachable storage medium 113 can hold the software code to be uploaded to memory 111.

[0202] The processor 103 may be any type of processor such as a general purpose central processing unit ("CPU") or a dedicated microprocessor such as an embedded microcontroller or a digital signal processor ("DSP").

[0203] In addition, apparatus 100 may also include other components typically found in computing systems, such as an operating system, queue managers, device drivers, or one or more network protocols that are stored in memory 111 and executed by the processor 103.

[0204] Although aspects herein have been described with reference to particular embodiments, it is to be understood that these embodiments are merely illustrative of the principles and applications of the present disclosure. It is therefore to be understood that numerous modifications can be made to the illustrative embodiments and that other arrangements can be devised without departing from the spirit and scope of the disclosure as determined based upon the claims and any equivalents thereof.

[0205] For example, the data disclosed herein may be stored in various types of data structures which may be accessed and manipulated by a programmable processor (e.g., CPU or FPGA) that is implemented using software, hardware, or combination thereof.

[0206] It should be appreciated by those skilled in the art that any block diagrams herein represent conceptual views of illustrative circuitry embodying the principles of the disclosure. Similarly, it will be appreciated that any flow charts, flow diagrams, state transition diagrams, and the like represent various processes which may be substantially implemented by circuitry.

[0207] Each described function, engine, block, step can be implemented in hardware, software, firmware, middleware, microcode, or any suitable combination thereof. If implemented in software, the functions, engines, blocks of the block diagrams and / or flowchart illustrations can be implemented by computer program instructions I software code, which may be stored or transmitted over a computer-readable medium, or loaded onto a general purpose computer, special purpose computer or other programmable processing apparatus and / or system to produce a machine, such that the computer program instructions or software code which execute on the computer or other programmable processing apparatus, create the means for implementing the functions described herein.

[0208] When provided by a processor, the functions may be provided by a single dedicated processor, by a single shared processor, or by a plurality of individual processors, some of which may be shared. Moreover, explicit use of the term "processor" or "controller" should not be construed to refer exclusively to hardware capable of executing software, and may implicitly include, without limitation, digital signal processor (DSP) hardware, network processor, application specific integrated circuit (ASIC), field programmable gate array (FPGA), read only memory (ROM) for storing software, random access memory (RAM), and non-volatile storage. Other hardware, conventional or custom, may also be included. Their function may be carried out through the operation of program logic, through dedicated logic, through the interaction of program control and dedicated logic, or even manually, the particular technique being selectable by the implementer as more specifically understood from the context.

[0209] In the present disclosure, the combined data from various sensors comprising smart meters and smart pill box are exploited to generate valuable insights over time. By continuously monitoring and comparing current data to a baseline, analyzing patterns and correlations between medication adherences, health measurements, health outcomes / results, daily activities patterns using meter / other data, the proposed solution contributes to help patients to follow their recommended activities and thereby improve their well-being and / or family members better care for their ill relatives.. It also allows to better understand how specific medications affect patients’ well-being. Indeed, the proposed method and system may contribute to provide valuable insights into how specific medications affect a patient’s well-being by continuously monitoring various health metrics and comparing them to the patient’s baseline behavior as I described above. For instance, if a patient with congestive heart failure shows improved physical activity levels and reduced symptoms when adhering to their medication schedule, this positive correlation can be identified.

Claims

Claims1. A method for monitoring a behavior of a patient (PT) in an environment (ENV), comprising the steps of: obtaining (S210) behavioral recommendations (BRC) for the patient by applying input data comprising one or more clinical conditions (CLC) of the patient and contextual data stored in one or more data sources (SRC) related to health and medicine, to a tuned large language model, LLM, said behavioral recommendations being provided as output, providing (S220) said behavioral recommendations to a user (PT, UT) through a user interface (III), obtaining (S230) sensor data (SD) from a plurality of sensors (C1-C5) placed in the environment (ENV), said sensor data comprising electrical data, environmental data and data representative of medication intake by the patient, determining (S240) a global adherence score (GSC) of the patient to the behavioral recommendations (BRC) based on an analysis of said sensor data, evaluating (S250) the global adherence score (GSC) based on one or more reference thresholds and one or more historic levels of the patient, reporting (S260) a result of evaluation (RS) to the user (PT, UT).

2. A method according to claim 1 , characterized in that said obtaining of behavioral recommendations (BRC) comprises: obtaining (S200) said one or more clinical conditions (CLC) of the patient, retrieving documents from said one or more data sources (SRC), by conducting a semantic search based on at least said one or more clinical conditions (CLC) of the patient, selecting a subset of the retrieved documents based on an estimated relevance, the one or more clinical conditions being applied (S212) along with the subset of retrieved documents to the LLM.

3. A method according to any one of the preceding claims, characterized in that the determining (S240) of a global adherence score (GSC) of the patient to the behavioral recommendations (BRC) comprises: determining (S242) a medication adherence score (MSC) based on the data representative of medical intake by the patient, and in that determining (S243) an activity adherence score (ASC) based on the electrical and environmental data, andin that the global adherence score is derived at least from the medication adherence score (MSC) an activity adherence score (ASC).

4. A method according to the preceding claim, characterized in that the determining (S240) of a global adherence score (GSC) further comprises:- Applying (S241) the sensor data as input to a machine learning model, said machine learning model having been trained to output a baseline daily living behavior from the sensor data collected over a period of time when the patient is in a stable condition applied as input, and in that said global adherence score is determined based on said medicine adherence score, said activity adherence score, said baseline daily living behavior and said behavior recommendations for the patient.

5. A method according to any one of the preceding claims, characterised in that the evaluating (S250) of the global adherence score (GSC) comprises:- comparing (S251) the global adherence score (ASC) against one or more given reference thresholds and one or more specific historic levels for the patient,- detecting (S252) an anomaly based on a result of evaluation (RS) and, when it has been decided that the anomaly detected is significant, the reporting (S260) comprises sending (S261) an alert to a caregiver (UT) of the patient.

6. An apparatus (100) for monitoring a behavior of a patient (PT) in a physical environment (ENV), said apparatus being configured to implement the following modules of: obtaining behavioral recommendations (BRC) for the patient by applying input data comprising one or more clinical conditions (CLC) of the patient and contextual data stored in one or more data sources (SRC) related to health and medicine, to a tuned large language model, LLM, said behavioral recommendations being provided as output, providing them to a user through a user interface (III), obtaining sensor data (SD) from a plurality of sensors (C1-C5) placed in the environment (ENV), said sensor data comprising electrical data, environmental data and data representative of medication intake by the patient, determining a global adherence score (GSC) of the patient to the behavioral recommendations based on an analysis of said sensor data, evaluating the global adherence score based on one or more reference thresholds and one or more historic levels of the patient, and reporting a result (RS) of evaluation to the user.

7. An apparatus (100) comprising: at least one processor;at least one memory storing instructions that, when executed by the at least one processor, cause the apparatus to perform the method according to any one of claims 1 to 6.

8. A user equipment (UE) configured to receive sensor data (SD) from a plurality of sensors (C1-C5) arranged in an environment of a patient, said user equipment (UE) comprising an apparatus (100) according to claim 7.

9. A system (S) comprising a user equipment (UE) according to the preceding claim, a plurality of sensors (C1-C5) arranged in the environment of the patient, and a LLM configured to output said behavioral recommendations for the patient.

10. A system (S) according to the preceding claim, said system further comprising a machine learning model (MLM), said machine learning model having been trained to output a baseline daily living behavior from the sensor data applied as input, said baseline daily living behavior being used to determine a global adherence score (GSC) of the patient to behavioral recommendations (BRO) output by the LLM.

11. A computer program comprising instructions for causing a computer comprising one or more processors, when said instructions are executed by said one or more processors, to perform the method for monitoring a behavior of a patient according to any one of claims 1 to 5.

12. A non-transitory computer-readable medium storing instructions thereon that, when executed by one or more processors of an apparatus for monitoring a behavior of a patient in an environment, cause the one or more processors to perform the method for monitoring a behavior of a patient according to any one of claims 1 to 5.

Images