Health intervention list generation method and device, electronic equipment and storage medium

Abstract

The application provides a health intervention list generation method and device, electronic equipment and a storage medium, applied to the field of artificial intelligence technology, medical intervention weights are determined according to medical symptom data of a target object, the more serious the illness indicated by the medical symptom data, the greater the medical intervention weight, so that the medical intervention weight can be determined according to the actual physical condition of the target object; when the target object is inclined to a strong medical order and a high-risk physical condition, the medical intervention weight is large, and medical intervention projects are mainly used; when the target object is inclined to a sub-health and stable physical condition, the medical intervention weight is small, and life intervention projects are mainly used. The adaptive intervention of 'treating the symptoms (emphasizing medicine) in an emergency and treating the root cause (emphasizing life)' can be realized, and the rationality of health intervention list recommendation is improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, apparatus, electronic device, and storage medium for generating a health intervention list. Background Technology

[0002] Community-based health management for the elderly is a crucial aspect of addressing the challenges of an aging population. To overcome the "illusion" and lack of professionalism inherent in Large Language Models (LLMs) in the medical field, intelligent health systems based on Retrieval-Augmented Generation (RAG) technology have become a mainstream research area.

[0003] Existing technologies primarily focus on the accuracy of clinical diagnosis, with applications mainly limited to single medical question-and-answer or hospital admission / discharge decisions. These technologies often employ unidirectional retrieval strategies based on semantic similarity or causal chains, achieving breakthroughs in the depth and accuracy of medical diagnostic reasoning. However, when applied to the specific scenario of daily, continuous health interventions for elderly people in the community, significant logical flaws and functional blind spots remain. In community-based elderly care, the elderly frequently switch between the "acute phase of illness" (requiring strict adherence to medical advice) and the "sub-healthy / stable phase" (requiring lifestyle adjustments). How to provide different health intervention lists at different stages to achieve adaptive intervention—"treating the symptoms in urgent cases (emphasizing medicine) and addressing the root causes in less urgent cases (emphasizing lifestyle)"—and improve the rationality of health intervention recommendations, has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and storage medium for generating a health intervention list, in order to improve the rationality of health intervention list recommendations.

[0005] This invention provides a method for generating a health intervention list, comprising the following steps.

[0006] Obtain medical symptom data of the target population; Based on the aforementioned medical symptom data, a sequence of medical intervention programs and a sequence of lifestyle intervention programs were determined. The weight of medical intervention is determined based on the medical symptom data, whereby the weight of medical intervention represents the importance of the medical intervention item; the more severe the condition indicated by the medical symptom data, the greater the weight of the medical intervention. The medical intervention project sequence and the life intervention project sequence are screened according to the medical intervention weight to obtain the target medical intervention project and the target life intervention project; Based on the target medical intervention projects and the target lifestyle intervention projects, a health intervention list is generated using a large language model.

[0007] According to the health intervention list generation method provided by the present invention, the step of determining the medical intervention weights based on the medical symptom data includes: The health risk score and medical order status of the target subject are determined based on the medical symptom data, wherein the health risk score represents the degree of abnormality of the target subject's physiological indicators, and the medical order status indicates whether a medical order exists; The weight of medical intervention is determined based on the status of the medical orders and the health risk score.

[0008] According to the health intervention list generation method provided by the present invention, the step of determining the medical intervention weight based on the medical order status and the health risk score includes: Obtain the preset weight parameters and preset bias terms; The medical order status, the health risk score, and the preset bias item are weighted according to the preset weight parameters, and the weighting results are normalized to obtain the medical intervention weight.

[0009] According to the health intervention list generation method provided by the present invention, the step of filtering the items in the medical intervention item sequence and the lifestyle intervention item sequence according to the medical intervention weight to obtain target medical intervention items and target lifestyle intervention items includes: The number of medical intervention items and the number of lifestyle intervention items are determined based on the aforementioned medical intervention weights. The target medical intervention project is obtained by selecting a corresponding number of medical intervention projects from the sequence of medical intervention projects according to the number of medical intervention projects mentioned above. If a medical order exists in the medical symptom data, improve the ranking of the lifestyle intervention items related to the medical order in the lifestyle intervention item sequence; The target life intervention program is obtained by selecting a corresponding number of life intervention programs from the current life intervention program sequence according to the stated number of life intervention programs.

[0010] According to the health intervention list generation method provided by the present invention, before selecting a corresponding number of life intervention items from the current life intervention item sequence according to the number of life intervention items to obtain the target life intervention items, the method further includes: Obtain the behavioral capability data of the target object; Based on the behavioral ability data, the behavioral ability level of the target object is determined; Adjust the sequence of life intervention programs according to the stated level of behavioral ability.

[0011] According to the health intervention list generation method provided by the present invention, the step of generating a health intervention list based on the target medical intervention items and the target lifestyle intervention items using a large language model includes: Generate prompts that include the target medical intervention program and the target lifestyle intervention program; The prompt words are input into a large language model to obtain a preliminary intervention plan; The large language model is used to analyze the preliminary intervention plan and the behavioral ability data of the target object. Items in the preliminary intervention plan that do not conform to the behavioral ability of the target object are revised to obtain a health intervention list.

[0012] The present invention also provides a health intervention list generation device, comprising the following modules: The object data acquisition module is used to acquire the medical symptom data of the target object; The project sequence determination module is used to determine the sequence of medical intervention projects and the sequence of lifestyle intervention projects based on the medical symptom data, wherein the medical intervention weight represents the importance of the medical intervention project; the more severe the condition indicated by the medical symptom data, the greater the medical intervention weight. An intervention weight determination module is used to determine the medical intervention weights based on the medical symptom data. The target project determination module is used to filter the projects in the medical intervention project sequence and the lifestyle intervention project sequence according to the medical intervention weight, so as to obtain the target medical intervention projects and the target lifestyle intervention projects. The intervention list generation module is used to generate a health intervention list based on the target medical intervention items and the target lifestyle intervention items using a large language model.

[0013] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the health intervention list generation method as described above.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the health intervention list generation method as described above.

[0015] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the health intervention list generation method as described above.

[0016] The health intervention list generation method, device, electronic device, and storage medium provided by this invention determine the weight of medical interventions based on the medical symptom data of the target subject. The more severe the condition indicated by the medical symptom data, the greater the weight of the medical intervention. This allows the weight of medical interventions to be determined based on the actual physical condition of the target subject. When the target subject is in a condition requiring strong medical orders or high risk, the weight of medical interventions is high, and medical intervention items are prioritized. When the target subject is in a sub-healthy or stable condition, the weight of medical interventions is low, and lifestyle intervention items are prioritized. This enables adaptive intervention that addresses symptoms in urgent cases (emphasizing medical interventions) and addresses underlying health issues in less urgent cases (emphasizing lifestyle interventions), improving the rationality of the recommended health intervention list. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 This is one of the flowcharts illustrating the health intervention list generation method provided by the present invention; Figure 2 This is the second flowchart illustrating the health intervention list generation method provided by the present invention; Figure 3 This is the third flowchart of the health intervention list generation method provided by the present invention; Figure 4 This is one possible implementation of step 104 in the health intervention list generation method provided by the present invention; Figure 5 This is the fourth flowchart of the health intervention list generation method provided by the present invention; Figure 6 This is a schematic diagram of the health intervention list generation device provided by the present invention; Figure 7 This is a functional diagram of each module in the health intervention list generation device provided by the present invention; Figure 8 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] RAG research in related technologies mainly focuses on enhancing the factuality and reasoning ability of models by introducing external knowledge bases. Its technological evolution mainly shows the following key trends: The leap from text retrieval to structured knowledge graph reasoning: To overcome the limitations of traditional RAG models in lacking structured reasoning capabilities, related technologies have begun to deeply integrate medical knowledge graphs (KG). A typical example is the Medical Retrieval-Augmented Generation (MedRAG) model, which is no longer limited to unstructured text retrieval but constructs a hierarchical diagnostic knowledge graph based on the "entity-relationship-entity" triple form and combines it with electronic health records (EHR) to achieve structured reasoning, significantly improving the accuracy and specificity of disease diagnosis. In addition, the Reinforced Reasoning-Augmented Generation (ReinRAG) model introduces a reinforcement learning-driven reasoning enhancement mechanism. Through a group-based retrieval optimization strategy, it guides the model to learn "reasoning leaps" across semantic clusters on the knowledge graph, in order to derive more complete discharge instructions from limited medical records.

[0021] Simulating Clinical Workflow and Causal Thinking Chains: To enable the system to possess a "doctor's" mindset, researchers proposed a framework combining causal reasoning. For example, the Medical Causal Chain-of-Thought Retrieval-Augmented Generation (MedCoT-RAG) model, targeting the problem of "hallucination generation," introduces causal perception retrieval and structured chain-of-thought (CoT), forcing the model to follow the clinical logical chain of "symptom analysis—pathological mechanism—differential diagnosis—evidence synthesis" for generation. The DoctorRAG (Medical RAG Fusing Knowledge with Patient Analogy through Textual Gradients) framework further pioneers a new paradigm that integrates "explicit medical knowledge" and "implicit clinical experience." By retrieving similar patient case databases and structured knowledge bases, and using the multi-agent text gradient mechanism of Medical Textual Gradient Optimization (Med-TextGrad) to iteratively optimize the generated results, it simulates the decision-making process of doctors combining experience and guidelines.

[0022] Enhanced system robustness and self-regulation: For high-risk medical scenarios, frameworks such as Turing-Complete Retrieval-Augmented Generation (TC-RAG) attempt to introduce Turing completeness theory into the system. They design a stack-based memory mechanism and operations such as "backtracking" and "summarizing" to give the model the ability to self-monitor and correct errors during the reasoning process, enabling it to stop or correct itself in time when it encounters an incorrect path.

[0023] It is evident that related technologies have evolved from simple text retrieval to advanced RAG systems that integrate knowledge graphs, causal reasoning, and experience bases, achieving significant progress in areas such as assisted diagnosis, pathological analysis, and discharge guidance generation.

[0024] Although the aforementioned technologies (such as MedRAG and DoctorRAG) have made breakthroughs in the depth and accuracy of medical diagnostic reasoning, significant logical flaws and functional blind spots still exist when applied to the specific scenario of daily continuous health intervention for elderly people in the community. These are mainly reflected in the following two aspects: I. Lack of a dynamic weighting mechanism for "medical-life" dual-domain knowledge makes it difficult to cope with state fluctuations in non-clinical scenarios: The core objective of related technologies mainly focuses on "accuracy of clinical diagnosis," and their application scenarios are mostly single medical question-and-answer or hospital admission / discharge decisions. They often employ unidirectional retrieval strategies based on semantic similarity or causal chains, lacking the ability to dynamically balance "medical rigor" and "lifestyle applicability." In community-based elderly care scenarios, the elderly often switch between "acute disease phase" (requiring strict adherence to medical advice) and "sub-health / stable phase" (requiring lifestyle adjustments). Related technologies cannot automatically calculate and adjust the weight ratio of medical intervention and lifestyle intervention in the generated plan based on real-time risk scores (such as blood pressure deviation) or medical advice status (whether there is a clear prescription). This can lead to situations where drug treatment is overemphasized for sub-healthy elderly while neglecting lifestyle adjustments, or where lifestyle suggestions lack enforcement power for high-risk elderly, failing to achieve adaptive intervention that prioritizes "treating the symptoms in acute cases (medical focus) and addressing the root cause in chronic cases (lifestyle focus)."

[0025] Second, the lack of a physical constraint layer based on "individual behavioral capabilities" leads to a lack of safety and feasibility in intervention programs: Although related technologies (such as DoctorRAG) introduce similar case retrieval, their essence remains semantic matching at the text level, generally lacking perception and logical constraints on the physical behavioral capabilities of the elderly (activity of daily living score, fall risk, exercise tolerance). When generating "exercise" or "rehabilitation" suggestions, they are often based solely on disease guidelines (such as "hypertension should involve aerobic exercise") without filtering for "ability suitability." This means that recommending "jogging" or "complex gymnastics" to an elderly person with fall risk or mobility impairment poses a serious safety hazard. Related technologies have failed to establish a causal cutoff or downgrading mechanism between "user ability score" and "intervention intensity" (e.g., automatically filtering exercise suggestions and increasing the weight of static care when low ability is detected), resulting in generated programs that, while theoretically correct, are unusable or even dangerous in actual user implementation.

[0026] To address at least one of the aforementioned problems, embodiments of the present invention provide a method, apparatus, electronic device, and storage medium for generating a health intervention list, which are described below in conjunction with... Figures 1 to 8 Please provide a detailed explanation.

[0027] Figure 1 This is one of the flowcharts illustrating the health intervention list generation method provided by the present invention, such as... Figure 1 As shown, the method includes the following: Step 101: Obtain medical symptom data of the target subject.

[0028] The target audience refers to those who require a health intervention list. For example, in a community-based elderly health management scenario, the target audience could be elderly individuals. Medical symptom data refers to medical data relevant to the target audience. This data includes at least one of structured and unstructured data. Structured data may include at least one of the following: real-time blood pressure, heart rate, sleep score (obtainable through a monitoring device worn by the target audience), and the most recent physical examination results. Unstructured data may include at least one of the following: user self-reports (e.g., "recently dizzy," "recently having trouble sleeping," etc.), and doctors' historical prescriptions or medical orders (obtainable through character recognition technology or text input).

[0029] Step 102: Determine the sequence of medical intervention projects and the sequence of lifestyle intervention projects based on medical symptom data.

[0030] The medical intervention sequence includes multiple medical interventions arranged sequentially. Each medical intervention represents an intervention in the target population using medical means. The earlier a medical intervention appears in the sequence, the stronger its correlation with medical symptom data. Similarly, the lifestyle intervention sequence includes multiple lifestyle interventions arranged sequentially. Each lifestyle intervention represents an intervention in the target population using lifestyle methods. The earlier a lifestyle intervention appears in the sequence, the stronger its correlation with medical symptom data. For details on how to retrieve medical and lifestyle intervention sequences from medical symptom data, please refer to the relevant technical documentation.

[0031] In one example, two sequences can be obtained using a parallel two-domain knowledge retrieval system. This system aims to process two completely different types of knowledge in parallel, providing a rich pool of candidates for subsequent filtering. Path 1: Structured Medical Knowledge Retrieval: Directly accesses a pre-built medical knowledge graph (Medical KG). Based on the user's medical symptom data, it retrieves corresponding disease mechanisms, contraindications, and standard treatment guidelines from the medical knowledge graph. It outputs a sequence of medical intervention projects (e.g., drug contraindications and complication risks for stage III hypertension). For an example, see MedRAG's "KG Searching" module for implementation details.

[0032] Path 2: Unstructured Lifestyle Retrieval: Construct a "Community Health Intervention Rule Base" (including dietary recipes, exercise rehabilitation movements, psychological adjustment techniques, sleep improvement skills, etc.). Use experience retrieval (Patient Base Retrieval), and add historical successful intervention programs for similar cases. Then, use semantic vector retrieval to find dietary and exercise suggestions suitable for the current symptoms. Finally, output a sequence of lifestyle intervention projects (e.g., low-sodium diet recipes, introductory Tai Chi movements, Baduanjin, etc.). An example implementation can be found by referring to DoctorRAG.

[0033] It is understandable that the "medical domain" of Path 1 is not limited to using knowledge graphs, and the "life domain" of Path 2 is not limited to using vector databases. Path 1 can also be vectorized and stored in a vector library, or a small language model (SLM) fine-tuned for medical applications can be used as a parameterized knowledge base. Path 2 can also be constructed as a structured rule tree or graph. Furthermore, Path 1 and Path 2 are not limited to merging after parallel retrieval. A sequential retrieval strategy can be adopted (e.g., first retrieving medical knowledge, and then retrieving appropriate life advice based on keywords in the medical results), or a hybrid search strategy combining keyword matching and semantic vector matching can be used to improve recall.

[0034] Step 103: Determine the weight of medical interventions based on medical symptom data.

[0035] Medical intervention weights indicate the importance of a medical intervention program; the higher the weight, the more important the intervention. Medical symptom data of the target population characterizes their physical condition; the more severe the condition indicated by the symptom data, the higher the intervention weight. Based on this data, it can be determined whether the target population is in a condition requiring strict medical supervision and facing high risk, or in a sub-healthy or stable state, thus further determining the intervention weight. The more the target population leans towards a condition requiring strict medical supervision and facing high risk, the higher the intervention weight; conversely, the more the target population leans towards a sub-healthy or stable state, the lower the intervention weight.

[0036] Step 104: Filter the items in the medical intervention project sequence and the life intervention project sequence according to the medical intervention weight to obtain the target medical intervention project and the target life intervention project.

[0037] The greater the weight of medical intervention, the greater the proportion of the number of target medical intervention items to the first total number, where the first total number is the total number of target medical intervention items and target life intervention items.

[0038] Step 105: Based on the target medical intervention projects and target lifestyle intervention projects, generate a health intervention list using a large language model.

[0039] Generate prompts that include target medical interventions and target lifestyle interventions. Input the prompts into a large language model, which will then output a health intervention list.

[0040] In this embodiment of the invention, medical intervention weights are determined based on the target subject's medical symptom data. The more severe the condition indicated by the medical symptom data, the greater the medical intervention weight. The more the medical symptom data indicates a target subject's condition leaning towards strict medical orders and high risk, the greater the medical intervention weight; conversely, the more the medical symptom data indicates a target subject's condition leaning towards sub-health or stable conditions, the smaller the medical intervention weight. This allows for the determination of medical intervention weights based on the target subject's actual physical condition. When the target subject leans towards a condition leaning towards strict medical orders and high risk, the medical intervention weight is high, with medical intervention programs taking precedence; when the target subject leans towards a condition leaning towards sub-health or stable conditions, the medical intervention weight is low, with lifestyle intervention programs taking precedence. This enables adaptive intervention that addresses symptoms in urgent cases (emphasizing medical intervention) and addresses underlying health issues in less urgent cases (emphasizing lifestyle intervention), improving the rationality of the recommended health intervention list.

[0041] Furthermore, this invention achieves deep synergy and complementarity between medical intervention and lifestyle intervention, resolving the problem of separation between medical and elderly care. The "medical-lifestyle collaborative retrieval mechanism" designed in this invention ensures that non-pharmacological interventions no longer exist in isolation, but rather serve as an effective supplement to drug treatment. It effectively solves the problem of the separation and even potential conflict between medical and lifestyle interventions, providing the elderly with a comprehensive care plan that conforms to the concept of "integrated medical and elderly care" and possesses a high degree of logical consistency.

[0042] exist Figure 1 Based on the illustrated embodiments, see also Figure 2 In one possible implementation, determining the weight of medical interventions based on medical symptom data includes: Step 1031: Determine the target subject's health risk score and medical advice status based on medical symptom data.

[0043] The medical order status indicates whether a medical order exists. The health risk score indicates the degree of abnormality in the target individual's physiological indicators. For example, a health risk score can be a normalized score derived by calculating the deviation of at least one of the vital signs data (such as blood pressure and heart rate) from normal medical reference values ​​using medical symptom data. A higher health risk score indicates a more severe abnormality in the physiological indicators.

[0044] Step 1032: Determine the weight of medical intervention based on the doctor's order status and health risk score.

[0045] When the medical order status is the same, the weight of medical intervention is positively correlated with the health risk score; that is, the higher the health risk score, the greater the weight of medical intervention. When the health risk scores are the same, the weight of medical intervention when the medical order status is "with medical order" is greater than the weight of medical intervention when the medical order status is "without medical order".

[0046] In one possible implementation, the weights of medical interventions can be obtained through weighting and normalization. Determining the weights of medical interventions based on the medical order status and health risk score includes: obtaining preset weight parameters and preset bias terms; weighting the medical order status, health risk score, and preset bias terms according to the preset weight parameters; and normalizing the weighted results to obtain the medical intervention weights.

[0047] The preset weighting parameters include a first weighting parameter for the medical order status and a second weighting parameter for the health risk score. Therefore, the formula for calculating the weight of the medical intervention can be: λ=σ(W1×F RX +W2×Risk score +b); Where λ is the weight of the medical intervention, σ() is the normalization function, which in one example can be the Sigmoid activation function, used to map the linearly weighted result to the interval (0, 1). W1 is the first weight parameter, F RX In the state of having a doctor's order, F is present when there is a doctor's order. RX =1, F when no doctor's order RX =0. W2 is the second weighting parameter, Risk. score For health risk scoring. b is a preset bias item used to adjust for baseline tendency.

[0048] The preset weight parameters and preset biases can be empirical or experimental values, and can be customized for different scenarios. The preset weight parameters are positive numbers, meaning both the first and second weight parameters are positive. From the above formula, it can be concluded that the higher the health risk score, the greater the weight of the medical intervention; and the weight of the medical intervention is significantly greater when there is a doctor's order compared to when there is no doctor's order. The medical intervention weight only affects the quantity of the target medical intervention items and the target lifestyle intervention items, not their specific content.

[0049] This approach achieves real-time dynamic updates and precise proportional adjustments to health intervention plans: Breaking through the limitations of traditional RAG technology's static and unidirectional knowledge fusion, it employs an adaptive weight controller based on the Sigmoid function, enabling real-time sensing of changes in the target individual's multidimensional health status (such as blood pressure fluctuations and sleep scores) and medical orders. By automatically calculating and adjusting the weight ratio of "medical intervention" and "lifestyle intervention," it solves the problem of a single strategy being unable to adapt to fluctuations in the patient's condition. When facing acute pathological risks or strong medical orders, it automatically and significantly increases the weight of medical intervention, emphasizing strict medical control; while in sub-healthy or stable periods, it automatically shifts towards lifestyle adjustments, achieving a qualitative leap from "diagnostic results" to "dynamic intervention plans."

[0050] In this embodiment of the invention, the medical order status and health risk score are strongly correlated with the target subject's physical condition. The medical order status and health risk score can indicate whether the target subject is more inclined to a strong medical order and high-risk physical condition, or more inclined to a sub-healthy and stable physical condition. Thus, the weight of medical intervention can be accurately obtained through the medical order status and health risk score, enabling adaptive intervention of "treating the symptoms in urgent cases (emphasizing medicine) and treating the root cause in chronic cases (emphasizing lifestyle)," thereby improving the rationality of the recommended health intervention list.

[0051] In other possible embodiments, in addition to using medical order status and health risk scores to calculate the weight of medical interventions, more influencing factors can be included, such as: the user's historical adherence record (if the user often does not take medication, the weight of lifestyle interventions may need to be temporarily increased as compensation), seasonal climate factors (such as the weight of cardiovascular medical attention may be automatically increased in winter), or one or more of the user's explicit preference settings (the user's subjective choice is biased towards conservative treatment or active exercise).

[0052] It is not limited to using health risk scores; the weight of medical interventions can also be derived based on medical advice, status, and other health risk indicators. For example: (a) Fuzzy logic control: Vacuum vital signs data such as blood pressure and heart rate and medical order status are fuzzified into fuzzy sets such as "high risk", "medium risk" and "with medical order". The weight coefficients are directly output through preset fuzzy inference rules (such as "IF high risk THEN medical weight is extremely large") to achieve nonlinear smooth control.

[0053] (b) Decision tree or machine learning classifier: Build a lightweight decision tree model (such as Extreme Gradient Boosting (XGBoost)) or random forest, take the features of historical cases as input, train the model to determine whether the current state is "urgently needing medical intervention", "mainly needing lifestyle adjustment" or "mixed intervention", and use the classification probability as the weight coefficient.

[0054] (c) Simple piecewise linear function or lookup table method: Set multiple risk threshold intervals, each interval corresponds to a fixed weight value, and the control effect can be achieved even on low computing power devices.

[0055] exist Figure 1 Based on the illustrated embodiments, see also Figure 3 In one possible implementation, the items in the medical intervention item sequence and the lifestyle intervention item sequence are screened according to medical intervention weights to obtain target medical intervention items and target lifestyle intervention items, including: Step 1041: Determine the number of medical intervention items and the number of lifestyle intervention items based on the weight of medical interventions.

[0056] A higher weight for medical interventions results in a larger number of medical intervention items and a smaller number of lifestyle intervention items. In one example, a preset threshold for the number of items can be obtained. The number of medical intervention items is then calculated by multiplying this threshold by the medical intervention weight, and the number of lifestyle intervention items is obtained by subtracting this threshold from the number of medical intervention items. This threshold can be set according to specific needs, such as 5, 8, or 10. Alternatively, the maximum context entry capacity preset by the large language model can also be used as the threshold for the number of items.

[0057] Step 1042: Select the appropriate number of medical intervention projects from the medical intervention project sequence according to the number of medical intervention projects to obtain the target medical intervention projects.

[0058] The medical intervention sequence includes multiple medical interventions arranged sequentially. The earlier a medical intervention appears in the sequence, the stronger its correlation with medical symptom data. When the number of medical interventions is n, the first n medical interventions are selected as the target medical interventions.

[0059] Step 1043: If no medical orders are available in the medical symptom data, select the corresponding number of lifestyle intervention items from the lifestyle intervention item sequence according to the number of lifestyle intervention items to obtain the target lifestyle intervention items.

[0060] The lifestyle intervention sequence includes multiple lifestyle interventions arranged sequentially. The earlier a lifestyle intervention appears in the sequence, the stronger its correlation with the medical symptom data. In the absence of medical orders in the medical symptom data, when the number of lifestyle interventions is m, the first m medical interventions are selected as the target medical interventions from the sequence.

[0061] Step 1044: If medical orders exist in the medical symptom data, improve the ranking of life intervention items related to medical orders in the life intervention item sequence.

[0062] When a medical order exists, extract the key entities from the order (such as "hypertension" or "diabetes") and prioritize and enhance the items that are medically related to the above key entities in the sequence of lifestyle intervention programs (for example, for a "hypertension" order, the order of "low-sodium diet" and "aerobic exercise" can be improved), to ensure that lifestyle intervention is an effective supplement to medical treatment, rather than existing in isolation.

[0063] Step 1045: Select the appropriate number of life intervention projects from the current life intervention project sequence according to the number of life intervention projects to obtain the target life intervention projects.

[0064] When the number of life intervention projects is m, the first m medical intervention projects are selected from the current medical intervention project sequence as target medical intervention projects.

[0065] In this embodiment of the invention, when a medical order exists, prioritizing the ranking of lifestyle intervention items related to the medical order in the lifestyle intervention item sequence can increase the correlation between the target lifestyle intervention item and the medical order, ensuring that the target lifestyle intervention item and the medical order are related, thereby improving the rationality of the recommended health intervention list. It not only increases the weight of medical interventions but also prioritizes matching lifestyle intervention items (such as "low-sodium diet" recommendations) that are medically related to key entities in the medical order (such as "hypertension" prescriptions) during the lifestyle domain retrieval. This effectively solves the problem of the disconnect or even potential conflict between medical interventions and lifestyle interventions, providing the elderly with a comprehensive care plan with high logical consistency that conforms to the concept of "integrated medical and elderly care."

[0066] exist Figure 3 Based on the illustrated embodiments, see also Figure 4 In one possible implementation, before selecting a corresponding number of lifestyle intervention items from the current lifestyle intervention item sequence according to the number of lifestyle intervention items to obtain the target lifestyle intervention items, the health intervention list generation method of this embodiment of the invention further includes: Step 1046: Obtain the behavioral capability data of the target object.

[0067] Behavioral ability data represents the activity level of a target individual; it can be any assessment scale or data indicator that reflects the target individual's physical function, exercise endurance, or fall risk. As an example, behavioral ability data could be an Activity of Daily Living (ADL) score. In other examples, behavioral ability data could be the Tinetti Balance and Gait Scale, the Frailty Index, or gait parameters (such as gait speed and stride variability) collected in real time via wearable devices / visual sensors.

[0068] Step 1047: Determine the target object's behavioral ability level based on the behavioral ability data.

[0069] Behavioral ability data is mapped to behavioral ability levels according to preset rules. For example, decision trees or machine learning classifiers can be used: a lightweight decision tree model or random forest is built, and the model is trained to determine whether the target object's behavioral ability level belongs to "low ability," "medium ability," or "high ability," using the features of the behavioral ability data as input. Alternatively, a simple piecewise linear function or lookup table method can be used: multiple intervals for behavioral ability levels are predefined, and the interval where the behavioral ability data is located is used as the target object's behavioral ability level.

[0070] Taking ADL (Activities of Daily Living) scores as an example, three behavioral ability levels are pre-defined: "Low Ability / High Risk Mode," "Medium Ability / Standard Mode," and "High Ability / Enhanced Mode." A first score threshold (low score threshold) and a second score threshold (high score threshold) are also set. When the target's ADL score is less than or equal to the first score threshold, the target's behavioral ability level is determined to be "Low Ability / High Risk Mode." When the target's ADL score is greater than the first score threshold but less than the second score threshold, the target's behavioral ability level is determined to be "Medium Ability / Standard Mode." When the target's ADL score is greater than or equal to the second score threshold, the target's behavioral ability level is determined to be "High Ability / Enhanced Mode."

[0071] Step 1048: Adjust the sequence of life intervention projects according to the level of behavioral ability.

[0072] Different behavioral ability levels correspond to different adjustment methods. The higher the behavioral ability level (the stronger the target's motor ability), the higher the ranking of the high-intensity exercise items in the life intervention program sequence; the lower the behavioral ability level (the weaker the target's motor ability), the higher the ranking of the low-intensity exercise items in the life intervention program sequence.

[0073] In one example, when the target's behavioral ability level is "low ability / high risk mode," exercise-related items are weighted lower (lower weight items are ranked lower) or disabled, while items focusing on sleep management, dietary regulation, and sedentary care are weighted higher. When the target's behavioral ability level is "medium ability / standard mode," the default weights are maintained without any additional changes or masking, emphasizing a balanced combination of safety and universality, and increasing the weight of items such as gentle exercise (e.g., walking), a balanced diet, and regular sleep recommendations. When the target's behavioral ability level is "high ability / enhanced mode," exercise-related items are weighted higher, and advanced exercise methods are allowed to be recommended, increasing the proportion of exercise and including proactive health interventions such as Tai Chi, Baduanjin, and brisk walking.

[0074] In this embodiment of the invention, the sequence of life intervention projects is adjusted using the behavioral ability data of the target object, and a hard constraint on behavioral ability is introduced. This can increase the adaptability between the target life intervention projects and the behavioral ability data of the target object, thereby reducing the mismatch between the health intervention list and the actual exercise ability of the target object, and increasing the rationality of the recommended health intervention list.

[0075] exist Figure 1 Based on the illustrated embodiments, see also Figure 5 In one possible implementation, a health intervention list is generated using a large language model based on target medical intervention programs and target lifestyle intervention programs, including: Step 1051: Generate prompts including target medical intervention projects and target lifestyle intervention projects.

[0076] A prompt template is pre-built, and the target medical intervention items and target life intervention items are filled into the prompt template to obtain the prompt.

[0077] In one example, the Prompt template could be: "You are a community health assistant. The current user's status weights are: medical intervention weight [λ], and lifestyle intervention weights [1-λ]. User's medical order status: [F]" RX (If applicable, please emphasize following doctor's orders and generate corresponding lifestyle recommendations based on those orders.) User behavior capability data [D] behav (Key note: Activities of daily living are rated as Ability) score (If the score is low, please avoid recommending exercise and focus on sedentary activities or diet and sleep; if the score is high, please provide comprehensive recommendations that include moderate exercise.) Please generate an intervention plan based on the knowledge retrieved below: [Target Medical Intervention Project]: K med (Including pharmacology, contraindications, etc.), [Target Lifestyle Intervention Program]: K life (Dual matching and screening based on medical order keywords and user behavior data has been performed.) Generation requirements: If there is a medical order, medication treatment must be listed as the primary task. At the same time, dietary / exercise recommendations that complement the drug's mechanism of action (e.g., dietary restrictions during medication) must be generated, and it must be ensured that all recommended actions are within the user's behavioral capabilities. Step 1052: Input the prompt words into the large language model to obtain a preliminary intervention plan.

[0078] Step 1053: Analyze the preliminary intervention plan and the behavioral ability data of the target subjects using a large language model, revise the items in the preliminary intervention plan that do not conform to the behavioral abilities of the target subjects, and obtain a health intervention list.

[0079] The initial intervention plan and the target individual's behavioral ability data (e.g., "needs a cane to walk") are used as input to check whether the actions in the initial intervention plan exceed the target individual's behavioral ability (e.g., the user has difficulty moving, but the plan suggests "jogging" → triggering a conflict). If a conflict is triggered, it is corrected. Correction here can be to delete conflicting items, or downgrade "exercise intervention" to "static care" or "passive rehabilitation"; correction can also be to penalize the relevance score of conflicting items (downweighting), placing them at the bottom of the health intervention list; correction can also be to automatically rewrite incompatible actions as compatible actions using a large language model (e.g., automatically rewriting "standing chest expansion" as "sitting chest expansion"), rather than simply filtering. The output health intervention list mainly includes two parts: key medical implementation (e.g., taking medication A at 8 am, monitoring blood pressure) and lifestyle modifications (e.g., reducing pickled foods at dinner, increasing walking by 20 minutes).

[0080] In this embodiment of the invention, a large language model is used to correct items that do not conform to the behavioral capabilities of the target group, and a safety self-checking mechanism is introduced, which significantly improves the physical safety and feasibility of the health intervention list. It can proactively identify and correct action conflicts in the health intervention list (e.g., downgrading a "jogging" recommendation for people with mobility impairments to "static care" or "passive rehabilitation"). This correction mechanism based on physical capability constraints can reduce the risk of secondary injury caused by inappropriate interventions, ensuring the practical implementation and safe execution of the health intervention list in community and home environments.

[0081] Related technologies often overlook the individual physical tolerance of the elderly, which can easily lead to falls due to exceeding their physical capabilities. In one possible implementation of this invention, it can be combined with... Figure 4 and Figure 5 The implementation of this method introduces both hard constraints on behavioral capabilities and a self-checking mechanism for safety, incorporates ADL scores into the decision-making logic, establishes a graded filtering and content degradation strategy, and, in conjunction with the safety self-checking module of the large language model, significantly improves the physical security and feasibility of the health intervention list.

[0082] The health intervention list generation device provided by the present invention is described below. The health intervention list generation device described below and the health intervention list generation method described above can be referred to in correspondence.

[0083] See Figure 6 The health intervention list generation device includes: The object data acquisition module 601 is used to acquire the medical symptom data of the target object; The project sequence determination module 602 is used to determine the sequence of medical intervention projects and the sequence of lifestyle intervention projects based on medical symptom data. The intervention weight determination module 603 is used to determine the medical intervention weight based on medical symptom data. The medical intervention weight represents the importance of the medical intervention item; the more severe the condition represented by the medical symptom data, the greater the medical intervention weight. The target project determination module 604 is used to filter the projects in the medical intervention project sequence and the life intervention project sequence according to the medical intervention weight to obtain the target medical intervention projects and the target life intervention projects. The intervention list generation module 605 is used to generate a health intervention list based on target medical intervention projects and target lifestyle intervention projects using a large language model.

[0084] In one possible implementation, the intervention weight determination module 603 is specifically used to: determine the health risk score and medical order status of the target subject based on medical symptom data, wherein the health risk score represents the degree of abnormality of the target subject's physiological indicators and the medical order status represents whether a medical order exists; and determine the medical intervention weight based on the medical order status and health risk score.

[0085] In one possible implementation, the intervention weight determination module 603 is specifically used to: obtain preset weight parameters and preset bias items; weight the medical order status, health risk score and preset bias items according to the preset weight parameters, and normalize the weighting results to obtain the medical intervention weight.

[0086] In one possible implementation, the target item determination module 604 is specifically used for: determining the number of medical intervention items and the number of lifestyle intervention items based on the weight of medical interventions; selecting a corresponding number of medical intervention items from the medical intervention item sequence according to the number of medical intervention items to obtain target medical intervention items; if there are no medical orders in the medical symptom data, selecting a corresponding number of lifestyle intervention items from the lifestyle intervention item sequence according to the number of lifestyle intervention items to obtain target lifestyle intervention items; if there are medical orders in the medical symptom data, improving the ranking of lifestyle intervention items related to medical orders in the lifestyle intervention item sequence; and selecting a corresponding number of lifestyle intervention items from the current lifestyle intervention item sequence according to the number of lifestyle intervention items to obtain target lifestyle intervention items.

[0087] In one possible implementation, the target item determination module 604 is further configured to: acquire behavioral ability data of the target object; determine the behavioral ability level of the target object based on the behavioral ability data; and adjust the sequence of life intervention items according to the behavioral ability level.

[0088] In one possible implementation, the intervention list generation module 605 is specifically used to: generate prompts including target medical intervention items and target lifestyle intervention items; input the prompts into a large language model to obtain a preliminary intervention plan; use the large language model to analyze the preliminary intervention plan and the behavioral ability data of the target object, and correct the items in the preliminary intervention plan that do not conform to the behavioral ability of the target object to obtain a health intervention list.

[0089] In one example, the functions of each module in the health intervention list generation device can be as follows: Figure 7 As shown. The object data acquisition module 601 is used for multi-dimensional health status perception and semantic encoding. The object data acquisition module 601 receives multimodal data, including structured vital sign data, unstructured text, and behavioral ability scores. The object data acquisition module 601 processes the multimodal data, converting it into state vectors. Using multimodal encoding, it encodes the unstructured text and structured vital sign data separately to obtain state vectors; and converts the behavioral ability scores into behavioral ability labels. Natural Language Processing (NLP) is used to extract entities and determine whether a "strong medical order" (e.g., mandatory use of antihypertensive drugs) exists. If it exists, a flag F is set to indicate the medical order status. RX =1, or 0 if there is no flag to set the status of the medical order.

[0090] The item sequence determination module 602 can specifically be a parallel dual-threshold knowledge retrieval device, used to implement dual-path retrieval: Path 1: Structured medical knowledge retrieval, obtaining a sequence of medical intervention items. Path 2: Unstructured lifestyle behavior retrieval, obtaining a sequence of lifestyle intervention items.

[0091] The intervention weight determination module 603 can specifically be an adaptive dynamic weight controller, used to calculate a normalized health risk score and calculate the medical intervention weight λ based on the health risk score and the medical order status.

[0092] The target item determination module 604 is used to dynamically truncate and deeply clean the search results (medical intervention item sequences and life intervention item sequences) based on the medical intervention weight λ and behavioral ability data. The target item determination module 604 dynamically allocates the number of items in the medical and life domains based on the medical intervention weight λ and the maximum context item capacity K preset by the large language model, and truncates the corresponding item sequences, selecting the top λK items from the medical intervention item sequences. When a medical order F exists... RX When the threshold is 1, the lifestyle intervention program sequence is updated according to the doctor's orders. A secondary screening of the lifestyle intervention program sequence is performed based on the ADL score, and then the top (1-λ)K items are selected from the current lifestyle intervention program sequence. This achieves reordering and assembly based on dynamic thresholds and behavioral ability constraints.

[0093] The intervention list generation module 605 is used to construct a Prompt, which is then input into the LLM to generate a preliminary intervention plan. The LLM uses behavioral ability tags to perform a safety self-check on the preliminary intervention plan to determine if any motor ability conflicts are triggered. If a motor ability conflict is triggered, the preliminary intervention plan is revised, and the safety self-check is performed again until no motor ability conflict is triggered. If no motor ability conflict is triggered, the current intervention plan is output as a health intervention list.

[0094] Figure 8 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8 As shown, the electronic device may include a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, communications interface 820, and memory 830 communicate with each other via the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a health intervention list generation method. This method includes: acquiring medical symptom data of a target object; determining a sequence of medical intervention items and a sequence of lifestyle intervention items based on the medical symptom data; determining medical intervention weights based on the medical symptom data, wherein the medical intervention weights represent the importance of the medical intervention items; filtering the items in the medical intervention item sequence and the lifestyle intervention item sequence according to the medical intervention weights to obtain target medical intervention items and target lifestyle intervention items; and generating a health intervention list using a large language model based on the target medical intervention items and the target lifestyle intervention items.

[0095] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0096] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the health intervention list generation method provided by the above methods. The method includes: acquiring medical symptom data of a target object; determining a sequence of medical intervention items and a sequence of lifestyle intervention items based on the medical symptom data; determining medical intervention weights based on the medical symptom data, wherein the medical intervention weights represent the importance of the medical intervention items; filtering the items in the sequence of medical intervention items and the sequence of lifestyle intervention items according to the medical intervention weights to obtain target medical intervention items and target lifestyle intervention items; and generating a health intervention list using a large language model based on the target medical intervention items and the target lifestyle intervention items.

[0097] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a method for generating a health intervention list provided by the methods described above. This method includes: acquiring medical symptom data of a target object; determining a sequence of medical intervention items and a sequence of lifestyle intervention items based on the medical symptom data; determining medical intervention weights based on the medical symptom data, wherein the medical intervention weights represent the importance of the medical intervention items; filtering items in the sequence of medical intervention items and the sequence of lifestyle intervention items according to the medical intervention weights to obtain target medical intervention items and target lifestyle intervention items; and generating a health intervention list based on the target medical intervention items and the target lifestyle intervention items using a large language model.

[0098] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0099] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0100] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for generating a health intervention list, characterized in that, include: Obtain medical symptom data of the target population; Based on the aforementioned medical symptom data, a sequence of medical intervention programs and a sequence of lifestyle intervention programs were determined. The weight of medical intervention is determined based on the medical symptom data, whereby the weight of medical intervention represents the importance of the medical intervention item; the more severe the condition indicated by the medical symptom data, the greater the weight of the medical intervention. The medical intervention project sequence and the life intervention project sequence are screened according to the medical intervention weight to obtain the target medical intervention project and the target life intervention project; Based on the target medical intervention projects and the target lifestyle intervention projects, a health intervention list is generated using a large language model.

2. The method according to claim 1, characterized in that, The step of determining the weight of medical intervention based on the medical symptom data includes: The health risk score and medical order status of the target subject are determined based on the medical symptom data, wherein the health risk score represents the degree of abnormality of the target subject's physiological indicators, and the medical order status indicates whether a medical order exists; The weight of medical intervention is determined based on the status of the medical orders and the health risk score.

3. The method according to claim 2, characterized in that, The determination of medical intervention weights based on the medical order status and the health risk score includes: Obtain the preset weight parameters and preset bias terms; The medical order status, the health risk score, and the preset bias item are weighted according to the preset weight parameters, and the weighting results are normalized to obtain the medical intervention weight.

4. The method according to claim 1, characterized in that, The step of filtering the medical intervention project sequence and the lifestyle intervention project sequence according to the medical intervention weight to obtain target medical intervention projects and target lifestyle intervention projects includes: The number of medical intervention items and the number of lifestyle intervention items are determined based on the aforementioned medical intervention weights. The target medical intervention project is obtained by selecting a corresponding number of medical intervention projects from the sequence of medical intervention projects according to the number of medical intervention projects mentioned above. If a medical order exists in the medical symptom data, improve the ranking of the lifestyle intervention items related to the medical order in the lifestyle intervention item sequence; The target life intervention program is obtained by selecting a corresponding number of life intervention programs from the current life intervention program sequence according to the stated number of life intervention programs.

5. The method according to claim 4, characterized in that, Before selecting a corresponding number of life intervention projects from the current life intervention project sequence according to the stated number of life intervention projects to obtain the target life intervention project, the method further includes: Obtain the behavioral capability data of the target object; Based on the behavioral ability data, the behavioral ability level of the target object is determined; Adjust the sequence of life intervention programs according to the stated level of behavioral ability.

6. The method according to claim 1, characterized in that, The health intervention list, generated using a large language model based on the target medical intervention project and the target lifestyle intervention project, includes: Generate prompts that include the target medical intervention program and the target lifestyle intervention program; The prompt words are input into a large language model to obtain a preliminary intervention plan; The large language model is used to analyze the preliminary intervention plan and the behavioral ability data of the target object. Items in the preliminary intervention plan that do not conform to the behavioral ability of the target object are revised to obtain a health intervention list.

7. A health intervention list generation device, characterized in that, include: The object data acquisition module is used to acquire the medical symptom data of the target object; The project sequence determination module is used to determine the sequence of medical intervention projects and the sequence of lifestyle intervention projects based on the medical symptom data. The intervention weight determination module is used to determine the medical intervention weight based on the medical symptom data, wherein the medical intervention weight represents the importance of the medical intervention item; the more severe the condition indicated by the medical symptom data, the greater the medical intervention weight; The target project determination module is used to filter the projects in the medical intervention project sequence and the lifestyle intervention project sequence according to the medical intervention weight, so as to obtain the target medical intervention projects and the target lifestyle intervention projects. The intervention list generation module is used to generate a health intervention list based on the target medical intervention items and the target lifestyle intervention items using a large language model.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the method as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 6.

Images