A gastric cancer early screening risk analysis and health management method

Abstract

The application discloses a gastric cancer early screening risk analysis and health management method, and relates to the technical field of gastric cancer early screening and health management. The method comprises the following steps: S1, constructing a multi-dimensional influence factor ontology knowledge base of a high-risk population of gastric cancer, wherein the knowledge base comprises a disease cognition ontology layer, an individual health characteristic ontology layer, an environmental characteristic ontology layer and a health management scheme ontology layer; and S2, establishing a gastroscopy screening compliance influence mechanism analysis model to construct a structural equation model in the knowledge dimension and the belief dimension. The application constructs the multi-dimensional influence factor ontology knowledge base of the high-risk population of gastric cancer and the gastroscopy screening compliance influence mechanism analysis model, integrates multi-source health information to generate an individual multi-dimensional influence factor portrait, matches and generates a personalized health management scheme in combination with a search enhancement generation technology, executes a closed-loop health management process, establishes a scheme iteration optimization mechanism, and comprehensively analyzes specific obstacles and interaction relationships of an individual in the knowledge, belief and behavior three dimensions.

Description

Technical Field

[0001] This invention relates to the field of early gastric cancer screening and health management technology, specifically a method for early gastric cancer screening risk analysis and health management. Background Technology

[0002] Gastric cancer is one of the malignant tumors with high incidence and mortality rates in my country. Early screening can significantly improve patients' survival rate and quality of life. Gastroscopy is currently the gold standard for early gastric cancer screening. Regular gastroscopy screening can detect early gastric cancer and precancerous lesions, allowing for timely intervention and treatment. To increase screening participation rates among high-risk groups for gastric cancer, various health education and management methods have emerged, including online science popularization, telephone follow-ups, and mobile application reminders, aiming to help the target population understand the importance of gastric cancer screening and actively participate in examinations.

[0003] Current methods for early gastric cancer screening and health management still suffer from a problem in practical application: they cannot achieve dynamic and precise matching between health management strategies and the multidimensional influencing factors of individuals. Existing methods typically generate intervention content based on limited demographic information or single risk factors, failing to comprehensively analyze the specific obstacles and their interactions within the three dimensions of knowledge, beliefs, and behavior. Furthermore, existing methods lack the ability to perceive and respond to changes in an individual's psychological state and behavioral intentions in real time. Once an intervention plan is generated, it is difficult to fine-tune it based on the individual's actual feedback and state changes, resulting in a discrepancy between the intervention content and the individual's true needs, and failing to effectively translate into actual screening behavior. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a method for early screening risk analysis and health management of gastric cancer, which solves the problems of low matching degree between intervention strategies and individual needs and lack of dynamic adjustment capability in existing technologies.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for early screening risk analysis and health management of gastric cancer, comprising the following steps: S1. Construct a multidimensional ontology knowledge base of influencing factors for high-risk groups of gastric cancer. The knowledge base includes a disease cognition ontology layer, an individual health characteristic ontology layer, an environmental characteristic ontology layer, and a health management program ontology layer. S2. Establish an analytical model of the mechanism influencing compliance with gastroscopy screening. Construct a structural equation model with knowledge and belief dimensions, taking knowledge dimension as an exogenous variable and behavioral attitude, subjective norms, perceived behavioral control, perceived susceptibility, perceived severity, perceived benefits and perceived obstacles as mediating variables, and gastroscopy screening behavior as an endogenous variable. S3. Collect multidimensional health information of individuals, guide individuals to fill out health questionnaires through an interactive interface, connect with hospital information systems to obtain clinical data, use large language models to perform semantic parsing on unstructured text, and generate multidimensional profiles of influencing factors for individuals; S4. Based on retrieval enhancement generation technology, health management plan matching is achieved. The content of the ontology knowledge base is converted into a high-dimensional vector representation through vector encoding and stored in the vector database. The individual profile is converted into a query vector for semantic similarity matching. The retrieval results are concatenated with the structured prompt template and input into the large language model to generate a personalized plan. S5. Implement a closed-loop health management process, including health assessment, intervention recommendation, intervention implementation, and intelligent follow-up. S6. Establish an iterative optimization mechanism for the solution, record the intervention implementation process and effect data, optimize the path coefficients of the structural equation model based on the effect data, and continuously update the content of the ontology knowledge base.

[0006] Preferably, the step of constructing the ontology knowledge base includes: An ontology construction process combining the skeleton method and a seven-step method is adopted, and a visual ontology development tool is used for development. Integrate disease knowledge, individual health characteristics, environmental factors, and health management plans related to gastric cancer screening; Consistency checks and expert evaluations are performed using an inference engine.

[0007] Preferably, the disease cognition ontology layer includes knowledge of gastric cancer epidemiological characteristics, risk factors, early warning symptoms, screening procedures, and treatment prognosis. The individual health characteristics ontology layer includes demographic information, medical history, family history, lifestyle habits, and psychological state characteristics; The environmental characteristics ontology layer includes social support, accessibility to medical resources, and cultural background factors; The ontology layer of the health management program includes knowledge education, belief guidance, and behavioral support intervention strategies.

[0008] Preferably, the step of establishing the analytical model for the mechanism influencing gastroscopy screening compliance includes: A theoretical framework for constructing structural equation modeling based on knowledge-belief-practice theory and health belief model; We set knowledge as an exogenous variable, and behavioral attitude, subjective norms, perceived behavioral control, perceived susceptibility, perceived severity, perceived benefit, and perceived impairment as mediating variables. We defined gastroscopy screening behavior as an endogenous variable; The goodness of fit between the theoretical model and actual data was tested, and the direct and indirect effects of various factors on screening behavior were quantified.

[0009] Preferably, the step of collecting individual multidimensional health information includes: The interactive interface of the front-end presentation layer guides individuals to fill out a standardized health questionnaire, which covers basic health information, medical history, family history, lifestyle habits, screening awareness level and psychological state. By connecting to the hospital information system through the backend application layer, electronic medical records, test results and previous medical records can be automatically obtained. We use a large language model to perform semantic parsing and feature extraction on the collected unstructured text information, and transform it into structured feature data.

[0010] Preferably, the step of matching health management plans based on search enhancement generation technology includes: All knowledge content in the ontology knowledge base is transformed into high-dimensional vector representations through a vector encoding model, stored in a vector database, and a semantic index is established. The multidimensional profile of an individual's influencing factors is transformed into a query vector. Semantic similarity matching is performed in the vector database to retrieve knowledge fragments and intervention strategies that have a semantic similarity greater than a preset threshold with the individual's features. The retrieved knowledge fragments are combined with structured prompt templates to form enhanced instructions that are then input into the large language model.

[0011] Preferably, the step of generating a personalized solution includes: The large language model combines enhanced instructions with a multi-dimensional profile of an individual's influencing factors to generate a health management plan, which covers three aspects: knowledge education, belief guidance, and behavioral support. The knowledge education content includes knowledge about disease risks, screening procedures, and coping mechanisms. Belief guidance includes strengthening risk perception, enhancing benefit perception, and weakening obstacle perception; Behavioral support includes screening program development, appointment assistance, and psychological counseling.

[0012] Preferably, the steps of executing the closed-loop health management process include: The health assessment process generates a health assessment report based on an individual's multidimensional health information, clarifying the individual's gastric cancer risk level and the type of mechanism affecting screening compliance; The intervention recommendation process generates personalized health management plans based on health assessment results using search-enhanced generation technology. During the intervention implementation phase, health education content, screening reminders, and behavioral guidance are pushed to individuals according to the generated plan, and intelligent Q&A and report interpretation services are provided. The intelligent follow-up process develops a follow-up plan based on the individual's screening results and health status, sends follow-up reminders regularly, and collects the individual's feedback and behavioral data.

[0013] Preferably, the step of establishing the scheme iterative optimization mechanism includes: Record the implementation process and effect data of each intervention, including changes in knowledge level, changes in belief state, transformation of screening behavior, and satisfaction with the program; The path coefficients and weight parameters of the influence mechanism analysis model are optimized through data analysis. If the data on the effectiveness of multiple consecutive interventions fails to reach the preset threshold, the system will automatically trigger a warning of program failure, increase the intensity of intervention according to preset rules, and introduce medical staff for manual intervention.

[0014] Preferably, the method is implemented based on a layered technical architecture, which includes: The front-end presentation layer, data storage layer, retrieval enhancement generation and large language model layer, and back-end application layer interact and collaborate through standardized interfaces. The front-end presentation layer is used for interactive interface display and solution push; The data storage layer is used to store the ontology knowledge base and the vector database; The retrieval enhancement generation and large language model layer are used for semantic retrieval and solution generation; The backend application layer is used to connect to the hospital information system and execute health management processes.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention constructs a multidimensional influencing factor ontology knowledge base for high-risk gastric cancer populations and an analysis model of the influencing mechanism of gastroscopy screening compliance. It integrates multi-source health information to generate an individual multidimensional influencing factor profile, combines retrieval-enhanced generation technology to match and generate personalized health management plans, executes a closed-loop health management process, and establishes an iterative optimization mechanism for the plan. It comprehensively analyzes the specific obstacles and their interactions among individuals in the three dimensions of knowledge, belief, and behavior, achieving dynamic and accurate matching between health management strategies and individual multidimensional influencing factors. It perceives and responds to changes in individual psychological state and behavioral intentions in real time, and refines the intervention plan based on actual individual feedback and state changes, reducing the deviation between intervention content and individual's actual needs, promoting the effective transformation of intervention content into actual gastroscopy screening behavior, and improving the participation rate of early gastric cancer screening and the efficiency of health management. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a diagram illustrating the method for constructing a knowledge base for health management of high-risk groups for gastric cancer according to the present invention. Figure 3 This is a diagram illustrating the adaptation strategy system for the mechanism influencing gastroscopy screening compliance in this invention. Figure 4 This is a flowchart illustrating the operation of the search enhancement generation technology of the present invention; Figure 5 This is a technical architecture diagram of the gastric cancer early screening health management platform based on retrieval enhancement generation and large language model of the present invention; Figure 6 This is a design diagram of the functional modules of the gastric cancer early screening health management platform of the present invention; Figure 7 This is a flowchart illustrating the operation of the gastric cancer early screening health management platform based on enhanced retrieval generation and a large language model, as described in this invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Please see Figure 1 This embodiment provides a method for early screening risk analysis and health management of gastric cancer, which includes the following steps: S1. Construct a multidimensional ontology knowledge base of influencing factors for high-risk groups of gastric cancer. The knowledge base includes a disease cognition ontology layer, an individual health characteristic ontology layer, an environmental characteristic ontology layer, and a health management program ontology layer. S2. Establish an analytical model of the mechanism influencing compliance with gastroscopy screening. Construct a structural equation model with knowledge and belief dimensions, taking knowledge dimension as an exogenous variable and behavioral attitude, subjective norms, perceived behavioral control, perceived susceptibility, perceived severity, perceived benefits and perceived obstacles as mediating variables, and gastroscopy screening behavior as an endogenous variable. S3. Collect multidimensional health information of individuals, guide individuals to fill out health questionnaires through an interactive interface, connect with hospital information systems to obtain clinical data, use large language models to perform semantic parsing on unstructured text, and generate multidimensional profiles of influencing factors for individuals; S4. Based on retrieval enhancement generation technology, health management plan matching is achieved. The content of the ontology knowledge base is converted into a high-dimensional vector representation through vector encoding and stored in the vector database. The individual profile is converted into a query vector for semantic similarity matching. The retrieval results are concatenated with the structured prompt template and input into the large language model to generate a personalized plan. S5. Implement a closed-loop health management process, including health assessment, intervention recommendation, intervention implementation, and intelligent follow-up. S6. Establish an iterative optimization mechanism for the solution, record the intervention implementation process and effect data, optimize the path coefficients of the structural equation model based on the effect data, and continuously update the content of the ontology knowledge base.

[0019] Specifically, this embodiment integrates multi-dimensional knowledge with artificial intelligence technology to achieve precise risk analysis and dynamic health management for high-risk groups of gastric cancer. First, an ontology knowledge base covering all influencing factors is constructed to provide a standardized knowledge foundation for subsequent analysis and intervention. Next, a quantitative model of compliance influence mechanisms is established to analyze the impact path of each factor on screening behavior. Then, multi-source health data of individuals is collected and analyzed to generate refined individual characteristic profiles. Next, retrieval-enhanced generation technology is used to achieve precise matching between knowledge and individual characteristics, generating personalized intervention plans. Subsequently, a closed-loop health management process is implemented to ensure the intervention measures are effectively implemented. Finally, continuous data feedback optimizes the model and knowledge base, forming a self-evolving management system. This process addresses the problems of low matching between intervention strategies and individual needs and lack of dynamic adjustment capabilities in existing technologies, improving the participation rate and management efficiency of early gastric cancer screening.

[0020] In one specific embodiment, the steps for constructing an ontology knowledge base include: An ontology construction process combining the skeleton method and a seven-step method is adopted, and a visual ontology development tool is used for development. Integrate disease knowledge, individual health characteristics, environmental factors, and health management plans related to gastric cancer screening; Consistency checks and expert evaluations are performed using an inference engine.

[0021] Specifically, this embodiment uses Protégé 5.2.0 as the visual ontology development tool, following the top-level conceptual design process of the skeleton method, while also referencing the detailed content construction specifications of the seven-step method. Knowledge sources include authoritative domestic and international gastric cancer screening guidelines, core academic literature, medical terminology standards, and opinions from domain experts. First, the application goal of the ontology is determined to support personalized health management program recommendations. Then, core concepts and relationships between concepts in the field of health management for high-risk gastric cancer populations are extracted to construct the top-level ontology framework. Next, the concepts and attributes of each ontology layer are refined layer by layer to form a complete ontology structure. After the initial construction is completed, a HermiT 1.3.8 inference engine is used for logical consistency checks to identify contradictions and redundancies in the ontology. Subsequently, at least five experts with associate senior or higher professional titles in gastroenterology and health management are invited to evaluate the ontology, and the content is revised based on their opinions to ensure the scientific validity and accuracy of the knowledge base.

[0022] In one specific embodiment, the disease cognition ontology layer includes knowledge of gastric cancer epidemiological characteristics, risk factors, early warning symptoms, screening procedures, and treatment prognosis. The individual health characteristics ontology layer includes demographic information, medical history, family history, lifestyle habits, and psychological state characteristics; The environmental characteristics ontology layer includes social support, accessibility to medical resources, and cultural background factors; The ontology layer of the health management program includes knowledge education, belief guidance, and behavioral support intervention strategies.

[0023] Specifically, in the disease cognition ontology layer, epidemiological characteristics cover data on the incidence and mortality rates of gastric cancer across different regions, ages, and genders; risk factors include biological factors such as Helicobacter pylori infection and chronic atrophic gastritis, as well as lifestyle factors such as smoking and high-salt diets; warning symptoms include typical manifestations such as indigestion, upper abdominal pain, and hematemesis; the screening process clarifies the screening frequency and examination methods for different risk levels; and treatment and prognosis knowledge introduces the treatment plans and survival rate differences between early-stage and advanced-stage gastric cancer. In the individual health characteristics ontology layer, demographic information includes age, gender, and place of residence; medical history records gastric-related diseases and other chronic illnesses; family history focuses on the gastric cancer incidence of first-degree relatives; lifestyle habits cover behaviors such as diet, exercise, smoking, and alcohol consumption; and psychological characteristics include emotional states such as anxiety and depression, as well as attitudes towards the disease. In the environmental characteristics ontology layer, social support includes the health awareness and level of support among family members; accessibility to medical resources involves the distance between residence and medical institutions, and medical insurance coverage; and cultural background factors include traditional concepts and acceptance of medical examinations. In the core layer of the health management program, the knowledge education strategy designs differentiated content for different cognitive levels; the belief guidance strategy changes individual cognition through risk warnings and benefit promotion; and the behavior support strategy provides specific action guidance and convenient services.

[0024] In a specific embodiment, the steps for establishing an analytical model of the mechanism influencing gastroscopy screening compliance include: A theoretical framework for constructing structural equation modeling based on knowledge-belief-practice theory and health belief model; We set knowledge as an exogenous variable, and behavioral attitude, subjective norms, perceived behavioral control, perceived susceptibility, perceived severity, perceived benefit, and perceived impairment as mediating variables. We defined gastroscopy screening behavior as an endogenous variable; The goodness of fit between the theoretical model and actual data was tested, and the direct and indirect effects of various factors on screening behavior were quantified.

[0025] Specifically, this embodiment uses structural equation modeling as the analysis tool. Its general form is derived from the standard theoretical framework of structural equation modeling, and the specific expression is as follows: ; in, The vector of endogenous variables includes behavioral attitudes, subjective norms, perceived behavioral control, perceived susceptibility, perceived severity, perceived benefits, perceived barriers, and gastroscopy screening behavior. This is a vector of exogenous variables, i.e., the knowledge dimension; This is the path coefficient matrix between endogenous variables; This is the path coefficient matrix of exogenous variables on endogenous variables; The model represents the residual vector. Initial path coefficients were calculated using maximum likelihood estimation and fitted using data from a previously collected survey of 200 individuals at high risk for gastric cancer. Goodness-of-fit was assessed using multiple indicators, including the chi-square degree-of-freedom ratio, root mean square error of approximation, comparison fit index, and Tucker-Lewis index. Model analysis quantifies the direct impact of knowledge dimensions on each belief dimension, as well as the direct effect of each belief dimension on endoscopy screening behavior. It also calculates the indirect effect of knowledge dimensions on screening behavior through belief-mediated variables, clarifying the pathways and strengths of each factor's influence. For example, the initial path coefficients of knowledge dimensions on perceived susceptibility can be calculated using pre-experimental data and will be iteratively optimized based on intervention effect data.

[0026] In one specific embodiment, the steps for collecting individual multidimensional health information include: The interactive interface of the front-end presentation layer guides individuals to fill out a standardized health questionnaire, which covers basic health information, medical history, family history, lifestyle habits, screening awareness level and psychological state. By connecting to the hospital information system through the backend application layer, electronic medical records, test results and previous medical records can be automatically obtained. We use a large language model to perform semantic parsing and feature extraction on the collected unstructured text information, and transform it into structured feature data.

[0027] Specifically, the standardized health questionnaire is designed based on previous research findings and includes scales that have undergone reliability and validity testing to ensure data reliability. Individuals complete the questionnaire through a front-end interactive interface, and the system verifies data integrity and logical consistency in real time. The back-end application layer interfaces with the hospital information system through standardized interfaces, supporting automatic synchronization of data from multiple systems such as HIS, EMR, and laboratory systems, avoiding errors and omissions caused by manual data entry. For unstructured text such as medical records and examination report descriptions in electronic medical records, a large language model fine-tuned for the medical field is used for semantic parsing. The model can identify medical entities in the text, such as disease names, symptoms, and examination results, and extract corresponding attribute information, transforming them into structured key-value pair data. For example, for the text stating that a patient experienced upper abdominal pain for two weeks in January 2025, and that gastroscopy revealed chronic atrophic gastritis with intestinal metaplasia, the model can extract key information such as symptoms, time, and examination results, generating structured features for subsequent individual profile construction.

[0028] In a specific embodiment, the steps for matching health management plans based on retrieval enhancement generation technology include: All knowledge content in the ontology knowledge base is transformed into high-dimensional vector representations through a vector encoding model, stored in a vector database, and a semantic index is established. The multidimensional profile of an individual's influencing factors is transformed into a query vector. Semantic similarity matching is performed in the vector database to retrieve knowledge fragments and intervention strategies that have a semantic similarity greater than a preset threshold with the individual's features. The retrieved knowledge fragments are combined with structured prompt templates to form enhanced instructions that are then input into the large language model.

[0029] Specifically, this embodiment uses BERT-base-chinese as the vector encoding model to transform each piece of knowledge content in the ontology knowledge base into a 768-dimensional high-dimensional vector representation, which is stored in the Milvus vector database. An IVF-FLAT semantic index is also established to improve retrieval efficiency. First, an end-to-end named entity recognition module processes the multi-dimensional influencing factor profile text of an individual, extracting a list of medical entities. ,in Indicates the first Each entity and its start and end index positions in the image text. For each extracted entity... It uses the same coding model as the ontology knowledge base. Encode it and calculate its relationship with all entities in the ontology knowledge base. The inner product similarity is calculated using a formula derived from the standard similarity calculation method in the entity linking domain. The specific expression is: ; in, The first of the individual characteristics Vector representation of an entity, For the first in the ontology knowledge base Vector representations of each entity. Then select the vectors with the highest similarity exceeding a preset threshold. entity and Linked, among which This is a threshold constant, determined through ROC curve analysis, for example, set to 0.7. For entities successfully linked to the ontology knowledge base... Retrieve related knowledge, including entity names, relationships, and attributes, and also link them with... Connected triples or subgraphs are incorporated as supplementary knowledge to enrich the background knowledge base. Finally, the retrieved relevant knowledge is standardized using a language model. It is temporarily stored as a knowledge chain, and then used as part of the prompt words to be input into a large language model to generate a personalized health management plan. The formula is derived from the standard generation framework of retrieval enhancement generation technology, and the specific expression is: ; in, This is a generative model that has been specifically fine-tuned for the medical field. The system retrieves text to create a multidimensional profile of an individual's influencing factors. The retrieved knowledge fragments are sorted by similarity from highest to lowest, and the top five most relevant pieces are concatenated with a structured prompt template. The prompt template includes task instructions, individual characteristic descriptions, a standardized knowledge chain, and output format requirements. This enhanced instruction is input into a large language model to ensure the generated solution conforms to the knowledge base and meets the individual's needs. For example, an individual's multidimensional profile might include the entity "chronic atrophic gastritis," which is encoded into a corresponding vector by the encoding model. The ontology knowledge base contains a vector corresponding to the "chronic atrophic gastritis" entity; the inner product similarity between the two is 0.85, greater than the preset threshold of 0.7, therefore the two entities are linked. Subsequently, knowledge related to "chronic atrophic gastritis" is retrieved, including screening frequency of once every 1-2 years and lifestyle adjustment recommendations, and this knowledge is standardized into... .Will These are combined with other individual characteristics to form enhancement instructions, which are then input into a large language model to generate a fragment of a personalized health management plan that includes targeted screening plans and dietary guidance.

[0030] In one specific embodiment, the steps for generating a personalized solution include: The large language model combines enhanced instructions with a multi-dimensional profile of an individual's influencing factors to generate a health management plan, which covers three aspects: knowledge education, belief guidance, and behavioral support. The knowledge education content includes knowledge about disease risks, screening procedures, and coping mechanisms. Belief guidance includes strengthening risk perception, enhancing benefit perception, and weakening obstacle perception; Behavioral support includes screening program development, appointment assistance, and psychological counseling.

[0031] Specifically, the large language model generates customized health management plans based on the requirements of enhanced instructions and the specific characteristics of each individual. The knowledge education section adjusts the depth of content according to the individual's cognitive level. For individuals with lower cognitive levels, it focuses on basic disease risks and screening procedures, using a combination of text and graphics to improve readability; for individuals with higher cognitive levels, it supplements the content with more evidence-based medicine and the latest research progress. The belief guidance section intervenes in the individual's main belief barriers. For example, for individuals with screening phobia, it emphasizes the safety and comfort of painless gastroscopy; for individuals who believe screening is useless, it provides comparative data on the survival rates of early-stage and advanced-stage gastric cancer. The behavioral support section provides specific and actionable guidance, including screening time suggestions, recommendations for nearby medical institutions, appointment process guidance, and a checklist for pre-examination preparation. Simultaneously, for individuals with psychological anxiety, it provides relaxation training methods and psychological counseling resources to help alleviate their tension. The generated plans are presented in natural language, with a clear structure, making them easy to understand and implement.

[0032] In one specific embodiment, the steps for implementing a closed-loop health management process include: The health assessment process generates a health assessment report based on an individual's multidimensional health information, clarifying the individual's gastric cancer risk level and the type of mechanism affecting screening compliance; The intervention recommendation process generates personalized health management plans based on health assessment results using search-enhanced generation technology. During the intervention implementation phase, health education content, screening reminders, and behavioral guidance are pushed to individuals according to the generated plan, and intelligent Q&A and report interpretation services are provided. The intelligent follow-up process develops a follow-up plan based on the individual's screening results and health status, sends follow-up reminders regularly, and collects the individual's feedback and behavioral data.

[0033] Specifically, the health assessment process comprehensively considers factors such as an individual's age, family history, medical history, and lifestyle habits to categorize gastric cancer risk into low, moderate, and high levels. Simultaneously, based on the analysis results of the adherence impact mechanism model, the main obstacle types for each individual are identified, such as knowledge deficiency, fear and anxiety, and economic hardship. The intervention recommendation process, based on the assessment results, utilizes a search-enhanced generation module to generate targeted intervention plans. The intervention implementation process pushes content to individuals through the front-end presentation layer, with the push frequency adjusted according to individual characteristics; for example, the push frequency is appropriately increased for individuals with low screening willingness. The intelligent Q&A function uses a large language model to answer common questions individuals have about gastric cancer screening and health management. The report interpretation function allows individuals to upload test reports, and the system automatically identifies abnormal indicators and provides interpretation suggestions. The intelligent follow-up process develops a follow-up plan based on the individual's screening results and risk level; for example, low-risk individuals are followed up annually, while high-risk individuals are followed up every 6 months. Follow-up activities include inquiries about health status, screening reminders, and lifestyle guidance. The system regularly sends follow-up reminders and collects individual feedback and behavioral data for subsequent program optimization.

[0034] In a specific embodiment, the steps for establishing a scheme iterative optimization mechanism include: Record the implementation process and effect data of each intervention, including changes in knowledge level, changes in belief state, transformation of screening behavior, and satisfaction with the program; The path coefficients and weight parameters of the influence mechanism analysis model are optimized through data analysis. If the data on the effectiveness of multiple consecutive interventions fails to reach the preset threshold, the system will automatically trigger a warning of program failure, increase the intensity of intervention according to preset rules, and introduce medical staff for manual intervention.

[0035] Specifically, the system automatically records the push time, content, individual viewing status, and feedback information for each intervention. Effectiveness data is obtained through periodically distributed follow-up questionnaires and clinical data. Changes in knowledge level are assessed by comparing before-and-after scores on a cognitive questionnaire; changes in belief status are assessed by changes in scores on a health belief scale; screening behavior conversion records whether individuals complete gastroscopy screening after the intervention; and protocol satisfaction is assessed through a satisfaction questionnaire. Model parameter optimization uses the gradient descent method, and the update formula for the path coefficients is: ; in, Set the learning rate, for example, to 0.01; This represents the difference between the actual screening behavior and the model's predicted behavior. This represents the change in the path coefficient. Through continuous iterative optimization, the model can gradually improve the predictive accuracy of screening behaviors. When the effect data of an individual still does not reach the preset threshold after three consecutive interventions, the system automatically triggers a program failure warning. At this time, the system will increase the intensity of the intervention, such as increasing the frequency of push notifications, providing more targeted content, and pushing the individual's information to medical staff, who will then conduct manual telephone follow-ups or face-to-face guidance to ensure the intervention effect.

[0036] In one specific embodiment, the method is implemented based on a layered technical architecture, which includes: The front-end presentation layer, data storage layer, retrieval enhancement generation and large language model layer, and back-end application layer interact and collaborate through standardized interfaces. The front-end presentation layer is used for interactive interface display and solution push; The data storage layer is used to store the ontology knowledge base and the vector database; The retrieval enhancement generation and large language model layer are used for semantic retrieval and solution generation; The backend application layer is used to connect to the hospital information system and execute health management processes.

[0037] Specifically, the front-end presentation layer is built using HTML, CSS, and JavaScript technologies, combined with the Vue.js framework and the Django template engine, providing a user-friendly interface that supports functions such as questionnaire completion, report uploading, and solution viewing. The data storage layer employs a hybrid storage architecture, using a PostgreSQL relational database to store structured individual health data and metadata from the ontology knowledge base; a MongoDB non-relational database to store unstructured text data and intervention records; and a Milvus vector database to store high-dimensional vector representations and semantic indexes. The retrieval enhancement generation and large language model layer provides standardized interfaces through FastAPI, enabling functions such as vector encoding, semantic retrieval, knowledge concatenation, and solution generation. The back-end application layer is developed based on the Django MTV architecture, implementing core business logic such as user management, data integration, health assessment, intervention execution, and follow-up management. Data interaction between layers is achieved through RESTful APIs, ensuring system decoupling and scalability. This architecture can support large-scale concurrent user access, guaranteeing system stability and response speed.

[0038] In summary, this invention constructs a multidimensional influencing factor ontology knowledge base for high-risk gastric cancer populations and an analysis model of the influencing mechanism of gastroscopy screening compliance. It integrates multi-source health information to generate individual multidimensional influencing factor profiles, combines retrieval-enhanced generation technology to match and generate personalized health management plans, executes a closed-loop health management process, and establishes an iterative optimization mechanism for the plans. It comprehensively analyzes the specific obstacles and their interactions among individuals in the three dimensions of knowledge, belief, and behavior, achieving dynamic and accurate matching between health management strategies and individual multidimensional influencing factors. It perceives and responds to changes in individual psychological state and behavioral intentions in real time, and refines the intervention plan based on actual individual feedback and state changes, reducing the deviation between intervention content and individual's actual needs, promoting the effective transformation of intervention content into actual gastroscopy screening behavior, and improving the participation rate of early gastric cancer screening and the efficiency of health management.

[0039] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0040] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for early screening risk analysis and health management of gastric cancer, characterized in that, Includes the following steps: S1. Construct a multidimensional ontology knowledge base of influencing factors for high-risk groups of gastric cancer. The knowledge base includes a disease cognition ontology layer, an individual health characteristic ontology layer, an environmental characteristic ontology layer, and a health management program ontology layer. S2. Establish an analytical model of the mechanism influencing compliance with gastroscopy screening. Construct a structural equation model with knowledge and belief dimensions, taking knowledge dimension as an exogenous variable and behavioral attitude, subjective norms, perceived behavioral control, perceived susceptibility, perceived severity, perceived benefits and perceived obstacles as mediating variables, and gastroscopy screening behavior as an endogenous variable. S3. Collect multidimensional health information of individuals, guide individuals to fill out health questionnaires through an interactive interface, connect with hospital information systems to obtain clinical data, use large language models to perform semantic parsing on unstructured text, and generate multidimensional profiles of influencing factors for individuals; S4. Based on retrieval enhancement generation technology, health management plan matching is achieved. The content of the ontology knowledge base is converted into a high-dimensional vector representation through vector encoding and stored in the vector database. The individual profile is converted into a query vector for semantic similarity matching. The retrieval results are concatenated with the structured prompt template and input into the large language model to generate a personalized plan. S5. Implement a closed-loop health management process, including health assessment, intervention recommendation, intervention implementation, and intelligent follow-up. S6. Establish an iterative optimization mechanism for the solution, record the intervention implementation process and effect data, optimize the path coefficients of the structural equation model based on the effect data, and continuously update the content of the ontology knowledge base.

2. The method for early screening risk analysis and health management of gastric cancer according to claim 1, characterized in that, The steps for constructing the ontology knowledge base include: An ontology construction process combining the skeleton method and a seven-step method is adopted, and a visual ontology development tool is used for development. Integrate disease knowledge, individual health characteristics, environmental factors, and health management plans related to gastric cancer screening; Consistency checks and expert evaluations are performed using an inference engine.

3. The method for early screening risk analysis and health management of gastric cancer according to claim 1, characterized in that, The disease cognition ontology layer includes knowledge on the epidemiological characteristics, risk factors, early warning symptoms, screening procedures, and treatment prognosis of gastric cancer. The individual health characteristics ontology layer includes demographic information, medical history, family history, lifestyle habits, and psychological state characteristics; The environmental characteristics ontology layer includes social support, accessibility to medical resources, and cultural background factors; The ontology layer of the health management program includes knowledge education, belief guidance, and behavioral support intervention strategies.

4. The method for early screening risk analysis and health management of gastric cancer according to claim 1, characterized in that, The steps for establishing the analytical model of the mechanism affecting gastroscopy screening compliance include: A theoretical framework for constructing structural equation modeling based on knowledge-belief-practice theory and health belief model; We set knowledge as an exogenous variable, and behavioral attitude, subjective norms, perceived behavioral control, perceived susceptibility, perceived severity, perceived benefit, and perceived impairment as mediating variables. We defined gastroscopy screening behavior as an endogenous variable; The goodness of fit between the theoretical model and actual data was tested, and the direct and indirect effects of various factors on screening behavior were quantified.

5. The method for early screening risk analysis and health management of gastric cancer according to claim 1, characterized in that, The steps for collecting individual multidimensional health information include: The interactive interface of the front-end presentation layer guides individuals to fill out a standardized health questionnaire, which covers basic health information, medical history, family history, lifestyle habits, screening awareness level and psychological state. By connecting to the hospital information system through the backend application layer, electronic medical records, test results and previous medical records can be automatically obtained. We use a large language model to perform semantic parsing and feature extraction on the collected unstructured text information, and transform it into structured feature data.

6. The method for early screening risk analysis and health management of gastric cancer according to claim 1, characterized in that, The steps for matching health management plans based on retrieval enhancement generation technology include: All knowledge content in the ontology knowledge base is transformed into high-dimensional vector representations through a vector encoding model, stored in a vector database, and a semantic index is established. The multidimensional profile of an individual's influencing factors is transformed into a query vector. Semantic similarity matching is performed in the vector database to retrieve knowledge fragments and intervention strategies that have a semantic similarity greater than a preset threshold with the individual's features. The retrieved knowledge fragments are combined with structured prompt templates to form enhanced instructions that are then input into the large language model.

7. The method for early screening risk analysis and health management of gastric cancer according to claim 1, characterized in that, The steps for generating a personalized solution include: The large language model combines enhanced instructions with a multi-dimensional profile of an individual's influencing factors to generate a health management plan, which covers three aspects: knowledge education, belief guidance, and behavioral support. The knowledge education content includes knowledge about disease risks, screening procedures, and coping mechanisms. Belief guidance includes strengthening risk perception, enhancing benefit perception, and weakening obstacle perception; Behavioral support includes screening program development, appointment assistance, and psychological counseling.

8. The method for early screening risk analysis and health management of gastric cancer according to claim 1, characterized in that, The steps for implementing the closed-loop health management process include: The health assessment process generates a health assessment report based on an individual's multidimensional health information, clarifying the individual's gastric cancer risk level and the type of mechanism affecting screening compliance; The intervention recommendation process generates personalized health management plans based on health assessment results using search-enhanced generation technology. During the intervention implementation phase, health education content, screening reminders, and behavioral guidance are pushed to individuals according to the generated plan, and intelligent Q&A and report interpretation services are provided. The intelligent follow-up process develops a follow-up plan based on the individual's screening results and health status, sends follow-up reminders regularly, and collects the individual's feedback and behavioral data.

9. The method for early screening risk analysis and health management of gastric cancer according to claim 1, characterized in that, The steps for establishing the iterative optimization mechanism for the scheme include: Record the implementation process and effect data of each intervention, including changes in knowledge level, changes in belief state, transformation of screening behavior, and satisfaction with the program; The path coefficients and weight parameters of the influence mechanism analysis model are optimized through data analysis. If the data on the effectiveness of multiple consecutive interventions fails to reach the preset threshold, the system will automatically trigger a warning of program failure, increase the intensity of intervention according to preset rules, and introduce medical staff for manual intervention.

10. A method for early screening risk analysis and health management of gastric cancer according to claim 1, characterized in that, The method is implemented based on a layered technical architecture, which includes: The front-end presentation layer, data storage layer, retrieval enhancement generation and large language model layer, and back-end application layer interact and collaborate through standardized interfaces. The front-end presentation layer is used for interactive interface display and solution push; The data storage layer is used to store the ontology knowledge base and the vector database; The retrieval enhancement generation and large language model layer are used for semantic retrieval and solution generation; The backend application layer is used to connect to the hospital information system and execute health management processes.

Images