Medical statistics large model agent, construction method and equipment thereof and storage medium

Abstract

The invention discloses a medical statistics large model agent, a construction method and equipment thereof and a storage medium. The method comprises the following steps: constructing a structured medical statistics knowledge base by performing semantic analysis and knowledge extraction on medical statistics related guidelines, regulatory literatures and academic papers; contextual information of a medical statistics target task is obtained, and the contextual information is provided for a large language model after being subjected to structured semantic conversion through a unified cognitive interface; generating and executing a structured task instruction based on the medical statistics knowledge base and the context information through a large language model; and verifying and optimizing an output result of executing the structured task instruction through a closed-loop reflection iteration mechanism, and improving a statistical analysis process in the large language model based on an optimization result, so as to obtain a medical statistical large model agent. According to the invention, statistics of medical data is realized through the medical statistics large model agent, and the efficiency and quality of medical statistics are improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a medical statistical large model intelligent agent, its construction method, device and storage medium. Background Technology

[0002] With the increasing demands for accuracy and efficiency in statistical analysis from fields such as clinical research, evidence-based medicine, and public health analysis, medical statistics has become an indispensable core support tool in modern medicine. However, current medical statistical work still relies heavily on manual labor. Statisticians not only need to be proficient in various statistical methods and software tools, but also need extensive domain knowledge and experience to complete the entire process from data preprocessing to result interpretation. This process is not only labor-intensive and has a steep learning curve, but it is also easily affected by subjective biases or cognitive limitations in aspects such as method selection, hypothesis construction, and result validation.

[0003] Currently, most existing medical statistical support tools rely on manual operation and single software platforms, making it difficult to automate and intelligently manage complex statistical tasks. While some large language models possess certain natural language understanding and code generation capabilities, the lack of a dedicated medical statistical knowledge base hinders their accurate understanding and application of professional statistical guidelines, regulations, and literature, resulting in a lack of professionalism and rigor in method selection and result interpretation. Furthermore, existing systems fail to fully perceive the contextual information of statistical tasks, such as data characteristics, analysis processes, and intermediate results, lacking a holistic understanding of the task and impacting the model's autonomous decision-making regarding the statistical process. In addition, the lack of an effective intelligent scheduling mechanism for the invocation and execution of statistical methods makes it difficult for models to autonomously retrieve and combine multiple statistical methods, hindering their ability to adapt to diverse statistical needs. More critically, existing systems generally lack closed-loop reflection and iteration mechanisms, making it difficult to perform logical consistency checks, bias analysis, and method suitability assessments on statistical results, and hindering automatic error correction and optimization, thus affecting the reliability and practicality of statistical results. Summary of the Invention

[0004] In view of this, the present invention provides a medical statistical large model intelligent agent and its construction method, device and storage medium, aiming to solve the technical problems of low automation, insufficient professional guarantee and lack of continuous optimization capability in existing medical data statistical work.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for constructing a large-scale medical statistical model agent, comprising: By performing semantic parsing and knowledge extraction on guidelines, regulations, and academic papers related to medical statistics, a structured medical statistics knowledge base is constructed. Receive and parse the specified medical statistical target task; obtain the context information of the medical statistical target task, and provide the context information to the large language model after structured semantic transformation through a unified cognitive interface; Based on the medical statistical knowledge base and the context information, the large language model generates an intent representation for the target task, and generates structured task instructions based on the intent representation; the structured task instructions are then executed to complete the target task and obtain the corresponding output. Through a closed-loop reflection and iteration mechanism, the output is analyzed, verified, and optimized to obtain optimized analysis conclusions. Based on the optimized analysis conclusions, the large language model is trained to improve its statistical analysis process planning ability, thus obtaining a medical statistical large model intelligent agent.

[0006] Secondly, this invention provides a medical statistical large-scale model intelligent agent, comprising: The Medical Statistics Knowledge Base module is used to construct a structured medical statistics knowledge base by performing semantic parsing and knowledge extraction on medical statistics-related guidelines, regulations, and academic papers. The cognitive interface module is used to obtain contextual information for medical statistical target tasks, and to provide the contextual information to the large language model after structured semantic transformation through a unified cognitive interface. The model invocation and execution module is used to invoke the large language model to generate an intent representation and structured task instructions for the target task based on the medical statistical knowledge base and the context information, and execute the structured task instructions to complete the target task and obtain the corresponding output. The closed-loop reflection and iteration module is used to analyze, verify and optimize the output through a closed-loop reflection and iteration mechanism to obtain optimized analysis conclusions, and to train the large language model based on the optimized analysis conclusions to improve its statistical analysis process planning capability.

[0007] Thirdly, the present invention 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 above-described method.

[0008] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method.

[0009] Beneficial effects Compared with the prior art, the present invention has at least the following beneficial effects: This invention constructs a professional medical statistics knowledge base, injecting domain knowledge into a general large language model. This solves the problems of arbitrary method selection and imprecise interpretation caused by a lack of knowledge in professional statistical tasks, and significantly improves the authority and accuracy of statistical analysis.

[0010] By achieving seamless information interaction with large language models through a unified cognitive interface, it can fully perceive the complete context of statistical tasks and make autonomous planning and decisions in conjunction with a knowledge base. This realizes full-process automation from problem understanding to method execution, greatly reducing the reliance on professional statisticians.

[0011] The innovative closed-loop reflection and iteration mechanism of this invention can automatically verify statistical results, analyze deviations, and generate optimization suggestions, enabling the agent to have the ability to self-evaluate and continuously improve, thereby continuously improving the accuracy and reliability of statistical analysis.

[0012] The intelligent agent of this invention can adapt to the complex statistical needs of multiple fields such as clinical research, medical paper writing, and drug review, effectively improving the efficiency and quality of medical statistical work. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart illustrating a method for constructing a large medical statistical model agent, as provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of the intelligent agent structure of the large medical statistical model provided in an embodiment of the present invention. Detailed Implementation

[0015] The embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0016] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. This application can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] Example 1: A Method for Constructing Intelligent Agents for Large-Scale Medical Statistical Models like Figure 1 As shown in the figure, the method for constructing a large medical statistical model agent proposed in this embodiment of the invention includes the following steps: Step S100: Construct a structured medical statistical knowledge base.

[0018] This step aims to transform the professional knowledge scattered across various medical statistics regulations, guidelines, and academic papers into structured, machine-readable, and easily accessible knowledge resources for intelligent agents. A structured medical statistics knowledge base is constructed by performing semantic parsing and knowledge extraction on medical statistics-related guidelines, regulations, and academic papers. Its specific implementation includes: First, we batch-obtained medical statistics-related regulations, guidelines, and academic papers in PDF format from public databases and internal resources. Then, using a PDF text extraction tool, we converted the unstructured content in the PDF documents into editable plain text, ensuring the integrity of the text content and the standardization of the format.

[0019] Secondly, for the extracted text, the system performs coarse-grained segmentation based on the hierarchical structure of medical literature (such as chapters, sections, and subsections), breaking the text down into multiple basic content blocks with clearly defined themes. Medical literature includes at least one of guidelines, regulatory documents, and academic papers. Next, a large language model (a large pre-trained language model) is invoked to perform more fine-grained semantic segmentation and parsing on each basic content block, identifying the semantic boundaries and specific knowledge units of the basic content block (e.g., statistical method definitions, applicable conditions, data requirements, formulas, etc.).

[0020] Finally, based on semantic boundaries and knowledge units, and combining the text content of each content block with the overall overview information of its document, embedding representation techniques (such as BERT, Sentence-Transformer, etc.) are used to construct high-dimensional semantic vectors from the text content. All embedded vectors, along with their corresponding original text and metadata, constitute the core of the medical statistical knowledge base, achieving structured storage and efficient semantic retrieval.

[0021] For example, when analyzing survival data, the knowledge base can provide detailed explanations, applicable scenarios, and key points for implementation of methods such as the Kaplan-Meier method and the Cox proportional hazards model.

[0022] Step S110: Obtain context information and provide it through the cognitive interface.

[0023] This step uses a unified cognitive interface to effectively transfer the target task context information from the medical statistics software system to the large language model. Specific implementation includes: First, the system receives and parses the user-specified medical statistical task. Users submit their specific medical statistical analysis needs to the system through natural language interfaces, structured forms, or file uploads. For example, a user might input a natural language command such as, "Please analyze whether there is a significant difference in the decrease in fasting blood glucose between the new drug trial group and the placebo control group after 12 weeks of treatment," or select "Comparison of the means of two independent samples (t-test)" from a preset task template. Upon receiving this input, the system uses natural language understanding technology or a rule parser to transform it into a structured initial task description object that the system can process internally. This clarifies the core problem that the agent needs to solve. For example, the types of medical statistical tasks include statistical model building and / or statistical result generation.

[0024] Then, a unified cognitive interface based on the WebSocket communication protocol is set up in the medical statistics software system (such as RStudio, Python environment, SPSS, etc.). Different request and response contents are defined on the cognitive interface based on the gRPC data format, realizing the structured transmission and semantic conversion of the context information of the medical statistics target task to the large language model. This interface can extract key context information of the medical statistics target task (i.e., the current statistical task) in real time. The context information mainly includes three categories: 1) the structure information of the associated data tables used in the current analysis, such as the table name, field names, data types, field descriptions, and number of data rows; 2) the set of statistical analysis methods that can be called in the current software environment, including the name, function description, parameter configuration requirements, and calling path of each analysis method; 3) the completed historical statistical analysis results, including the analysis type, generated formulas, statistical tables, visualizations, and conclusive descriptions.

[0025] After obtaining the aforementioned context information, the cognitive interface module standardizes and encodes the structured semantically transformed context information into JSON format, resulting in standardized JSON data. Subsequently, a built-in conversion engine automatically parses this JSON data and, based on a preset template, reassembles it into clearly structured and semantically explicit Markdown or specific format prompts. These prompts organize information through titles, lists, tables, code blocks, etc., ensuring that the large language model can comprehensively and accurately understand the complete context of the current statistical task, supporting subsequent intelligent analysis and response generation.

[0026] For example, when analyzing data from a multicenter clinical trial of cardiovascular disease, the system cognitive interface extracts the demographic data tables (data structure) of patients from each center, lists the available methods such as t-tests, ANOVA, and linear regression (set of available methods), and attaches the completed descriptive statistical results (historical results), and finally integrates them into a detailed Markdown report and sends it to the large language model.

[0027] Step S120: The large language model autonomously performs statistical analysis.

[0028] In this step, the large language model automatically generates an intent representation for the aforementioned target task based on the received contextual information (prompt words) and a query of the medical statistics knowledge base, and generates structured task instructions based on the intent representation; the structured task instructions are then executed to complete the target task and obtain the corresponding output. Specifically, this includes: The large language model first understands the task objective, then searches and reasons within a medical statistics knowledge base, autonomously selects, combines, and plans a complete statistical analysis process, and generates an intent representation of the target task based on this process. This intent representation can be a structured JSON object, which specifies the proposed sequence of analysis steps, the statistical methods to be used in each step, and the logical relationships between the methods.

[0029] Then, a highly structured task instruction is generated based on this intent representation. Specifically, the structured task instruction is typically represented in JSON format with nested dictionaries or a specific schema, which includes the selected one or more statistical methods (such as ANOVA, post-hoc tests), the parameter names required for each method (such as dependent variable, independent variable, significance level), the corresponding specific data fields (such as systolic blood pressure, treatment group), the set parameter values ​​(such as alpha=0.05), and possible advanced configuration information (such as missing value handling strategy = multiple imputation).

[0030] Finally, the structured task instructions are parsed and converted into code or API calls that can be directly understood and executed by underlying statistical software (such as R or Python's statsmodels library). Subsequently, the structured task instructions are executed via remote procedure calls or script execution to complete the target task and obtain the corresponding output.

[0031] During execution, the analysis progress is monitored in real time through event listening or asynchronous response mechanisms. Upon completion, the results output by the statistical software are obtained, typically including statistical tables, model parameter estimates, graphical visualizations, formula derivations, and analytical conclusions. These multimodal outputs are then subjected to structured parsing and semantic enhancement, integrating them into a unified and well-organized structured analysis report as the output of this round of execution. This achieves intelligent statistical analysis, improving both efficiency and accuracy.

[0032] For example, when faced with the task of "analyzing the differences in the efficacy of drug A in hypertensive patients of different age groups," the large language model, based on the method descriptions and contextual information in the knowledge base, may automatically plan a combination of "grouped descriptive statistics -> analysis of variance -> Kaplan-Meier survival analysis" and generate corresponding executable code or analysis requests. After execution, a complete report containing mean blood pressure, F-value, p-value, survival curves, and conclusion text for each age group will be returned.

[0033] Step S130: Closed-loop reflection iteration and agent improvement.

[0034] This step utilizes a closed-loop reflective iterative mechanism to analyze, verify, and optimize the output of step S120, obtaining optimized analytical conclusions. Based on these conclusions, the large language model is trained to improve its statistical analysis workflow planning capabilities, resulting in a medical statistical large-scale model intelligent agent. Specific implementation includes: First, the output is subjected to a logical consistency check. For example, the logical consistency check includes at least one of the following checks: checking whether the selected statistical method (such as the chi-square test) matches the overall hypothesis of the study (such as comparing means); checking whether the confidence interval (such as 95%) in the report output corresponds to the declared significance level (such as 0.05); and determining whether the data type is consistent with the model premise.

[0035] Secondly, method suitability assessment and bias analysis are performed. The suitability of the selected method to the current data characteristics is evaluated. For example, parametric tests may be unsuitable for small sample data that is clearly not normally distributed. Bias analysis combines statistical diagnostic methods to assess the stability of the results and the model's fit. For instance, residual analysis checks for systematic bias in the model, sensitivity analysis examines the dependence of the conclusions on outliers or parameter variations, bias-variance decomposition observes overfitting or underfitting of the model, and propensity score weighting or stratified analysis can determine estimation bias caused by uneven data distribution.

[0036] Based on the results of logical consistency checks, method suitability assessments, and deviation analysis, if potential problems are found, the system automatically generates optimization and adjustment suggestions for the statistical analysis process based on the built-in rule base and knowledge base (e.g., if the data is not normally distributed, it is recommended to use the Mann-Whitney U test). The system updates the structured task instructions according to the suggestions (returning to step S120) and re-executes the analysis to obtain an optimized analysis conclusion.

[0037] Finally, the complete optimization process—from initial task description to problematic output, optimization suggestions, and post-optimization conclusions—is constructed into high-quality training sample data. This sample data is used to perform supervised fine-tuning (training) of the large language model, updating its internal parameters. Specifically, the training objective is to enable the large language model to learn and generate more accurate and reasonable intent representations and structured task instructions by comparing the analysis processes before and after optimization. Through multiple iterative training iterations, the quality of intent representation generation (i.e., statistical analysis process planning capability) of the large language model will be continuously improved when facing similar statistical scenarios, ultimately resulting in a mature and professional medical statistical large-scale intelligent model.

[0038] Example 2: Intelligent Agent for Large-Scale Medical Statistical Models like Figure 2 As shown in the illustration, this invention provides a medical statistical large-scale model intelligent agent, which includes four core modules. These modules work collaboratively to realize a complete intelligent statistical process from knowledge injection, task understanding, intelligent execution to continuous optimization. Each module interacts with data and instructions through clearly defined interfaces, collectively forming the intelligent agent.

[0039] Module 20, the Medical Statistics Knowledge Base, forms the professional knowledge foundation for the intelligent agent. It constructs a structured medical statistics knowledge base by performing semantic parsing and knowledge extraction on guidelines, regulations, and academic papers in the field of medical statistics. Specifically, this module first acquires relevant PDF documents and converts them into plain text; then, it performs coarse-grained segmentation based on the document's title hierarchy to form basic content blocks; next, it uses a large language model to perform fine-grained semantic segmentation and parsing on each content block, identifying specific knowledge units (such as definitions of statistical methods, applicable conditions, etc.); finally, combining the text content and document overview information, it uses embedding representation technology to construct the knowledge into high-dimensional semantic vectors, forming a structured knowledge base that supports efficient semantic retrieval. The function of this module corresponds to step S100 in the above construction method, providing authoritative and structured domain knowledge support for subsequent analysis.

[0040] Cognitive Interface Module 21: This module serves as a unified bridge for information interaction between the intelligent agent and the external medical statistics software environment. Its core function is to establish and maintain a unified cognitive interface (WebSocket communication). This interface can obtain the context information of the current medical statistics target task and provide this information to the large language model after performing structured semantic transformation. The context information includes: the data structure associated with the current analysis task (such as data table fields and types), the set of statistical analysis methods available in the current software environment, and the completed historical statistical analysis processes and results. This module standardizes and encodes the above information (e.g., converts it to JSON format) and reorganizes it into a clear and semantically explicit prompt word format (e.g., Markdown) according to a preset template, thereby ensuring that the large language model can fully and accurately understand the complete context of the task. This module implements the function of step S110 in the above construction method.

[0041] Model Invocation and Execution Module 22: This module is the task execution engine of the intelligent agent. It is used to invoke the large language model, receive instructions from the large language model, and transform them into specific statistical software executable actions. Specifically, this module invokes the large language model to drive it to generate an intent representation for the target task based on the context information provided by the cognitive interface module 21 and the professional knowledge of the medical statistics knowledge base module 20. This generates a structured task instruction containing information such as specific statistical methods, parameters, and data fields. Then, this module is responsible for parsing the structured instruction, converting it into code or API calls that can be executed by the underlying statistical software (such as statistical libraries in R or Python), and triggering the execution process. After execution, this module obtains the multimodal results (such as statistical tables, graphs, and conclusion text) output by the statistical software, and performs parsing and structuring processing to form a unified output report. This module corresponds to step S120 in the above construction method.

[0042] Closed-loop reflection and iteration module 23: This module is the core of the agent's self-optimization and continuous improvement. It is responsible for in-depth analysis, verification, and optimization of the results output by the model call and execution module 22. The specific workflow includes: first, performing logical consistency verification on the output results (such as whether the method selection matches the assumptions), method suitability judgment (such as whether the method is suitable for the current data type), and deviation analysis (such as performing residual analysis, sensitivity analysis, etc.). Based on the analysis results, if potential problems or optimization space are found, this module will automatically generate adjustment suggestions for the statistical analysis process. These suggestions are fed back to the system to update the structured task instructions and re-execute the analysis, thereby obtaining optimized analysis conclusions. More importantly, this module uses the complete optimization process (initial task, problem output, optimization suggestions, optimization conclusions) as training data to train the large language model, thereby continuously improving the model's ability to plan the statistical process. This module implements the closed-loop reflection and iteration and agent improvement functions in step S130 of the above construction method.

[0043] The above four modules are connected sequentially and logically, together realizing all the functions of the medical statistical big data model intelligent agent, ensuring the core capabilities of the intelligent agent in terms of professionalism, automation, adaptability and continuous evolution.

[0044] Example 3: Electronic Devices and Storage Media This invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the medical statistical large model intelligent agent construction method as described in Embodiment 1.

[0045] This invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the medical statistical large model intelligent agent construction method as described in Embodiment 1.

[0046] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, various modifications and variations can be made to the embodiments of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for constructing an intelligent agent for a large-scale medical statistical model, characterized in that, include: By performing semantic parsing and knowledge extraction on guidelines, regulations, and academic papers related to medical statistics, a structured medical statistics knowledge base is constructed. Receive and parse the specified medical statistics target task; The contextual information of the medical statistical target task is obtained, and the contextual information is provided to the large language model after being transformed into structured semantics through a unified cognitive interface; Based on the medical statistical knowledge base and the context information, the large language model generates an intent representation for the target task, and generates structured task instructions based on the intent representation; the structured task instructions are then executed to complete the target task and obtain the corresponding output. Through a closed-loop reflection and iteration mechanism, the output is analyzed, verified, and optimized to obtain optimized analysis conclusions. Based on the optimized analysis conclusions, the large language model is trained to improve its statistical analysis process planning ability, thus obtaining a medical statistical large model intelligent agent.

2. The method according to claim 1, characterized in that, The process involves analyzing, verifying, and optimizing the output through a closed-loop reflective iteration mechanism to obtain optimized analytical conclusions. Based on these optimized conclusions, the large language model is trained to improve its statistical analysis process planning capabilities, resulting in a medical statistical large-scale model intelligent agent. Specifically, this includes: The output is subjected to logical consistency verification, method adaptability judgment and deviation analysis; Based on the results of the logical consistency check, method adaptability judgment, and deviation analysis, adjustment suggestions for the statistical analysis process are generated, and the structured task instructions are updated and the analysis is re-executed according to the adjustment suggestions to obtain the optimized analysis conclusions. The optimized analysis conclusions and corresponding optimization processes are used as training data to train the large language model, thereby improving the planning capability of the statistical analysis process in the large language model.

3. The method according to claim 2, characterized in that, The logical consistency check includes at least one of the following: whether the overall hypothesis matches the selected statistical test method, whether the significance level corresponds to the confidence interval, and whether the data type matches the model premise; the bias analysis includes at least one of the following: residual analysis, sensitivity analysis, bias-variance decomposition, propensity score weighting, or stratified analysis.

4. The method according to claim 1, characterized in that, The structured task instructions include the selected statistical method, parameter name, data fields, set parameter values, and advanced configuration information.

5. The method according to claim 1, characterized in that, The process of obtaining contextual information for the medical statistical target task and providing this contextual information to the large language model after structured semantic transformation through a unified cognitive interface specifically includes: Establish a unified cognitive interface in the medical statistics software system; The cognitive interface performs structured semantic transformation on the contextual information of the medical statistics target task. The contextual information includes the associated data structure, the set of available statistical analysis methods, and historical statistical analysis results. The contextual information, after being transformed into structured semantics, is provided to the large language model.

6. The method according to claim 5, characterized in that, The structured semantic transformation includes: standardizing and encoding the context information in JSON format to obtain standardized JSON data; and parsing and recombining the standardized JSON data into a prompt word format.

7. The method according to claim 1, characterized in that, The process involves semantic parsing and knowledge extraction from guidelines, regulations, and academic papers related to medical statistics to construct a structured medical statistics knowledge base, specifically including: Coarse-grained segmentation is performed based on the title hierarchy of medical literature, breaking down the document into multiple basic content blocks; wherein, the medical literature includes at least one of guidelines, regulatory documents and academic papers; The large language model is invoked to perform semantic fine-grained segmentation and parsing on each basic content block, and the semantic boundaries and knowledge units of the basic content block are identified. Based on the semantic boundaries and knowledge units, and combined with the overall overview information of the documents to which the basic content blocks belong, a semantic vector is constructed to obtain the medical statistical knowledge base.

8. A medical statistical large-scale model intelligent agent, characterized in that, include: The Medical Statistics Knowledge Base module is used to construct a structured medical statistics knowledge base by performing semantic parsing and knowledge extraction on guidelines, regulations, and academic papers related to medical statistics. The cognitive interface module is used to obtain contextual information for medical statistical target tasks, and to provide the contextual information to the large language model after structured semantic transformation through a unified cognitive interface. The model invocation and execution module is used to invoke the large language model to generate an intent representation and structured task instructions for the target task based on the medical statistical knowledge base and the context information, and execute the structured task instructions to complete the target task and obtain the corresponding output. The closed-loop reflection and iteration module is used to analyze, verify and optimize the output through a closed-loop reflection and iteration mechanism to obtain optimized analysis conclusions, and to train the large language model based on the optimized analysis conclusions to improve its statistical analysis process planning capability.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A 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 of any one of claims 1 to 7.

Images