A medical insurance DRG grouping prediction method and system based on artificial intelligence technology

Abstract

The present invention discloses a medical insurance DRG grouping prediction method and system based on artificial intelligence technology, which relates to the field of medical artificial intelligence technology. The prediction method includes the following steps: S100, feature importance analysis: using a feature importance evaluation algorithm to comprehensively analyze and display the distribution of important features in different groups, and analyzing the correlation between important features and other features. The present invention introduces artificial intelligence technology, uses a feature importance evaluation algorithm for comprehensive analysis, and uses layer-by-layer correlation propagation technology to explain the decision-making process. It can conduct in-depth analysis of medical insurance data from multiple angles, which makes the prediction results more accurate, better understands the impact of different factors on DRG grouping, avoids one-sided decisions caused by a single factor, and provides a more scientific decision-making basis for medical insurance institutions and medical service providers. At the same time, the detailed explanation of the decision-making path and the case comparison analysis are helpful to discover potential rules and problems.

Description

Technical Field

[0001] The present invention relates to the field of medical artificial intelligence technology, and specifically to a medical insurance DRG grouping prediction method and system based on artificial intelligence technology. Background Art

[0002] With the continuous development of medical technology and the continuous growth of medical expenses, the reform of medical insurance payment methods has become an important challenge faced by medical systems in various countries. Medical insurance payment based on Diagnosis Related Groups (DRGs) is an important payment method reform. It groups diseases according to clinical characteristics, treatment methods and other factors, and determines a fixed payment standard for each group, thereby achieving refined management of medical insurance payments. In the medical insurance DRG grouping prediction, accurately allocating patients to appropriate groups is crucial for both medical insurance institutions and medical service providers. This is not only related to the rationality of medical insurance payments, but also affects the allocation of medical resources and the quality of medical services.

[0003] At present, the traditional medical insurance DRG grouping prediction and analysis method lacks in-depth explanation of the prediction results, making it difficult for medical insurance institutions and medical service providers to understand the basis of the prediction results. It is difficult to provide personalized explanations when faced with complex medical data and diverse patient conditions. Different patients have different disease characteristics, treatment histories, and socioeconomic backgrounds, and existing technologies can usually only give general prediction results and cannot provide detailed explanations for the specific circumstances of each patient, making it difficult for medical insurance institutions and medical service providers to understand the basis and logic of the prediction results, limiting the promotion and acceptance of the medical insurance DRG grouping prediction method in practical applications, and also increasing the uncertainty and risk in the decision-making process.

[0004] To sum up, the existing medical insurance DRG grouping prediction technology has many problems in the explanatory feedback mechanism, which not only affects the rationality of medical insurance payment and the transparency of decision-making, but also is not conducive to the improvement of medical service quality and the optimal allocation of resources. Therefore, there is an urgent need for a medical insurance DRG grouping prediction method and system based on artificial intelligence technology, which can introduce an effective explanatory feedback mechanism, improve the comprehensibility and transparency of the prediction results, and provide more powerful decision-making support for medical insurance institutions and medical service providers. Summary of the invention

[0005] The purpose of the present invention is to make up for the shortcomings of the existing technology and to provide a medical insurance DRG grouping prediction method and system based on artificial intelligence technology. It can more accurately analyze and predict the medical insurance DRG grouping results, process more complex data relationships, and identify more subtle patterns by applying artificial intelligence technology, thereby improving the prediction accuracy. This method not only provides grouping prediction results, but also can reasonably explain the prediction results, providing more scientific and transparent support for decision-making.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: On the one hand, a medical insurance DRG grouping prediction method based on artificial intelligence technology, the prediction method comprises the following steps:

[0007] S100, feature importance analysis: Use the feature importance evaluation algorithm to comprehensively analyze and display the distribution of important features in different groups, and analyze the correlation between important features and other features;

[0008] S200, Decision Path Explanation: Use layer-by-layer relevance propagation technology to explain the decision process, present the decision rules based on rule-based algorithms, and provide case illustrations of decision paths;

[0009] S300, case comparison analysis: expand the scope of case comparison and conduct statistical analysis, introduce time dimension analysis, and visualize the case comparison results;

[0010] S400, User Interaction Explanation: Provides an interactive interface between users and the explanation mechanism, allows users to customize feature importance weights, records user interaction history, and provides personalized explanations and suggestions;

[0011] S500, Uncertainty Analysis: Quantify the uncertainty of forecast results, analyze sources of uncertainty, and provide suggestions for coping with uncertainty.

[0012] Furthermore, the specific steps of the S100 feature importance analysis are:

[0013] S101, Application of feature importance evaluation algorithm: Use the random forest algorithm in the feature importance evaluation algorithm as the basis for training, and automatically calculate the importance score of the feature during the training process;

[0014] S102, Visualization of Important Features: For important features, not only the numerical value of their contribution to the prediction results is given, but also their distribution in different DRG groups is displayed through statistical analysis methods;

[0015] S103, correlation analysis: calculate the correlation coefficient matrix between the important features and other features, and find out other features that are highly correlated with the important features;

[0016] S104, Feature Importance Verification: Verify the stability and reliability of feature importance evaluation, so that users can understand the importance of different features on different data subsets;

[0017] S105, feature importance report generation: Generate a detailed feature importance report, including feature importance scores, comprehensive feature importance rankings, distribution of features in different groups, and impact descriptions of related features;

[0018] S106, feature importance adjustment: Provide a user interface to allow users to adjust the importance weight of features according to actual conditions. The system updates the prediction results in real time and regenerates an explanation report to show the impact of the adjustment.

[0019] Furthermore, the application of the S101 feature importance evaluation algorithm uses the random forest algorithm to comprehensively evaluate the importance of features and obtain a feature importance score based on gradient boosting. ,Right now ,in, Indicates Features Indicates the number of trees generated during model training, Indicated in Features in a tree Importance rating, Indicates The importance weight of each tree is based on the results of the comprehensive feature importance evaluation algorithm. The robustness of the score is improved by weighted average to obtain the comprehensive feature importance score. ,Right now ,in, represents the number of features of the feature importance evaluation algorithm, Indicates Algorithms for features Importance rating, Indicates The importance weight of the algorithm.

[0020] Furthermore, the visualization of the important features in S102 calculates the statistical information of the important features in different DRG groups and displays the distribution of the features in different groups, so as to help users understand the differences in the features in different groups, i.e., feature distribution statistical visualization. , Indicates DRG groups, Representation characteristics In group The mean value in Representation characteristics In group The median of Representation characteristics In group The range in, based on Visualization results S103 calculates the correlation coefficient matrix between features, finds other features that are highly correlated with important features, and explains the potential impact of these related features on the prediction results in the explanation report to establish a feature correlation matrix ,Right now ,in, and Respectively represent Features in samples and The value of and Respectively represent characteristics and The mean value of It indicates the number of samples in the DRG grouping. Based on the feature correlation matrix, one important feature is positively correlated with another feature, and the correlated feature has a greater impact on the DRG grouping. This correlation and the direction of influence are pointed out in the interpretation report.

[0021] Furthermore, the S100 feature importance analysis monitors the importance changes of important features in real time, and verifies the stability of feature importance. ,Right now , Indicates the number of verifications, Indicates Verification features Importance rating, Indicates The importance weight of the validation is displayed to enable users to understand the importance of different features on different data subsets, and generate a comprehensive feature importance report (CFIR), including feature importance scores, comprehensive feature importance rankings, distribution of features in different groups, and impact descriptions of related features. , and provides dynamic adjustment of feature importance FIDA, allowing users to adjust the importance weight of features according to actual conditions and update the prediction results in real time. , represents the feature weight adjusted by the user, It represents the adjustment coefficient, which is set according to the adjustment range of the user. After the user adjusts the feature weight, the system updates the prediction result in real time and regenerates the explanation report to show the impact of the adjustment.

[0022] Furthermore, the specific steps of the S200 decision path interpretation are:

[0023] S201, data processing: reading the classified DRG grouping data and case feature information from the database, and based on the feature importance analysis results, selecting features that have a significant impact on the DRG grouping as input features of the explanation path;

[0024] S202, building a decision tree model: using the selected feature data to train the decision tree model to ensure that the model can accurately reflect the DRG grouping logic;

[0025] S203, generating a decision path: receiving case data to be interpreted, inputting it into a trained decision tree model, using layer-by-layer correlation propagation technology to start from the root node of the decision tree, traversing downward along the decision path according to the characteristic values ​​of the case until reaching a leaf node, i.e., the corresponding DRG group, and recording the complete path from the root node to the leaf node, including the decision conditions and path direction at each node;

[0026] S204, Path explanation: Explain each node feature on the decision path to explain how the feature affects the decision-making process of DRG grouping. The decision path will show the entire decision logic from case input to DRG grouping, and rank the features on the decision path in order of importance, highlighting the key features that have the greatest impact on DRG grouping;

[0027] S205, user feedback mechanism for decision path explanation: providing an interactive interface between the user and the decision path explanation module, and further interpretation and analysis will be performed based on user feedback.

[0028] Furthermore, the specific steps of the S300 case comparative analysis are:

[0029] S301, extensive case collection and screening: collect a large amount of medical insurance case data from multiple data sources, including cases from different regions, different medical institutions, and different time periods, conduct preliminary screening of the collected cases, remove cases with incomplete and abnormal data, and classify the cases into different categories according to the DRG grouping standards;

[0030] S302, multi-dimensional feature extraction and analysis: for each case, extract multi-dimensional features including basic information of the patient, disease diagnosis, treatment process, and cost expenditure, and find out the correlation between different features and the correlation between these features and the DRG grouping results through correlation analysis;

[0031] S303, time series comparative analysis: arrange the cases in chronological order, and analyze the characteristic changes of the cases in different time periods and the changing trends of the DRG grouping results;

[0032] S304, cross-regional case comparison: select cases from different regions for comparative analysis to study the differences in case characteristics, treatment methods, and DRG grouping results between different regions;

[0033] S305, case comparison result visualization and report generation: Visualize the results of case comparison analysis, use charts to intuitively present the similarities and differences between different cases, and generate a detailed case comparison analysis report, including analysis methods, result summary, conclusions and suggestions.

[0034] Furthermore, the S301 extensive case collection and screening collects a large amount of clinical text data, structured data and time series data of medical insurance cases and patients from multiple data sources, specifically:

[0035] Clinical text data: including medical records, examination reports, and doctors’ diagnosis opinions;

[0036] Structured data: including ICD diagnosis codes, ICD-9-CM3 surgical operation codes, and drug use records;

[0037] Time series data: including the patient’s hospital stay, changes in the condition over time, and status records at each stage of the treatment process.

[0038] On the other hand, a medical insurance DRG grouping prediction system based on artificial intelligence technology includes the following components:

[0039] Feature importance analysis module: Use feature importance evaluation algorithm to comprehensively analyze and display the distribution of important features in different groups, and analyze the correlation between important features and other features;

[0040] Decision path explanation module: Use layer-by-layer correlation propagation technology to explain the decision process, present decision rules based on rule algorithms, and provide case explanations of decision paths;

[0041] Case comparison and analysis module: expand the scope of case comparison and conduct statistical analysis, introduce time dimension analysis, and visualize the case comparison results;

[0042] User interaction explanation module: provides an interactive interface between users and the explanation mechanism, allows users to customize feature importance weights, records user interaction history, and provides personalized explanations and suggestions;

[0043] Uncertainty Analysis Module: Quantify the uncertainty of prediction results, analyze the sources of uncertainty, and provide suggestions for dealing with uncertainty.

[0044] Furthermore, the user interactive interpretation module determines the main diagnosis category MDC based on the input main diagnosis using the ICD-10 coding rules, and preliminarily divides the cases into three categories: "surgical", "non-surgical" and "operational" according to the treatment methods. Under each category, through clinical similarity and statistical analysis, cases with the same main diagnosis and main operation are merged into core disease diagnosis-related groups ADRG. Taking into account other individual characteristics, comorbidities and complications of the cases, statistical analysis methods and expert experience are used to further subdivide ADRG into DRG groups.

[0045] Compared with the existing technology, the medical insurance DRG grouping prediction method and system based on artificial intelligence technology has the following beneficial effects:

[0046] 1. The present invention introduces artificial intelligence technology, uses feature importance evaluation algorithm for comprehensive analysis, and uses layer-by-layer correlation propagation technology to explain the decision-making process. It can conduct in-depth analysis of medical insurance data from multiple angles, which makes the prediction results more accurate and better understands the impact of different factors on DRG grouping, avoids one-sided decisions caused by a single factor, and provides medical insurance institutions and medical service providers with a more scientific decision-making basis. At the same time, the detailed explanation of the decision-making path and the case comparative analysis are helpful to discover potential patterns and problems, adjust the decision-making strategy in time, and improve the accuracy and reliability of the decision.

[0047] 2. This system greatly enhances the transparency of the decision-making process by providing a detailed explanatory feedback mechanism, including decision path explanations, case comparative analysis, and user interactive explanations. Users can ask questions through the interactive interface, and the system will automatically generate detailed explanations based on the questions, including the distribution of key features, correlation analysis between features, and case descriptions of decision paths. This interactivity not only improves the user experience, but also helps users adjust feature weights based on their own professional knowledge and observe changes in prediction results in real time, so as to better understand and utilize the prediction results of artificial intelligence and provide support for medical insurance decision-making.

[0048] Other advantages, objectives and features of the present invention will be set forth in part in the following description and, in part, will be apparent to those skilled in the art based on an examination of the following or may be taught from the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0050] Figure 1 This is an operational flow chart of a medical insurance DRG grouping prediction method based on artificial intelligence technology;

[0051] Figure 2 This is a flow chart of feature importance analysis of a medical insurance DRG grouping prediction method based on artificial intelligence technology. DETAILED DESCRIPTION

[0052] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0053] Embodiment 1

[0054] This embodiment provides a specific operation process of a medical insurance DRG grouping prediction method based on artificial intelligence technology, aiming to improve the accuracy and interpretability of medical insurance DRG grouping, and comprehensively analyze medical insurance case data through technical means such as feature importance analysis, decision path interpretation, case comparison analysis, user interaction interpretation and uncertainty analysis, to achieve refined DRG grouping prediction and provide a scientific basis for medical insurance payment and management.

[0055] First, enter the feature importance analysis stage (S100), and use the gradient boosting algorithm of the random forest algorithm to evaluate the feature importance. It not only calculates the contribution of each feature to the prediction result, but also displays the distribution of important features in different DRG groups through visualization, and displays the mean, median, range and other statistical information of important features in each group, to help users intuitively understand the feature differences. The system also calculates the correlation coefficient matrix between important features and other features, finds out other features that are highly correlated with important features, and explains the potential impact of these related features on the prediction results in the explanation report, monitors the changes in the importance of important features in real time, and automatically recalculates the feature importance when new data is added or model parameters are adjusted. The trend of the importance of important features over time is displayed in a visual way, and a sensitivity analysis of the feature importance is performed. The values ​​of important features are changed one by one to observe their impact on the prediction results. For each important feature, the change range of the prediction result when it changes within a certain range is calculated. The sensitivity of each important feature and the practical significance of this sensitivity to DRG group prediction are explained in detail in the explanation report.

[0056] Then, we enter the decision path interpretation stage (S200). In addition to using the layer-by-layer correlation propagation technology, we introduce the attention mechanism visualization method to analyze the attention weight distribution between different layers, and visualize the degree of attention paid to different parts when processing the input data. For medical image data, we can highlight the areas in the image that have a greater impact on DRG grouping prediction, helping users understand how the model makes decisions based on image features. For the rule-based algorithm of the decision tree, we not only display the decision rules in the form of a flowchart, but also add detailed annotations to each decision node, including the meaning of the feature corresponding to the node, the value range, and the basis for the decision. At the same time, Provide case analysis to show how the model performs group prediction along the decision path under different feature value combinations. Perform decision path analysis on the same group of cases to compare the decision differences of different models when processing the same input data, including feature selection and decision logic. Track changes in the decision path in real time. When the input data changes slightly, observe how the model's decision path is adjusted. For example, when a patient's examination indicator changes, the system can show how the model updates the decision path based on the new data and gives the corresponding DRG group prediction results. At the same time, record the change history of the decision path so that users can review and analyze the dynamic decision-making process of the model.

[0057] Next, we enter the case comparison and analysis stage (S300), select representative case samples from the DRG grouping data set, covering cases of different disease types, treatment methods and complications, extract key information from the selected cases, including the main diagnosis, surgical operation, complications and comorbidities, hospitalization time, cost, etc., to ensure the integrity and accuracy of the data, and preliminarily group the selected cases into corresponding DRG groups according to the CHS-DRG grouping rules, check whether the grouping of each case is consistent with its clinical characteristics and resource consumption, and make manual adjustments or reviews if necessary to ensure that the grouping is correct, arrange the cases in chronological order, and analyze the changes in the characteristics of the cases in different time periods and the changing trends of the DRG grouping results. It intuitively displays the changes of various characteristics and DRG groupings over time, analyzes the factors that may affect these changes, such as medical technology advances and medical insurance policy adjustments, selects cases from different regions for comparative analysis, studies the differences between different regions in case characteristics, treatment methods, DRG grouping results, etc., and analyzes the reasons for these differences, such as regional economic development level, distribution of medical resources, medical habits, etc. Through cross-regional comparison, it can provide a reference for the formulation and adjustment of medical insurance policies and promote the balanced development of medical resources among regions. The results of case comparison analysis are visualized to show the similarities and differences between different cases, and generate a detailed case comparison analysis report, including analysis methods, result summary, conclusions and suggestions.

[0058] Subsequently, the user interaction explanation stage (S400) is entered. An intuitive and easy-to-use user interaction interface is designed to display the prediction results of the medical insurance DRG grouping and related explanatory information. The interface should include clear charts, text descriptions and operation buttons to facilitate users to view and understand the prediction results, display the probability distribution of different DRG groups, and use text descriptions to explain the meaning and characteristics of each group. A question input box is provided in which users can enter questions about the prediction results. The system should be able to parse the user's question, extract key information, and determine the type of question. For example, if the user asks about the impact of a certain feature on the prediction result, the system should identify the two key elements of the feature and the prediction result in the question, and determine the question type as feature impact analysis. According to the user's question type, the system automatically generates For feature impact analysis, the system can display the distribution of the feature in different DRG groups, its correlation with other features, and its specific contribution to the prediction results. Based on the user's questions and prediction results, the system provides users with personalized suggestions. For example, if the user has questions about the medical insurance reimbursement of a treatment plan, the system can provide suggestions on treatment plan selection, cost control, etc. based on the patient's specific situation and DRG group prediction results. The suggestions should be targeted and practical, and help users make better decisions. The system records the interaction history between the user and the system, including question input, explanation content viewing, and suggestion adoption. Through the analysis of the interaction history, the system can understand the user's needs and concerns, and further optimize the explanation content and suggestion provision.

[0059] Finally, we enter the uncertainty analysis stage (S500). By performing multiple random sampling on the model to obtain different prediction results, we combine prior knowledge and data to obtain the posterior distribution of the prediction results to quantify uncertainty and determine appropriate uncertainty metrics. We analyze aspects such as data, model, and environment to determine the main sources of uncertainty. In terms of data, we consider the impact of factors such as data quality, quantity, noise, and missing values ​​on uncertainty. For the model, we analyze factors such as model complexity, parameter uncertainty, and assumption rationality. In terms of the environment, we consider the impact of external factors such as policy changes and medical technology advances on uncertainty. Through sensitivity analysis and other methods, the contribution of each factor to uncertainty is determined, and the propagation process of uncertainty in the model is analyzed. According to the results of uncertainty analysis, corresponding response strategies are formulated. For data uncertainty, data cleaning, data enhancement, multi-source data fusion and other methods can be used to improve data quality and quantity. For model uncertainty, model optimization, parameter adjustment, model integration and other operations can be performed to improve the accuracy and stability of the model. In terms of environmental uncertainty, a dynamic monitoring mechanism is established to adjust the model and strategy in time to adapt to external changes. The results of uncertainty analysis are visualized and a detailed uncertainty analysis report is generated, including analysis methods, result summary, response strategy recommendations and other contents. The report is provided to medical insurance institutions and relevant departments to provide a reference for decision-making.

[0060] In summary, this embodiment specifically demonstrates the effectiveness and practicality of the medical insurance DRG grouping prediction method based on artificial intelligence technology, can provide detailed explanatory feedback, help users understand the basis of the prediction results, improve the transparency and credibility of decision-making, and reduce decision-making risks through uncertainty analysis and response suggestions, providing strong support for the development of the medical insurance field.

[0061] Embodiment 2

[0062] This embodiment describes in detail the specific implementation process of a medical insurance DRG grouping prediction system based on artificial intelligence technology. By integrating the grouping logic of CHS-DRG and artificial intelligence technology, the grouping process is automated and intelligent, the accuracy and fairness of the grouping results are improved, and the rationality and efficiency of medical insurance payment are ensured by using the steps of feature importance evaluation, decision path transparency, and extensive case comparison and analysis.

[0063] In the specific implementation, medical insurance case data is collected from multiple data sources, including cases from different regions, medical institutions and time periods. At the same time, hospital medical record data, medical insurance platform settlement data, hospital disease diagnosis and surgical operation coding library and other information are collected, and the data is cleaned and standardized to ensure the integrity, rationality and standardization of the data. For example, check whether the patient's medical record homepage and cost details are complete, determine whether the diagnosis and cost are consistent, etc., convert the data according to the national medical insurance standard coding to prepare for subsequent grouping predictions.

[0064] The feature importance analysis module uses feature importance evaluation algorithms such as the random forest algorithm to comprehensively analyze and display the distribution of important features in different groups to obtain a comprehensive feature importance score. ,Right now By calculating the statistical information of important features in different DRG groups and displaying the distribution of features in different groups, it helps users understand the differences in features in different groups, i.e., feature distribution statistics visualization , calculate the correlation coefficient matrix between important features and other features ,Right now , find highly correlated features, and visualize the distribution of important features in different DRG groups to help users understand the differences in features in different groups. For example, for features such as disease diagnosis and treatment costs, analyze their statistical information such as mean, median and range in different DRG groups. At the same time, verify the stability of feature importance Verify the stability of feature importance evaluation, i.e. , verify the stability of feature importance evaluation and generate a comprehensive feature importance report ,Right now , allowing users to adjust feature importance weights according to actual conditions and update prediction results in real time.

[0065] Then, the decision path interpretation module reads the classified DRG grouping data and case feature information from the database. Based on the feature importance analysis results, the features that have a significant impact on the DRG grouping are selected as the input features of the interpretation path. The decision tree model is trained using these feature data to ensure that the model can accurately reflect the DRG grouping logic. When receiving the case data to be interpreted, the layer-by-layer correlation propagation technology is used, starting from the root node of the decision tree, and traversing downward along the decision path according to the feature values ​​of the case until reaching the leaf node, that is, the corresponding DRG grouping. Each node feature on the decision path is interpreted to explain its decision-making process for the DRG grouping, and the features are ranked by importance, highlighting the key features. In addition, a user feedback mechanism is provided to conduct further interpretation and analysis based on user feedback, and to collect a wide range of Collect medical insurance case data from different regions, medical institutions and time periods, conduct preliminary screening of the collected cases, remove cases with incomplete and abnormal data, and classify the cases into different categories according to the DRG grouping standards. For each case, extract multi-dimensional features such as patient basic information, disease diagnosis, treatment process, and cost expenditure. Through correlation analysis, find out the correlation between different features and the correlation between these features and the DRG grouping results. Arrange the cases in chronological order, analyze the changes in case characteristics and the changing trends of DRG grouping results in different time periods, select cases from different regions for comparative analysis, and study the differences in case characteristics, treatment methods, and DRG grouping results between different regions. Visually display the results of case comparative analysis and generate a detailed case comparative analysis report.

[0066] Finally, the user interactive explanation module provides an interactive interface between the user and the explanation mechanism. The user can use this interface to ask questions about the prediction results. The system determines the main diagnosis category MDC based on the input main diagnosis using the ICD-10 coding rules, and preliminarily divides the cases into three categories according to the treatment methods: "surgery", "non-surgical" and "operation". Under each category, through clinical similarity and statistical analysis, the cases with the same main diagnosis and main operation are merged into core disease diagnosis related groups ADRG. Taking into account other individual characteristics, comorbidities and complications of the cases, ADRG is further subdivided into DRG groups. The system will automatically generate detailed explanations based on the user's questions, including the distribution of key features, the relationship between features, and the relationship between them. Case studies of correlation analysis and decision paths. Users can also customize the weights of feature importance, observe changes in prediction results in real time, quantify the uncertainty of prediction results, estimate the uncertainty range of prediction results using probability distribution or confidence interval methods, analyze the source of uncertainty, and determine which factors have the greatest impact on the uncertainty of prediction results through sensitivity analysis, such as measurement errors of certain features, missing data, or complexity of the model. Based on the results of uncertainty analysis, provide medical insurance agencies and medical service providers with suggestions for dealing with uncertainty, such as adopting a more conservative medical insurance payment strategy or further collecting more medical data to improve the accuracy of predictions when uncertainty is high.

[0067] In summary, this embodiment demonstrates the comprehensive implementation process of a medical insurance DRG grouping prediction system based on artificial intelligence technology. The system not only provides accurate prediction results, but also enhances the transparency of decision-making and user trust through a detailed explanatory feedback mechanism, providing strong support for medical insurance institutions and medical service providers, and helping to achieve refined management of medical insurance payments and improve the quality of medical services.

[0068] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A medical insurance DRG grouping prediction method based on artificial intelligence technology, characterized in that: The prediction method includes the following steps: S100, feature importance analysis: Use the feature importance evaluation algorithm to comprehensively analyze and display the distribution of important features in different groups, and analyze the correlation between important features and other features. The specific steps of the S100 feature importance analysis are: S101, Application of feature importance evaluation algorithm: Use the random forest algorithm in the feature importance evaluation algorithm as the basis for training, and automatically calculate the importance score of the feature during the training process; S102, Visualization of Important Features: For important features, not only the numerical value of their contribution to the prediction results is given, but also their distribution in different DRG groups is displayed through statistical analysis methods; S103, correlation analysis: calculate the correlation coefficient matrix between the important features and other features, and find out other features that are highly correlated with the important features; S104, Feature Importance Verification: Verify the stability and reliability of feature importance evaluation, so that users can understand the importance of different features on different data subsets; S105, feature importance report generation: Generate a detailed feature importance report, including feature importance scores, comprehensive feature importance rankings, distribution of features in different groups, and impact descriptions of related features; S106, feature importance adjustment: Provide a user interface to allow users to adjust the importance weight of features according to actual conditions. The system updates the prediction results in real time and regenerates an explanation report to show the impact after the adjustment. S200, decision path explanation: using layer-by-layer correlation propagation technology to explain the decision process, presenting the decision rules based on the rule algorithm, and providing case explanations of the decision path. The specific steps of the S200 decision path explanation are: S201, data processing: read the classified DRG grouping data and case feature information from the database, and based on the feature importance analysis results, select the features that have a significant impact on the DRG grouping as the input features of the explanation path; S202, building a decision tree model: using the selected feature data to train the decision tree model to ensure that the model can accurately reflect the DRG grouping logic; S203, generating a decision path: receiving case data to be interpreted, inputting it into a trained decision tree model, using layer-by-layer correlation propagation technology to start from the root node of the decision tree, traversing downward along the decision path according to the characteristic values ​​of the case until reaching a leaf node, i.e., the corresponding DRG group, and recording the complete path from the root node to the leaf node, including the decision conditions and path direction at each node; S204, Path explanation: Explain each node feature on the decision path to explain how the feature affects the decision-making process of DRG grouping. The decision path will show the entire decision logic from case input to DRG grouping, and rank the features on the decision path in order of importance, highlighting the key features that have the greatest impact on DRG grouping; S205, user feedback mechanism for decision path explanation: providing an interactive interface between the user and the decision path explanation module, and further explanation and analysis will be performed based on user feedback; S300, case comparison analysis: expand the case comparison scope and perform statistical analysis, introduce time dimension analysis, and visualize the case comparison results. The specific steps of the S300 case comparison analysis are: S301, extensive case collection and screening: collect a large amount of medical insurance case data from multiple data sources, including cases from different regions, different medical institutions, and different time periods, conduct preliminary screening of the collected cases, remove cases with incomplete and abnormal data, and classify the cases into different categories according to the DRG grouping standards; S302, multi-dimensional feature extraction and analysis: for each case, extract multi-dimensional features including basic information of the patient, disease diagnosis, treatment process, and cost expenditure, and find out the correlation between different features and the correlation between these features and the DRG grouping results through correlation analysis; S303, time series comparative analysis: arrange the cases in chronological order, and analyze the characteristic changes of the cases in different time periods and the changing trends of the DRG grouping results; S304, cross-regional case comparison: select cases from different regions for comparative analysis to study the differences in case characteristics, treatment methods, and DRG grouping results between different regions; S305, case comparison result visualization and report generation: Visualize the results of case comparison analysis, use charts to intuitively present the similarities and differences between different cases, and generate a detailed case comparison analysis report, including analysis methods, result summary, conclusions and suggestions; S400, User Interaction Explanation: Provides an interactive interface between users and the explanation mechanism, allows users to customize feature importance weights, records user interaction history, and provides personalized explanations and suggestions; S500, Uncertainty Analysis: Quantify the uncertainty of forecast results, analyze sources of uncertainty, and provide suggestions for coping with uncertainty.

2. According to the method for predicting medical insurance DRG grouping based on artificial intelligence technology in claim 1, it is characterized in that: The application of the S101 feature importance evaluation algorithm uses the random forest algorithm to comprehensively evaluate the importance of features and obtain the feature importance score GBFI based on gradient boosting, that is, Among them, f i represents the i-th feature, T represents the number of trees generated during model training, Imp t( f i) Indicates that feature f in the tth tree i The importance score of t Represents the importance weight of the t-th tree. Based on the results of the comprehensive feature importance evaluation algorithm, the robustness of the score is improved by weighted averaging to obtain the comprehensive feature importance score CFS, that is, Among them, J represents the number of features of the feature importance evaluation algorithm, Imp j ( f i ) Indicates the jth algorithm for feature f i The importance score, α j Represents the importance weight of the j-th algorithm.

3. The medical insurance DRG grouping prediction method based on artificial intelligence technology according to claim 1 is characterized in that: The visualization of important features in S102 calculates the statistical information of important features in different DRG groups and displays the distribution of features in different groups, so as to help users understand the differences between features in different groups, i.e., feature distribution statistical visualization. g k represents the kth DRG group, Represents feature f i In group g k The mean value in Represents feature f i In group g k The median of Represents feature f i In group g k In the range, based on the FDSV visualization results, S103 calculates the correlation coefficient matrix between features, finds other features that are highly correlated with important features, and explains the potential impact of these related features on the prediction results in the explanation report, and establishes the feature correlation matrix FCM, that is, Among them, f i,n and f j,n Respectively represent the feature f in the nth sample i and f j The value of and Respectively represent the features f i and f j The mean of N represents the number of samples in the DRG grouping. Based on the feature correlation matrix, an important feature is positively correlated with another feature, and the correlated feature has a greater impact on the DRG grouping. This association and the direction of impact are pointed out in the interpretation report.

4. The method for predicting medical insurance DRG grouping based on artificial intelligence technology according to claim 1 is characterized in that: The S100 feature importance analysis monitors the importance changes of important features in real time and verifies FISV through feature importance stability, i.e. K represents the number of verifications, Imp test,k( f i) Indicates feature f in the kth verification i The importance score of k It represents the importance weight of the kth validation. The validation results are displayed so that users can understand the importance of different features on different data subsets and generate a comprehensive feature importance report CFIR, including feature importance scores, comprehensive feature importance rankings, distribution of features in different groups, and impact descriptions of related features, i.e. It also provides dynamic feature importance adjustment FIDA, allowing users to adjust the importance weight of features according to actual conditions and update the prediction results in real time. ( f i ,w i) =CFS ( f i) +δ i ·w i , w i represents the feature weight adjusted by the user, δ i It represents the adjustment coefficient, which is set according to the adjustment range of the user. After the user adjusts the feature weight, the system updates the prediction result in real time and regenerates the explanation report to show the impact of the adjustment.

5. The method for predicting medical insurance DRG grouping based on artificial intelligence technology according to claim 1 is characterized in that: The S301 extensive case collection and screening collects a large amount of clinical text data, structured data and time series data of medical insurance cases and patients from multiple data sources, specifically: Clinical text data: including medical records, examination reports, and doctors’ diagnosis opinions; Structured data: including ICD diagnosis codes, ICD-9-CM3 surgical operation codes, and drug use records; Time series data: including the patient’s hospital stay, changes in the condition over time, and status records at each stage of the treatment process.

6. A medical insurance DRG grouping prediction system based on artificial intelligence technology, characterized in that: The group prediction system includes the following components: Feature importance analysis module: Use the feature importance evaluation algorithm to comprehensively analyze and display the distribution of important features in different groups, and analyze the correlation between important features and other features. The specific steps of the feature importance analysis module are: Application of feature importance evaluation algorithm: Use the random forest algorithm in the feature importance evaluation algorithm as the basis for training, and automatically calculate the importance score of the feature during the training process; Visual display of important features: For important features, not only the numerical value of their contribution to the prediction results is given, but also their distribution in different DRG groups is displayed through statistical analysis methods; Correlation analysis: Calculate the correlation coefficient matrix between important features and other features, and find other features that are highly correlated with important features; Feature importance verification: Verify the stability and reliability of feature importance evaluation, so that users can understand the importance of different features on different data subsets; Feature importance report generation: Generate a detailed feature importance report, including feature importance scores, comprehensive feature importance rankings, distribution of features in different groups, and impact descriptions of related features; Feature importance adjustment: Provides a user interface that allows users to adjust the importance weight of features according to actual conditions. The system updates the prediction results in real time and regenerates an explanation report to show the impact of the adjustment; Decision path explanation module: Use layer-by-layer correlation propagation technology to explain the decision process, present decision rules based on rule algorithms, and provide case descriptions of decision paths. The specific steps of the decision path explanation module are as follows: Data processing: Read the classified DRG grouping data and case feature information from the database, and based on the feature importance analysis results, select the features that have a significant impact on the DRG grouping as the input features of the explanation path; Construct a decision tree model: Use the selected feature data to train the decision tree model to ensure that the model can accurately reflect the DRG grouping logic; Generate decision paths: Receive case data to be interpreted, input it into the trained decision tree model, use layer-by-layer correlation propagation technology to start from the root node of the decision tree, traverse downward along the decision path according to the characteristic values ​​of the case until reaching the leaf node, that is, the corresponding DRG group, and record the complete path from the root node to the leaf node, including the decision conditions and path direction at each node; Path explanation: Explain each node feature on the decision path to explain how the feature affects the decision-making process of DRG grouping. The decision path will show the entire decision logic from case input to DRG grouping, and rank the features on the decision path in order of importance, highlighting the key features that have the greatest impact on DRG grouping; User feedback mechanism for decision path explanation: Provides an interactive interface between users and the decision path explanation module, and further explanation and analysis will be performed based on user feedback; Case comparison and analysis module: expand the scope of case comparison and conduct statistical analysis, introduce time dimension analysis, and visualize the case comparison results. The specific steps of the case comparison and analysis module are as follows: Extensive case collection and screening: Collect a large amount of medical insurance case data from multiple data sources, including cases from different regions, different medical institutions, and different time periods. Perform preliminary screening on the collected cases, remove cases with incomplete and abnormal data, and classify the cases into different categories according to the DRG grouping standards; Multi-dimensional feature extraction and analysis: For each case, extract multi-dimensional features including basic patient information, disease diagnosis, treatment process, and cost expenditure. Through correlation analysis, find out the correlation between different features, as well as the correlation between these features and DRG grouping results; Time series comparative analysis: Arrange the cases in chronological order and analyze the characteristic changes of the cases in different time periods and the changing trends of the DRG grouping results; Cross-regional case comparison: Select cases from different regions for comparative analysis to study the differences in case characteristics, treatment methods, and DRG grouping results between different regions; Visualization of case comparison results and report generation: Visualize the results of case comparison analysis, use charts to intuitively present the similarities and differences between different cases, and generate a detailed case comparison analysis report, including analysis methods, result summary, conclusions and suggestions; User interaction explanation module: provides an interactive interface between users and the explanation mechanism, allows users to customize feature importance weights, records user interaction history, and provides personalized explanations and suggestions; Uncertainty Analysis Module: Quantify the uncertainty of prediction results, analyze the sources of uncertainty, and provide suggestions for dealing with uncertainty.

7. The medical insurance DRG grouping prediction system based on artificial intelligence technology according to claim 6 is characterized in that: The user interactive interpretation module determines the major diagnosis category MDC based on the input main diagnosis using the ICD-10 coding rules, and preliminarily divides the cases into three categories: "surgery", "non-surgical" and "operation" according to the treatment method. Under each category, through clinical similarity and statistical analysis, cases with the same main diagnosis and main operation are merged into core disease diagnosis-related groups ADRG. Taking into account other individual characteristics, comorbidities and complications of the cases, ADRG is further subdivided into DRG groups using statistical analysis methods and expert experience.

Images