Automatic identification and intelligent error correction method and system for DRG grouping error

Abstract

The invention relates to the field of disease diagnosis related grouping, and discloses an automatic identification and intelligent error correction method and system for DRG grouping errors. According to the method, based on case original data, the total hospitalization cost, the drug cost, the consumable cost and the number of hospitalization days are standardized, a consumption index is constructed, an initial DRG score is generated, suspected grouped error cases are identified through distribution fitting and probability calculation, and an optimization recommendation result is generated by using case home page information and a hybrid error correction engine. And an interactive confirmation interface is provided to realize intelligent error correction, finally, the data adopted by the user is re-scored, and a recommendation strategy is dynamically optimized through a self-learning module. The system is composed of a data processing and scoring module, an anomaly detection and error correction recommendation module and a score updating and self-learning module, efficient recognition and intelligent error correction of DRG grouping errors are achieved, and the accuracy and efficiency of medical management are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of disease diagnosis related grouping, which can identify errors in grouping and perform intelligent error correction, and particularly relates to a DRG grouping error automatic identification and intelligent correction method and system. BACKGROUND

[0002] Disease diagnosis related grouping (DRG) is increasingly widely used in medical systems. DRG divides patients into diagnostic groups for management according to factors such as age, disease diagnosis, comorbidity, complications, treatment method, severity of illness, and outcome and resource consumption. However, since the DRG grouping rules are not publicly disclosed by the state, the DRG systems used by various hospitals are often provided by third-party software companies, and the quality of the grouping lacks a control condition. In the actual application of clinical practice, there is currently a lack of effective checking and verifying schemes for the correctness of DRG grouping.

[0003] Existing DRG grouping error correction relies on manual review or simple rule verification, which has the problems of low efficiency, dependence on experience, and difficulty in covering complex cases, and cannot achieve rapid and intelligent identification and correction of massive cases. At the same time, the existing methods lack a dynamic learning and optimization mechanism for the correction results, and it is difficult to improve accuracy and reliability as data accumulates. Therefore, there is an urgent need for a system method that can automatically identify DRG grouping errors, intelligently recommend corrections, and has self-learning ability to improve the accuracy of DRG grouping and the efficiency of medical management. SUMMARY

[0004] In order to achieve the above-mentioned application purpose, the present application provides the following technical scheme: a DRG grouping error automatic identification and intelligent correction method, comprising the following steps:

[0005] Step S1, based on a case original data set D raw , the total hospitalization cost C tot of the case, the drug cost C med , the consumable cost C mat , the length of stay L day is standardized, the consumption index vector V ind corresponding to each index is calculated, and the weight matrix W ind is normalized and compared with the reference group mean data M ref to obtain the initial DRG score S0.

[0006] Step S2, distribution fitting and interval division are performed on the DRG score S0, and the quartile model and the threshold interval [Q low , Q high ] set by the user are used to calculate the case abnormal probability P abn , and when the case abnormal probability P abnWhen the number exceeds the set range, it is marked as a suspected case of grouping error;

[0007] Step S3: For suspected incorrectly grouped cases, retrieve the information from the medical record cover page F. rec This includes diagnostic information, surgical procedures, and the generation of a ranking feature matrix, which is then processed by a hybrid error correction engine E based on rule-based reasoning and a deep recommendation model. corr Generate optimized recommendation results R opt It also provides an interactive confirmation interface for users to decide whether to adopt the solution.

[0008] Step S4: When the user accepts the recommendation result R opt Afterwards, the updated case data D new Input the scoring model from step S1 again to obtain the corrected score S1, and compare S0 with S1. If |S1-M ref |<|S0-M ref If the error is corrected, the error correction is considered successful; otherwise, it is marked as a failure. The failed sample and the error correction path are stored in the self-learning module for optimization of subsequent recommendation strategies.

[0009] Preferably, the specific process of step S1 is as follows:

[0010] S1(1), regarding the original case dataset D raw ={C tot C med C mat ,L day Preprocessing and standardization are performed:

[0011]

[0012] Among them, X i The original indicator values ​​for the cases, μ ref,i σ ref,i The mean and standard deviation of the same indicator in the reference group are respectively, and the total hospitalization cost C is... tot Sourced from hospital billing system, drug cost C med Sourced from the hospital's drug management system and consumable costs C mat Sourced from the hospital supplies management system and the number of days of hospital stay (L). day It is derived from the electronic medical record system and calculated based on the difference between the admission time and the discharge time;

[0013] S1(2), Construct the consumption exponent vector V ind :

[0014]

[0015] Among them, M tot M med Mmat , M day are the reference group data of total hospitalization expenses, drug expenses, consumable expenses, and hospitalization days, respectively, M ref = {M tot , M med , M mat , M day};

[0016] and the initial DRG score S0 is obtained by weighted superposition:

[0017]

[0018] wherein the weight matrix W ind is dynamically updated according to the big data regression learning result, I j , I tot , I med , I mat , and I day .

[0019] Preferably, the specific process of step S2 is as follows:

[0020] S2(1), distribution modeling is performed on the DRG score set of all cases, and a skew normal distribution function is used for fitting:

[0021]

[0022] wherein φ(·) is a standard normal density function, Φ(·) is a standard normal distribution function, ζ is a location parameter, ω is a scale parameter, and α is a skewness parameter, and ζ, ω, and α are derived from statistical modeling and fitting of the score set of step S1;

[0023] S2(2), according to the fitted distribution, the abnormal probability of each case is calculated:

[0024] P abn = 1-F(S0);

[0025] wherein F(S0) is a cumulative distribution value of the distribution function, derived from the distribution fitting result, and if S0 does not belong to [Q low , Q high ] and P abn <0.9, then it is marked as a suspected grouping error case F err .

[0026] Preferably, the specific process of step S3 is as follows:

[0027] For abnormal cases, the diagnosis information D rec and the surgery information O dia are extracted from the front page field information F oprand generates a ranking feature matrix M:

[0028] M sort = {Rank(D dia ), Rank(O opr )};

[0029] The ranking feature matrix M sort is input into a hybrid error correction engine E corr , which consists of two parts:

[0030] A rule reasoning engine E rule based on clinical pathways and medical insurance audit rules, which is used to check the logical consistency of the diagnosis and operation fields and generate preliminary correction suggestions;

[0031] A deep recommendation model E model based on graph neural networks, which is used to learn the implicit relationship between diagnosis and surgical operations and output the optimal prediction probability;

[0032] The recommendation function of the hybrid error correction engine E corr is:

[0033]

[0034] The final recommendation result R opt is generated and provided to the user through an interactive interface for selection, thereby realizing intelligent correction of case grouping errors.

[0035] Preferably, the specific process of step S4 is:

[0036] S4(1), when the user confirms the recommendation result R corr output by the hybrid error correction engine E opt in the interactive interface, the system generates new case data D new according to the following update rules:

[0037] D new = (F rec \F err ) U R opt ;

[0038] Where F rec represents the front page information F rec , including diagnosis information D dia and operation information O opr , and F err is the suspected case of grouping error determined in step S2 abnormality detection;

[0039] S4(2), the updated case data D newInputting the data into the scoring model yields the corrected score S1:

[0040]

[0041] Wherein, the weight matrix W ind Dynamically updated based on big data regression learning results. C′ tot C′ med C′ mat L′ day To achieve this through the hybrid error correction engine E corr For C tot C med C mat L day The data was re-analyzed after correction;

[0042] S4(3) Compare the difference in scores before and after correction. If |S1-M ref |<|S0-M ref If the result is |, then the error correction is considered successful; otherwise, it is considered a failure, and the system will discard the failed sample (D). new ,R opt S1) Store the data in the self-learning module and use an incremental learning algorithm to optimize the recommendation strategy parameters:

[0043]

[0044] Where θ represents the model parameters and η represents the learning rate. The loss function is used to achieve dynamic evolution of the system's error correction capability.

[0045] This invention also provides an intelligent error correction system for automatic identification of DRG grouping errors, comprising the following modules:

[0046] The data processing and scoring module is used to preprocess and standardize the original case data, calculate the consumption index of various costs and hospitalization days, and perform normalization calculation based on the weight matrix and the mean of the reference group to output the initial DRG score for subsequent anomaly detection.

[0047] The anomaly detection and error correction recommendation module is used to perform distribution modeling and probability calculation based on the initial DRG score, identify suspected misgrouped cases, and generate a feature matrix by calling the information on the medical record front page. The matrix is ​​then input into the hybrid error correction engine to generate intelligent error correction recommendation results, while providing an interactive confirmation interface for users to adopt the recommendations.

[0048] The score updating and self-learning module is used for receiving the error correction result adopted by the user, re-inputting the updated case data into the scoring model, calculating the corrected score, and storing the error correction samples of success and failure into the self-learning module, dynamically optimizing the recommendation strategy through the incremental learning algorithm, and realizing continuous improvement of the intelligent error correction capability of the system.

[0049] The data processing and scoring module comprises:

[0050] The data preprocessing unit is used for cleaning, standardizing and missing value processing of the original case data, and guarantees the completeness and calculability of the input data.

[0051] The consumption index calculation unit is used for normalizing each expense and the hospitalization days with the mean value of the reference group, forming a consumption index vector, and generating the initial DRG score through weighted superposition according to the weight matrix.

[0052] The score output unit is used for transmitting the initial DRG score to the abnormality detection and error correction recommendation module, and providing basic data for subsequent abnormality probability calculation.

[0053] The abnormality detection and error correction recommendation module comprises:

[0054] The abnormality identification unit is used for distribution fitting, probability calculation and interval division of the initial DRG score, and labels suspected grouping error cases.

[0055] The feature construction unit is used for extracting diagnosis and operation data from the case history information, and generating a sorting feature matrix for the error correction model.

[0056] The mixed error correction engine unit is composed of a rule reasoning engine and a deep recommendation model, and generates intelligent error correction recommendation results by combining clinical rules and data-driven algorithms, and provides the user with confirmation and adoption through an interactive interface.

[0057] The score updating and self-learning module comprises:

[0058] The data updating unit is used for fusing the error correction result confirmed and adopted by the user with the original case data, and generating an updated case data set.

[0059] The corrected score unit is used for re-inputting the updated case data into the scoring model to calculate the corrected score, and comparing the original score to judge whether the error correction is successful or not.

[0060] The self-learning unit is used for storing the error correction samples of success and failure, dynamically adjusting the recommendation strategy parameters in combination with the incremental learning algorithm, and realizing continuous optimization and iteration of the intelligent error correction capability of the system in actual application.

[0061] Compared with the prior art, the present application has the following beneficial effects:

[0062] Improve DRG grouping accuracy: by standardizing the total hospitalization cost, drug cost, consumable cost and hospitalization days of the case and calculating the consumption index, and combining distribution fitting and abnormal probability analysis, the application can quickly and accurately identify suspected grouping error cases, significantly reduce the dependence on manual review, and improve the overall accuracy of DRG grouping.

[0063] Intelligent error correction and recommendation: The application combines rule-based reasoning with deep recommendation models through a hybrid error correction engine, making full use of the implicit relationships between the case information and diagnosis and surgical operations to achieve intelligent error correction and optimized recommendations for suspected error cases. An interactive confirmation interface is also provided to allow users to independently adopt or adjust the recommended results, improving error correction efficiency and reliability.

[0064] Dynamic self-learning capability: By storing successful and failed samples after error correction in the self-learning module and continuously optimizing recommendation strategy parameters using incremental learning algorithms, the application can gradually improve error correction accuracy as data volume increases, enabling continuous intelligent iteration and self-improvement of the system to meet large-scale medical data processing needs. BRIEF DESCRIPTION OF DRAWINGS

[0065] Figure 1 Method step flowchart provided for the present application;

[0066] Figure 2 System module diagram provided for the present application. DETAILED DESCRIPTION

[0067] In order to better understand the technical scheme of the present application, the technical scheme in the embodiments of the present application will be described in detail below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0068] Reference Figure 1 The embodiments of the present application provide an automatic identification and intelligent correction method for DRG grouping errors, comprising the following steps:

[0069] Step S1, based on the original data set D of the case raw The total hospitalization cost C of the case tot The drug cost C med The consumable cost C mat The hospitalization days L day are standardized, the consumption index vector V corresponding to each index is calculated ind , and the weight matrix Wind with reference group mean data M ref Normalization comparison is performed to obtain initial DRG score S0.

[0070] In step S1, the system first processes the case original data set, which includes total hospitalization expenses, drug expenses, consumable expenses and hospitalization days, respectively from the hospital billing system, drug management system, material management system and electronic medical record system. The system pre-processes each index, including missing value filling, outlier filtering and standardization operation. The standardization formula is where X i is the case original index value, μ ref,i and σ ref,i are the corresponding index mean and standard deviation of the reference group. Subsequently, the system constructs the consumption index vector where each I j is calculated by the ratio of the case index to the reference group mean, and the initial DRG score S0is obtained by weighted superposition of each index through the weight matrix W ind . In this process, the weight matrix W ind is dynamically adjusted according to the historical big data regression learning results, so that the scoring model can adapt to the cost characteristics of different hospitals and different diseases.

[0071] Step S2, distribution fitting and interval division are performed on the DRG score S0. The quartile model and the user-set threshold interval [Q low , Q high ] are used to calculate the case abnormal probability P abn . When the case abnormal probability P abn exceeds the set range, it is marked as a suspected grouping error case.

[0072] In step S2, the system performs distribution modeling on the initial DRG score S0set generated in step S1, and uses the skew normal distribution function for fitting, where φ(·) is the standard normal density function, Φ(·) is the standard normal distribution function, ζ, ω, α are parameters obtained by statistics of the score set in step S1. The system calculates the abnormal probability P abn of each case using the fitting result, where F(S0) is the cumulative distribution function value. Subsequently, the system screens the abnormal probability according to the user-set threshold interval [Q low , Q high ]. If S0is not in the threshold interval and P abn <0.9, it is determined as a suspected grouping error case.

[0073] Step S3, for the suspected grouping error case, the front page information F rec, including diagnosis information, surgical operation, and generating a ranking feature matrix, through a hybrid error correction engine E corr , generating an optimized recommendation result R opt , and providing an interactive confirmation interface for the user to decide whether to adopt.

[0074] The system calls the medical record front page information F rec , including diagnosis information D dia and surgical operation O opr , for the suspected grouping error cases identified in step S2. The system first generates a ranking feature matrix M sort = {Rank(D dia ), Rank(O opr ), reflecting the logical order and priority of diagnosis and surgical operation in the case, and then inputs M sort into the hybrid error correction engine E corr , which consists of a rule-based reasoning engine E rule and a deep recommendation model E model . The rule-based reasoning engine performs logical consistency checking on the diagnosis and operation fields according to the clinical pathway and medical insurance audit rules, generating preliminary correction suggestions. The deep recommendation model learns the implicit relationship between diagnosis and surgical operation using graph neural networks, calculates the optimal prediction probability, and outputs the recommended result R The system displays R opt through an interactive interface, and the user can choose to adopt or adjust the recommendation.

[0075] Step S4, after the user adopts the recommended result R opt , the updated case data D new is input into the scoring model of step S1 again to obtain the corrected score S1, and S0 and S1 are compared. If |S1-M ref | < |S0-M ref |, it is determined that the error correction is successful, otherwise it is marked as error correction failure, and the failed sample and error correction path are stored in the self-learning module for optimizing subsequent recommendation strategies.

[0076] In step S4, after the user confirms the adoption of the recommended result of step S3 in the interactive interface, the system generates new case data D new = (F rec \F err ) ∪ R opt according to the update rule, and then D new is re-input into the scoring model to calculate the corrected DRG score where I j ′The system compares the consumption index, adjusted by the hybrid error correction engine, with the initial score S0. If |S1-M ref |<|S0-M ref If the error is corrected, the correction is considered successful; otherwise, it is marked as a failure. All successful and failed samples and their corresponding correction paths are stored in the self-learning module. The system employs an incremental learning algorithm. Optimize recommendation strategy parameters to enable intelligent error correction capabilities to continuously improve with data accumulation, forming a closed-loop learning system.

[0077] refer to Figure 2 This invention provides an intelligent error correction system for automatic identification of DRG grouping errors, comprising the following modules:

[0078] The data processing and scoring module is used to preprocess and standardize the original case data, calculate the consumption index of various costs and length of hospital stay, and perform normalization calculations based on the weight matrix and the mean of the reference group to output the initial DRG score for subsequent anomaly detection.

[0079] The data processing and scoring module is primarily used for cleaning, preprocessing, standardizing, and scoring raw case data. The module first obtains raw case data from the hospital's billing system, drug management system, material management system, and electronic medical record system, including total hospitalization costs, drug costs, consumable costs, and length of stay. The data preprocessing unit handles missing values, filters outliers, and performs consistency checks on the raw data to ensure data integrity and computability. Subsequently, the consumption index calculation unit normalizes each indicator with the mean of the reference group to form a consumption index vector, and then weights and superimposes it according to the weight matrix to generate the initial DRG score. The scoring output unit of this module provides the initial DRG score to the anomaly detection and correction recommendation module, while retaining the standardized values ​​and consumption indices of each indicator for subsequent probability calculations and anomaly identification. The entire module realizes a complete process from raw data to standardized scoring. By dynamically updating the weight matrix, the scoring model can adapt to the cost characteristics of different hospitals or diseases, ensuring the reliability and interpretability of the initial score, while providing a solid data foundation for the system's intelligent error correction.

[0080] The anomaly detection and error correction recommendation module is used to perform distribution modeling and probability calculation based on the initial DRG score, identify suspected misgrouped cases, and generate a feature matrix by calling the information on the medical record front page. The matrix is ​​then input into the hybrid error correction engine to generate intelligent error correction recommendation results, while providing an interactive confirmation interface for users to adopt the recommendations.

[0081] The anomaly detection and correction recommendation module identifies suspected misgrouped cases and generates intelligent correction suggestions. First, the module fits the initial DRG scores output from step S1 to a distribution, using a skewed normal distribution function to fit the score set of all cases, thus obtaining location, scale, and skewness parameters. Then, the anomaly identification unit calculates the anomaly probability of each case and, combined with a user-defined threshold range, filters the initial scores, marking suspected misgrouped cases. The feature construction unit extracts diagnostic information and surgical operation data from the medical record homepage, generating a ranking feature matrix to describe the logical relationship and priority between diagnoses and operations. The hybrid error correction engine unit combines rule-based reasoning and a deep recommendation model to process the ranking feature matrix. The rule-based reasoning engine generates preliminary correction suggestions based on clinical pathways and medical insurance review rules. The deep recommendation model uses a graph neural network to learn the implicit relationship between diagnoses and surgical operations, outputting the optimal predicted probability and generating recommendation results. The module provides the correction results to the user for confirmation or adjustment through an interactive interface, completing the human-computer interaction loop and achieving a close integration of anomaly identification and intelligent recommendation, providing complete input data for subsequent score updates and self-learning modules.

[0082] The scoring update and self-learning module receives the error correction results adopted by the user, re-inputs the updated case data into the scoring model, calculates the corrected score, and stores the successful and unsuccessful error correction samples in the self-learning module. Through incremental learning algorithms, it dynamically optimizes the recommendation strategy to continuously improve the system's intelligent error correction capabilities.

[0083] The scoring update and self-learning module processes the error correction results adopted by users and continuously optimizes system performance. The data update unit merges the user-confirmed error correction results with the original case data to generate an updated case data set and removes suspected erroneous data marked in step S2. The score correction unit inputs the updated case data into the scoring model, calculates the corrected DRG score, and compares it with the initial score to determine whether the error correction was successful. The self-learning unit stores successful and failed samples in the database and dynamically adjusts the recommendation strategy parameters using an incremental learning algorithm, enabling the hybrid error correction engine to refer to historical experience to improve prediction accuracy in subsequent error corrections. This module achieves iterative upgrades to the system's intelligent error correction capabilities through continuous data accumulation and model optimization, ensuring that the scoring model and recommendation strategy can adapt to different diseases, different hospitals, and cost characteristics that change over time, thereby improving the robustness and scalability of the entire system and providing reliable decision support for hospital management and medical insurance review.

[0084] It should be noted that, unless otherwise specified, the embodiments and features and technical solutions in the present invention can be combined with each other.

[0085] Obviously, the above-described embodiments are only some embodiments but not all the embodiments of the present application, the preferred embodiments of the present application are shown in the drawings, but do not limit the patent scope of the present application. The present application can be implemented in many different forms, and conversely, the purpose of providing these embodiments is to make the disclosure of the present application more thorough and comprehensive. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions recorded in the foregoing specific embodiments, or make equivalent replacements to some technical features therein. Any equivalent structure made by using the content of the present application specification and drawings, directly or indirectly applied to other related technical fields, is also within the patent protection scope of the present application.

Claims

1. An automatic identification and intelligent error correction method for DRG grouping errors, characterized in that, Includes the following steps: Step S1: Based on the original case dataset D raw The total hospitalization cost C for the case tot Drug costs C med Consumable costs C mat , number of days of hospitalization L day Standardize the data and calculate the consumption index vector V corresponding to each indicator. ind And through the weight matrix W ind Compared with the mean data of the reference group M ref Normalized comparison is performed to obtain the initial DRG score S0; Step S2: Perform distribution fitting and interval division on the DRG score S0, and use the quartile model and the user-defined threshold interval [Q] to... low Q high ] Calculate the probability of abnormality P of the case. abn When the probability of an abnormal case P abn When the number exceeds the set range, it is marked as a suspected case of grouping error; Step S3: For suspected incorrectly grouped cases, retrieve the information from the medical record cover page F. rec This includes diagnostic information, surgical procedures, and the generation of a ranking feature matrix, which is then processed by a hybrid error correction engine E based on rule-based reasoning and a deep recommendation model. corr Generate optimized recommendation result R opt It also provides an interactive confirmation interface for users to decide whether to adopt the solution. Step S4: When the user accepts the recommendation result R opt Afterwards, the updated case data D new Input the scoring model from step S1 again to obtain the corrected score S1, and compare S0 with S1. If |S1-M ref |<|S0-M ref If the error is corrected, the error correction is considered successful; otherwise, it is marked as an error correction failure. The failed sample and the error correction path are stored in the self-learning module for optimization of subsequent recommendation strategies.

2. The automatic identification and intelligent error correction method for DRG grouping errors according to claim 1, characterized in that, The specific process of step S1 is as follows: S1(1), regarding the original case dataset D raw ={C tot C med C mat ,L day Preprocessing and standardization are performed: Among them, X i The original indicator values ​​for the cases, μ ref,i σ ref,i The mean and standard deviation of the same indicator in the reference group are respectively, and the total hospitalization cost C is... tot Sourced from hospital billing system, drug cost C med Sourced from the hospital's drug management system and consumable costs C mat Sourced from the hospital supplies management system and the number of days of hospital stay (L). day It is derived from the electronic medical record system and calculated based on the difference between the admission time and the discharge time; S1(2), Construct the consumption exponent vector V ind : Among them, M tot M med M mat M day The data are for the reference group, which includes total hospitalization costs, drug costs, consumable costs, and length of hospital stay. ref =*M tot M med M mat M day }; The initial DRG score S0 is obtained by weighted summation: Wherein, the weight matrix W ind Dynamically updated based on big data regression learning results, I j Take values ​​I respectively tot I med I mat I day .

3. The automatic identification and intelligent error correction method for DRG grouping errors according to claim 1, characterized in that, The specific process of step S2 is as follows: S2(1), the set of DRG scores for all cases Distribution modeling is performed using a skewed normal distribution function for fitting: Where φ(·) is the standard normal density function, Φ(·) is the standard normal distribution function, ζ is the location parameter, ω is the scale parameter, and α is the skewness parameter. ζ, ω, and α are obtained by statistical modeling and fitting the score set of step S1. S2(2) Calculate the abnormal probability of each case based on the fitted distribution: P abn =1-F(S0); Where F(S0) is the cumulative distribution value of the distribution function, derived from the distribution fitting result. If S0 does not belong to [Q... low Q high And P abn If the value is less than 0.9, it is marked as a suspected case of grouping error, F. err .

4. The automatic identification and intelligent error correction method for DRG grouping errors according to claim 1, characterized in that, The specific process of step S3 is as follows: For abnormal cases, the F field information on the medical record cover page is used. rec Extracting diagnostic information D dia Surgical Information O opr And generate a sorted feature matrix: M sort =*Rank(D dia ),Rank(O opr ); The sorting feature matrix M sort Input to hybrid error correction engine E corr The hybrid error correction engine consists of the following two parts: E, a rule reasoning engine based on clinical pathways and medical insurance review rules rule It is used to perform logical consistency checks on diagnostic and operational fields and generate preliminary correction suggestions; E, a deep recommendation model based on graph neural networks model It is used to learn the implicit relationship between diagnosis and surgical procedures and output the optimal prediction probability; The hybrid error correction engine E corr The recommended function is: The final recommendation result R is generated. opt It provides users with the option to adopt or not to adopt the approach through an interactive interface, thereby enabling intelligent error correction for incorrect case grouping.

5. The automatic identification and intelligent error correction method for DRG grouping errors according to claim 1, characterized in that, The specific process of step S4 is as follows: S4(1), when the user interacts with the hybrid error correction engine E in the interface corr The output recommendation result R opt After confirmation, the system generates new case data D according to the following update rules. new : D new =(F rec \F err )∪R opt ; Among them, F rec This indicates the information on the first page of the medical record (F). rec Including diagnostic information D dia Surgical Information O opr F err These are cases identified as suspected grouping errors during the anomaly detection process in step S2. S4(2), Update the case data D new Inputting the data into the scoring model yields the corrected score S1: Wherein, the weight matrix W ind Dynamically updated based on big data regression learning results. C′ tot C′ med C′ mat L′ day To achieve this through the hybrid error correction engine E corr For C tot C med C mat L day The data was re-analyzed after correction; S4(3) Compare the difference in scores before and after correction. If |S1-M ref |<|S0-M ref If the result is |, then the error correction is considered successful; otherwise, it is considered a failure, and the system will discard the failed sample (D). new ,R opt S1) Store the data in the self-learning module and use an incremental learning algorithm to optimize the recommendation strategy parameters: Where θ represents the model parameters and η represents the learning rate. The loss function is used to achieve dynamic evolution of the system's error correction capability.

6. An intelligent error correction system for automatic identification of DRG grouping errors, characterized in that, Includes the following modules: The data processing and scoring module is used to preprocess and standardize the original case data, calculate the consumption index of various costs and hospitalization days, and perform normalization calculation based on the weight matrix and the mean of the reference group to output the initial DRG score for subsequent anomaly detection. The anomaly detection and error correction recommendation module is used to perform distribution modeling and probability calculation based on the initial DRG score, identify suspected misgrouped cases, and generate a feature matrix by calling the information on the medical record front page. The matrix is ​​then input into the hybrid error correction engine to generate intelligent error correction recommendation results, while providing an interactive confirmation interface for users to adopt the recommendations. The scoring update and self-learning module receives the error correction results adopted by the user, re-inputs the updated case data into the scoring model, calculates the corrected score, and stores the successful and unsuccessful error correction samples in the self-learning module. Through incremental learning algorithms, it dynamically optimizes the recommendation strategy to continuously improve the system's intelligent error correction capabilities.

7. The automatic identification and intelligent error correction system for DRG grouping errors according to claim 6, characterized in that, The data processing and scoring module includes: The data preprocessing unit is used to clean, standardize, and handle missing values ​​of the original case data to ensure the integrity and computability of the input data. The consumption index calculation unit is used to normalize various costs and length of hospital stay with the mean of the reference group to form a consumption index vector, and then perform weighted summation based on the weight matrix to generate the initial DRG score. The scoring output unit is used to pass the initial DRG score to the anomaly detection and error correction recommendation module and provide basic data for subsequent anomaly probability calculation.

8. The automatic identification and intelligent error correction system for DRG grouping errors according to claim 6, characterized in that, The anomaly detection and error correction recommendation module includes: An anomaly identification unit is used to perform distribution fitting, probability calculation, and interval division on the initial DRG score, and to mark suspected misgrouped cases. The feature construction unit is used to extract diagnostic and surgical operation data from the medical record homepage information and generate a sorted feature matrix for use by the error correction model. The hybrid error correction engine unit consists of a rule-based reasoning engine and a deep recommendation model. It combines clinical rules and data-driven algorithms to generate intelligent error correction and recommendation results, which are then presented to the user for confirmation and adoption through an interactive interface.

9. The automatic identification and intelligent error correction system for DRG grouping errors according to claim 6, characterized in that, The scoring update and self-learning module includes: The data update unit is used to merge the user-confirmed and adopted error correction results with the original case data to generate an updated case data set. The corrected scoring unit is used to re-enter the updated case data into the scoring model to calculate the corrected score and compare it with the original score to determine whether the error correction was successful. The self-learning unit stores failed and successful error correction samples, and dynamically adjusts the recommendation strategy parameters in conjunction with the incremental learning algorithm, so as to realize the continuous optimization and iteration of the system's intelligent error correction capability in practical applications.

Images