Consistency evaluation-based disease grouping diagnosis and treatment path deviation risk assessment method and system

Abstract

The invention relates to the technical field of medical insurance, in particular to a risk assessment method and system for grouping diagnosis and treatment paths according to disease categories based on consistency evaluation, and the method comprises the steps: introducing a diagnosis and treatment path structured coding method oriented to medical insurance semantics; a path consistency modeling mechanism fusing structure clustering and behavior stability weight is constructed, the structure deviation degree of a case in a disease category can be identified and quantified, and meanwhile misjudgment of a low-frequency path is restrained through sparse regularization; through a medical insurance payment elastic adjustment function based on a deviation degree, a behavior deviation result is linked with a payment amount, and nonlinear mapping and a sensitivity regular term are introduced, so that the payment result can reflect reasonable variation of a path and can inhibit structural expansion; and finally, providing a combined risk scoring and structured label generation mechanism, comprehensively considering behavior deviation, payment loss and path sparsity, constructing a risk scoring model and outputting risk levels and type labels, and realizing a full-link closed loop.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical insurance, and in particular to a diagnosis and treatment path deviation risk assessment method and system based on consistency evaluation by disease grouping. BACKGROUND

[0002] In the current medical insurance payment system, the DRG / DIP mode of grouping by disease has become an important direction of medical insurance payment reform in China, and the core goal is to improve the efficiency of medical insurance fund use and management transparency through the same payment for the same disease.

[0003] However, in actual application, the complexity of diagnosis and treatment path, the diversity of behavior coding and the inconsistency of hospital implementation standards make it difficult for the existing payment system to accurately depict the real diagnosis and treatment process of the case. The existing method relies on the coding of the front page of the medical record or the path similarity calculation based on the mean to identify deviations, but this way ignores the structural difference and semantic complexity of medical insurance data in the behavior sequence. For example, some medical institutions may add unnecessary examinations or excessive treatment items in the diagnosis and treatment process, although they meet the coding requirements of medical insurance payment, but the path structure has deviated from the main path within the disease group. If this deviation is not identified, the medical insurance fund will be consumed invisibly, leading to the decline of system fairness and fund use efficiency. At the same time, the existing payment mechanism generally uses fixed standards for cost settlement, lacks the ability to dynamically adjust according to the degree of individual diagnosis and treatment path deviation, and cannot effectively suppress the excessive expenditure brought by abnormal path. The more prominent problem is that in the risk supervision and audit link, the medical insurance department usually relies on manual audit or simple detection means based on cost abnormalities, and it is difficult to establish a systematic connection between behavior deviation, payment risk and fund loss, leading to lagging risk early warning and obvious short board in fund risk control.

[0004] Therefore, there is an urgent need for a technical solution that can start from the structure of diagnosis and treatment path, combine the characteristics of medical insurance payment mechanism, model and quantify the risk of deviation behavior, so as to realize the closed-loop linkage of path expression, payment adjustment and risk assessment, and improve the scientific supervision level of medical insurance fund. SUMMARY

[0005] The purpose of the present application is to provide a diagnosis and treatment path deviation risk assessment method and system based on consistency evaluation by disease grouping to solve the problems raised in the background art.

[0006] To achieve the above purpose, the technical solution adopted by the present application is: The diagnosis and treatment path deviation risk assessment method based on consistency evaluation by disease grouping comprises the following steps: The structured diagnosis and treatment behavior data collected in the medical insurance system is converted into a vector expression from three dimensions of medical insurance project coding, behavior type and behavior sequence position, and the behavior vector is obtained after conversion and combination. All behavior vectors are spliced in the order of occurrence to form a case path vector matrix; Each case path vector matrix corresponds to a disease label. The structural consistency between any two paths is calculated for the disease label to obtain K similar paths under the disease label, and the deviation degree of the path set is calculated based on the K similar paths and the optimal matching path index corresponding to the disease label of the case; Based on the deviation degree of the path set, the final payment amount is calculated by combining the nonlinear payment adjustment function, and the final payment amount is input into the loss estimation model to output the structural payment loss estimation value, which is used to predict the excess expenditure or loss of medical insurance fund caused by behavior deviation; The comprehensive risk score of the case is calculated by combining the risk score function, and the corresponding structured risk label is generated based on the comprehensive risk score, which includes a risk level field and a risk type field.

[0007] Preferably, the structured diagnosis and treatment behavior data are periodically field aggregated by a data center interface, and are collected from a hospital medical record front page management system, a medical insurance settlement interface system and a medical insurance data governance platform.

[0008] More preferably, the structured diagnosis and treatment behavior data all have a unified field structure, including medical insurance directory coding, behavior category label and occurrence sequence number. The behavior sequence is sorted according to the occurrence time of the project to form a behavior sequence.

[0009] Preferably, in the calculation of the structural consistency, a position-independent structure clustering method is introduced to extract a representative path.

[0010] Preferably, in the deviation degree calculation process, a sparse regularization term of the representative path is introduced, which is the reciprocal of the proportion of the path matching the cases in the training set, and is used to punish the consistency with rare paths.

[0011] Preferably, in the nonlinear payment adjustment function, an indicator function is included, which is triggered when the deviation index exceeds the 95th percentile value of the historical deviation distribution of the disease, forming a rigid downward adjustment mechanism.

[0012] As preferred, the structural payment loss estimation value is calculated by a static offset loss term and a payment response sensitivity regularization term, wherein the static offset loss term is an excess part of the actual payment relative to the population mean of the no-offset path samples, and the payment response sensitivity regularization term is a derivative of the payment amount to the offset degree, used to reflect the sensitivity of the payment function at the case.

[0013] As preferred, the joint risk score function includes a nonlinear regularization function, which is used to reflect the proportional dominance of the behavior deviation indicator in all risk dimensions, and the greater the regularization term, the higher the score if the behavior deviation is much higher than the payment loss; if the payment anomaly is dominant and the behavior deviation is weak, it means that it may be a cost-type anomaly, and the score is adjusted lower.

[0014] As preferred, the risk level field is set by the intra-disease quintile segmentation method, and the threshold values are 60%, 85% and 97%, which are divided into four grades of low risk, medium risk, high risk and extremely high risk.

[0015] The diagnosis and treatment path deviation risk assessment system based on consistency evaluation and grouped by disease includes sequentially connected: The data acquisition and processing module is used for converting each data from the medical insurance system into a vector expression from three dimensions of medical insurance project coding, behavior type and behavior sequence position, obtaining a behavior vector after conversion and combination, splicing all behavior vectors according to the occurrence sequence to form a case path vector matrix; The path consistency deviation calculation module is used for corresponding to a disease label in each case path vector matrix, obtaining K similar paths under the disease label by calculating the structural consistency between any two paths, and calculating the deviation degree of the path set and the optimal matching path index corresponding to the case belonging to the disease label based on the K similar paths; The medical insurance flexible payment module is used for calculating the final payment amount based on the deviation degree of the path set and combining a nonlinear payment adjustment function, inputting the final payment amount into the loss estimation model, and outputting a structural payment loss estimation value, which is used to predict the excess expenditure or loss of medical insurance fund caused by behavior deviation; The risk calculation and label output module is used for calculating the comprehensive risk score of the case by a joint risk score function, and generating a corresponding structured risk label based on the comprehensive risk score, wherein the structured risk label includes a risk level field and a risk type field.

[0016] Compared with the prior art, the present application has the following advantages: The application provides a diagnosis and treatment path deviation risk assessment method and system based on consistency evaluation, which can overcome the deficiencies of the prior art in path identification, payment adjustment and risk expression. By introducing a diagnosis and treatment path structured coding method oriented to medical insurance semantics, standardized vectorization of case diagnosis and treatment behavior is realized, avoiding the instability and uninterpretability of general semantic models in the medical insurance field. On this basis, a path consistency modeling mechanism is constructed by combining structural clustering and behavior stability weight, which can identify and quantify the structural deviation degree of cases within the disease category, and suppress the misjudgment of low-frequency paths through sparse regularization. Further, the application designs a medical insurance payment elasticity adjustment function based on deviation degree, which links the behavior deviation result with the payment amount, and introduces a nonlinear mapping and sensitivity regularization term, so that the payment result can reflect the reasonable variation of the path and suppress the structural expansion behavior, and further quantify the loss value of the excessive expenditure. Finally, the application proposes a joint risk score and structured label generation mechanism, which comprehensively considers behavior deviation, payment loss and path sparsity, constructs a risk score model and outputs risk level and type label, realizing the full-link closed loop from path modeling to risk expression. The overall scheme of the application can realize dynamic quantification of path deviation, fine adjustment of payment and interpretable expression of fund risk in the medical insurance payment system, effectively improving the use efficiency and supervision ability of the medical insurance fund. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A flowchart of the diagnosis and treatment path deviation risk assessment method based on consistency evaluation according to an embodiment of the application is shown in the figure. Figure 2 A block diagram of the diagnosis and treatment path deviation risk assessment system based on consistency evaluation according to an embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the accompanying drawings of the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of the application.

[0019] Please refer to Figure 1 The application provides a diagnosis and treatment path deviation risk assessment method based on consistency evaluation, which includes: Step 1: The structured diagnosis and treatment behavior data collected in the medical insurance system are converted into vector expressions from three dimensions of medical insurance item coding, behavior type and behavior sequence position, respectively. After conversion and combination, the behavior vectors are obtained, and all the behavior vectors are spliced according to the occurrence order to form a case path vector matrix, which specifically includes: This step aims to transform the structured diagnosis and treatment behavior data collected in the medical insurance system into a computable and comparable path vector expression form as the input for subsequent path consistency modeling. Under the payment system by disease grouping (DRG / DIP), medical insurance data has the characteristics of complex coding structure, various types of item behavior, and chaotic behavior order, which cannot be directly used for deviation identification modeling. Therefore, this step proposes a structured path representation method for the characteristics of medical insurance data, which realizes the unified vectorization expression of diagnosis and treatment behavior without introducing external semantic models.

[0020] The original data mainly comes from three types of interface systems: hospital medical record management system (HIS system), medical insurance settlement interface system, and medical insurance data governance platform. These systems regularly gather data through the data center interface to form a unified case diagnosis and treatment behavior record structure. Taking a high-age pneumonia patient as an example, the structured behavior sequence includes: admission diagnosis code J18.9; examination behavior: chest CT (medical insurance item code A1020401); treatment behavior: intravenous injection of antibiotics, inhalation (medical insurance item code B3010104, B2010102); drug use: ceftriaxone sodium injection (medical insurance drug code Y0011001); no surgical operation; discharge diagnosis is the same as admission.

[0021] All the above diagnosis and treatment behaviors have a unified field structure: medical insurance catalog code (denoted as ), behavior category label (denoted as ), and occurrence order number (denoted as ). The system sorts the behavior sequence according to the time of item occurrence to form the behavior sequence , where each is a diagnosis and treatment behavior.

[0022] Considering the high discreteness of medical insurance catalog codes and the large semantic differences between behavior categories, this step represents each behavior from three dimensions: medical insurance item code, behavior type, and behavior order position, and constructs the merged behavior vector : ; where represents the lookup vector corresponding to the medical insurance catalog code , which is derived from the national medical insurance service item and drug catalog standard and is a fixed-length dense vector; represents the enumeration mapping vector of the behavior type label , which includes five types of diagnosis, examination, treatment, medication, and surgery, and each type of behavior is mapped to an independent vector; represents the order number of the behavior in the entire hospitalization process Linearly mapped vectors, used to preserve the time information of the occurrence order of behaviors.

[0023] Concatenate all behavior vectors in the order of occurrence to form a case path vector matrix: ; Where, represents the diagnosis and treatment path of the th case, is the number of behaviors, is the vector dimension. If the number of behaviors of a certain case is less than the maximum length set by the system, zero vectors are added on the right side to make the length consistent, which is convenient for subsequent model batch processing. During the vector generation process, the three sub-vectors are normalized before mapping to ensure the numerical consistency of the encoding space and avoid interference of a certain dimension on the overall path representation.

[0024] For example, if a case has 6 behaviors (such as 2 examinations, 3 treatments, and 1 medication) during diagnosis and treatment, the dimension of its path matrix is ; when the system uniformly sets the maximum path length to 20, the final completed standard vector is a matrix. This matrix serves as the input for the next step of modeling and evaluating path consistency, with structural uniformity, complete encoding, and medical insurance business traceability.

[0025] The key innovation of this step is to design a path structure representation method centered on medical insurance behavior semantics, which does not rely on general semantic models or natural language processing mechanisms, ensuring the encoding stability of the path vector and the consistency of the regulatory semantics from the source. The three-dimensional joint embedding method not only improves the expression ability of the diagnosis and treatment path in medical insurance business, but also significantly enhances the explainability and accuracy of subsequent deviation modeling.

[0026] Step one finally obtains a case path vector matrix , each row of which represents the encoding result of a diagnosis and treatment behavior, and the entire matrix serves as the input feature of the consistency deviation evaluation model.

[0027] Step two: In each case path vector matrix, a disease label is correspondingly assigned. By calculating the structural consistency of the disease label between any two paths, K similar paths under the disease label are obtained, and based on the K similar paths, the deviation degree of the path set and the optimal matching path index of the case corresponding to the disease label are calculated. Specifically, it includes: The purpose of step two is to establish a path consistency modeling mechanism with medical insurance semantic understanding ability on the basis of the case path vector matrix constructed in step one, to measure whether the diagnosis and treatment path of each case in its corresponding disease grouping (DRG / DIP) exists structural deviation. Due to the nonlinear and branching evolution characteristics of the medical insurance path structure itself, the conventional scheme based on average path or path similarity cannot accurately reflect the real distribution of the "standard path". This step introduces an innovative method that combines structural clustering and behavior stability modeling, focusing on identifying potential deviation behaviors that "meet the coding rules but deviate from the main path", and finally outputs a deviation index with quantifiable explanatory power.

[0028] The case path vector obtained in the first step Each corresponds to a disease label The label is automatically generated by the grouping device (such as CHS-DRG or DIP engine) deployed by the medical insurance bureau during the case grouping stage. All paths are aggregated according to the disease label to form a path set and the intra-group consistency analysis is carried out accordingly.

[0029] Unlike traditional schemes based on sample mean or K-means path center representation, this scheme fully considers the characteristics of "behavior point misalignment" and "path diversity" of medical insurance paths, and introduces a position-independent structural clustering method for representative path extraction. Specifically, the structural similarity between any two paths in the disease group and is calculated, defined as follows: ; Where represents the th behavior embedding vector of the case , is the position importance weight (such as diagnosis behavior weight higher than examination and treatment behavior), is the behavior index of the case , represents the closest behavior to . This similarity measure allows for incomplete alignment of behaviors between paths, adapts to the reality of path behavior order, missing conditions, and is more sensitive to "behavior structure disturbance" in the actual medical insurance scenario. After clustering, we can obtain representative paths under the disease as the reference baseline for subsequent deviation calculation.

[0030] For the calculation of case ​The deviation index is introduced as follows with respect to the deviation degree of the path set: ; wherein, is the behavior embedding vector of the representative path , is the behavior consistency weight of the behavior position within the disease type (obtained by normalizing the variance of behaviors within the disease type), is a normalization factor, is a sparse regularization term of the representative path, defined as the reciprocal of the proportion of the path matching cases in the training set, to punish consistency with rare paths. is a sparsity adjustment parameter, which prevents the model from misjudging the "minority path" as deviating from the main path, and embodies the support logic for medical insurance reform "type management". The deviation degree

[0031] combines the behavior level structural difference and the path frequency penalty mechanism, not only depicting the point position deviation degree of the case path, but also automatically identifying and suppressing "structural rare paths", with strong medical insurance business interpretability. The main innovations are as follows: Firstly, the minimum matching mechanism of the misaligned behavior position, aiming at the common non-structural disturbance of "behavior loss" and "behavior order change" in medical insurance path, improves the robustness of path modeling. Secondly, the weighted strategy based on behavior stability is introduced ( ), highlighting the "key behavior consistency" rather than the overall path consistency of the medical insurance system. Thirdly, the path sparse regularization term avoids model bias to "edge path", and the regularization factor comes from the real coverage proportion of the case, which can be counted by the daily cumulative number of cases on the platform, without additional parameter estimation, with strong engineering deployability.

[0032] Finally, this step obtains: ① the deviation value of each case, which will be an important input of the risk score model; and ② the optimal matching path index of the case, which can be used for diagnosis and treatment path typing, auditing visualization and other business scenarios. This step, as the core logic layer of the whole system, establishes a pathway from medical insurance path vector to deviation value, directly connecting "path expression" and "risk judgment", and provides key data support for subsequent medical insurance payment risk estimation and regulation.

[0033] ​Step three: based on the degree of deviation of the path set, the final payment amount is calculated by combining the nonlinear payment adjustment function, and the final payment amount is input into the loss estimation model to output the structural payment loss estimation value, which is used to predict the excess expenditure or loss of the medical insurance fund caused by behavior deviation, specifically including: This step is based on the behavior deviation index and the corresponding representative path index , to build a medical insurance payment elasticity adjustment mechanism and a structural loss estimation model.

[0034] This step is the "payment decision mapping" core part of the present application scheme, which is used to convert the upstream "structural deviation identification" into the specific operation logic of "payment behavior adjustment" and "fund risk quantification", and to connect the path deviation modeling result and the medical insurance payment system, so that the whole system has a closed-loop execution capability.

[0035] Under the DRG / DIP payment system by disease group, the medical insurance payment amount is highly dependent on the average cost value of the disease group, and does not consider whether the individual diagnosis and treatment path structure deviates, which is easy to cause the medical insurance fund to make non-discriminatory expenditure on the path structure. Especially in the actual business scene, there are some medical institutions that evade traditional compliance review methods through "path extension" and "unnecessary service coverage", etc., causing structural expansion of medical insurance expenditure. This step designs a path deviation adjustment function to build a dynamic flexible payment model, and introduces a segmented regular and offset modulation mechanism, so that the payment result can truly reflect the structural risk of the diagnosis and treatment path, and then output the quantitative loss index for medical insurance fund audit and early warning.

[0036] Based on the behavior deviation index which is derived from the weighted measurement result of the structural difference between the case path and the main path in the last step; the representative path index is used to obtain the corresponding standard payment benchmark value ; and the historical payment distribution data within the case group , which is generated after cleaning processes such as outlier elimination, disease grouping filtering, and regional balance normalization, has good stability and referenceability.

[0037] In order to depict the adjustment response of medical insurance payment to behavior deviation, the following nonlinear payment adjustment function is constructed: ; Wherein, represents the final payment amount, the medical insurance payment amount is used for fee settlement with the medical insurance settlement system, a payment value of the matched trunk path standard for the case, a regular coefficient of the conventional deviation, reflecting a normal adjustment degree of the structural deviation to the payment amount, a deviation starting adjustment threshold value; an extreme deviation penalty coefficient, an indicator function, triggered when the deviation index exceeds the 95th percentile value of the historical deviation distribution of the disease , forming a rigid downward adjustment mechanism. This design embodies the combined payment strategy of "mild adjustment + rigid punishment", which not only takes into account the reasonable variation of the path, but also identifies structural avoidance behavior, and has a high degree of medical insurance reality fit.

[0038] Further, to estimate the excess expenditure or loss of medical insurance fund caused by behavior deviation, a structural loss estimation model is constructed as follows: ; wherein, represents the structural payment loss value of the case , and the structural payment loss estimation value is used for the fund risk control, abnormal case mining, payment system optimization and other business applications of the medical insurance bureau or supervision platform. The first item is the static deviation loss, that is, the excess part of the actual payment relative to the mean value of the "no deviation path" sample group, which can be calculated by the sample set with a deviation index less than the threshold value in the disease group ; the second item is a payment response sensitivity regular term, which represents the derivative of the payment amount to the deviation degree, reflecting the sensitivity of the payment function at the case, and the response regular weight. This innovative design can effectively identify those "over-concentrated payment adjustment" structural sensitive areas, and prevent the problem of mispenalty or misaward caused by the oversteep structure of the payment function.

[0039] This loss model provides key risk assessment capabilities for the present application. On the one hand, it is based on objective medical insurance policy data (standard payment, deviation degree, historical sample), and does not rely on subjective scoring; on the other hand, it introduces a regular term by the derivative of the payment function, controls the "reaction degree of payment to deviation" of the medical insurance system, and improves the continuity and controllability of the adjustment.

[0040] Step four: calculate the comprehensive risk score of the case by combining the risk score function, and generate the corresponding structured risk label based on the comprehensive risk score, wherein the structured risk label includes a risk level field and a risk type field, and specifically includes: After completing the first three steps of "structured path modeling - behavior deviation identification - payment elasticity adjustment", in order to realize the closed-loop supervision of the risk of medical insurance fund use, based on the existing quantitative results of structural risk, a unified joint risk scoring mechanism is further constructed in this step, and a structured risk label is generated for the medical insurance audit system, intelligent audit platform, risk control model or policy regulation system to call and use.

[0041] This step is the last link of "perception - regulation - expression" in the technical chain of the present application, and is also the key bridge to realize the interface landing from technical analysis results to management decisions. Considering the regulatory attributes of medical insurance payment scenarios and the diversity of decision-making scenarios (such as automatic audit, classification early warning, medical insurance negotiation, etc.), the "index correlation regularization" "sparse path activation term" "quantile segmentation scoring strategy" and other structures are introduced in the scoring modeling to improve the explainability of the score, the continuity of the regulation and the adaptability of the model.

[0042] This step first constructs a joint risk scoring function to comprehensively measure the "structural behavior heterogeneity" "financial expenditure abnormality" and "system behavior rarity" of medical insurance settlement risk. The scoring function is defined as follows: ; Wherein, is the comprehensive risk score of the case , , , is the weight of the scoring item, which can be configured by the regulatory system or learned by regression of historical audit samples; is the standard payment value of the main path to which the case belongs ; is the historical sample coverage rate of the main path in the disease, reflecting its behavior representativeness; is the index balance regularization coefficient, is a very small constant used to balance the scale. In particular, the fourth term is a structural innovation designed in this step: it is a nonlinear regularization function that reflects the proportion of behavior deviation indicators in all risk dimensions. If the behavior deviation is much higher than the payment loss, the regularization term will be larger, and the score will be higher. If the payment anomaly is dominant and the behavior deviation is weak, it means that it may be a cost-type anomaly, and the score is adjusted lower. This design strengthens the "dominance" of structural risk in joint scoring, prevents the score from being dominated by short-term cost fluctuations, and improves the identification sensitivity of the scoring model to structural fraud behavior.

[0043] Subsequently, this step designs a structured risk label generation mechanism to output the label field . The label consists of two parts: risk level field : Adopting the in-disease sub-quantile segmentation method to set, setting the threshold value as: , , , respectively, divided into four grades of “low risk”, “medium risk”, “high risk” and “extremely high risk”. The sub-quantile calculation is based on the distribution in the disease group, automatically adapts to the risk structure of different diseases, and has good cross-disease consistency.

[0044] Risk type field : According to the relative intensity of and , set: If and , it is “behavior deviation type”; If and , it is “payment overflow type”; If both are high, it is marked as “compound risk type”; If both are low, it is marked as “risk not significant”.

[0045] Each parameter can be optimized and set by the platform according to the statistical characteristics of different disease groups.

[0046] Finally, the and will be stored in the case risk data set, which is used for medical insurance risk control engine calling, audit task scheduling, intelligent audit strategy adjustment and other business processes. Among them can be directly used in the sorting type risk early warning model, can be directly used for structured decision tree, rule system or artificial audit system.

[0047] Please refer to Figure 2 , in the second aspect of the present application, a diagnosis and treatment path deviation risk assessment system based on consistency evaluation is proposed according to disease groups, which comprises sequentially connected: Data acquisition and processing module, for converting the structured diagnosis and treatment behavior data collected from the medical insurance system into vector expression from three dimensions of medical insurance project code, behavior type and behavior sequence position, respectively, obtaining behavior vector after conversion and combination, splicing all behavior vectors according to the occurrence order to form a case path vector matrix; Path consistency deviation calculation module, for corresponding to a disease label in each case path vector matrix, calculating the structural consistency between any two paths under the disease label, obtaining K similar paths under the disease label, and calculating the deviation degree of the path set and the optimal matching path index corresponding to the case belonging to the disease label based on the K similar paths.​ The medical insurance elastic payment module is used for calculating a final payment amount based on the degree of deviation of the path set in combination with a nonlinear payment adjustment function, inputting the final payment amount into a loss estimation model, and outputting a structural payment loss estimation value, which is used for predicting the excess expenditure or loss of the medical insurance fund caused by behavior deviation; The risk calculation and label output module is used for calculating a comprehensive risk score of the case through a joint risk scoring function, and generating a corresponding structured risk label based on the comprehensive risk score, wherein the structured risk label includes a risk level field and a risk type field. The above embodiments only describe the preferred embodiments of the present application, and do not limit the scope of the present application. Without departing from the design spirit of the present application, various modifications and improvements of the technical solutions of the present application made by the ordinary engineering technicians in the art shall fall within the protection scope determined by the claims of the present application.

Claims

1. A method for evaluating the risk of deviation from a diagnosis and treatment path by disease category based on consistency evaluation, characterized by, Includes the following steps: The structured diagnosis and treatment behavior data collected from the medical insurance system are converted into vector representations for each data item from three dimensions: medical insurance item code, behavior type, and behavior sequence position. After conversion and merging, behavior vectors are obtained. All behavior vectors are concatenated in the order of occurrence to form a case path vector matrix. Each case path vector matrix corresponds to a disease label. By calculating the structural consistency between any two paths for that disease label, K similar paths under that disease label are obtained. Based on the K similar paths, the deviation of the path set and the index of the optimal matching path to which the case corresponding to that disease label belongs are calculated. Based on the degree of deviation of the path set, the final payment amount is calculated by combining the nonlinear payment adjustment function. The final payment amount is then input into the loss estimation model to output a structural payment loss estimate. The structural payment loss estimate is used to predict the excess expenditure or loss of the medical insurance fund caused by behavioral deviation. The comprehensive risk score of a case is calculated by a joint risk scoring function, and a corresponding structured risk label is generated based on the comprehensive risk score. The structured risk label includes a risk level field and a risk type field. 2.The consistency evaluation-based diagnosis and treatment path grouping by disease type deviation risk assessment method according to claim 1, characterized in that, The structured diagnosis and treatment behavior data is collected periodically by the data platform interface from the hospital's medical record homepage management system, medical insurance settlement interface system, and medical insurance data governance platform. 3.The consistency evaluation-based diagnosis and treatment path grouping by disease type deviation risk assessment method according to claim 2, characterized in that, The structured medical behavior data all have a unified field structure, including medical insurance catalog code, behavior category label and occurrence sequence number. The behavior sequence is sorted according to the order of occurrence of the items to form a behavior sequence. 4.The consistency evaluation-based diagnosis and treatment path grouping by disease type deviation risk assessment method according to claim 1, characterized in that, In the calculation of the structural consistency, representative paths are extracted by introducing a position-independent structural clustering method. 5.The consistency evaluation-based diagnosis and treatment path grouping by disease type deviation risk assessment method according to claim 1, characterized in that, In the process of calculating the degree of deviation, a sparse regularization term representing the path is introduced, which is the inverse of the proportion of cases matching the path in the training set, and is used to penalize consistency with rare paths. 6.The consistency evaluation-based diagnosis and treatment path grouping by disease type deviation risk assessment method according to claim 1, characterized in that, The nonlinear payment adjustment function includes an indicator function, which is triggered when the deviation index exceeds the 95th percentile of the historical deviation distribution of the disease, thus forming a rigid downward adjustment mechanism. 7.The consistency evaluation-based diagnosis and treatment path grouping by disease type deviation risk assessment method according to claim 1, characterized in that, The estimated structural payment loss is calculated using a static offset loss term and a payment response sensitivity regularization term. The static offset loss term is the excess of the actual payment relative to the mean of the sample group without deviation, and the payment response sensitivity regularization term is the derivative of the payment amount with respect to the deviation, used to reflect the sensitivity of the payment function at this case. 8.The consistency evaluation-based diagnosis and treatment path grouping by disease type deviation risk assessment method according to claim 1, characterized in that, The joint risk scoring function includes a nonlinear regularization function, which is used to reflect the proportion of behavioral deviation indicators in all risk dimensions. If the behavioral deviation is much higher than the payment loss, the larger the regularization term, the higher the score. If the payment anomaly is dominant and the behavioral deviation is weak, it indicates that it may be a cost-type anomaly, and the score is lowered. 9.The consistency evaluation-based diagnosis and treatment path grouping by disease type deviation risk assessment method according to claim 1, characterized in that, The risk level field is set using a disease-specific quantile segmentation method, with thresholds set at 60%, 85%, and 97%, respectively, dividing the risk into four levels: low risk, medium risk, high risk, and extremely high risk.

10. A risk assessment system for deviation from a diagnosis and treatment pathway by disease group based on consistency evaluation, characterized by, Including those connected sequentially: The data acquisition and processing module is used to convert the structured diagnosis and treatment behavior data collected from the medical insurance system into vector expressions for each data item from three dimensions: medical insurance item code, behavior type and behavior sequence position. After conversion and merging, the behavior vector is obtained. All behavior vectors are concatenated in the order of occurrence to form a case path vector matrix. The path consistency deviation calculation module is used to calculate the structural consistency of the disease label between any two paths in each case path vector matrix, obtain K similar paths under the disease label, and calculate the deviation degree of the path set and the optimal matching path index of the case corresponding to the disease label based on the K similar paths. The medical insurance flexible payment module is used to calculate the final payment amount based on the degree of deviation of the path set and combined with the nonlinear payment adjustment function. The final payment amount is then input into the loss estimation model to output a structural payment loss estimate. The structural payment loss estimate is used to predict the excess expenditure or loss of the medical insurance fund caused by behavioral deviation. The risk calculation and label output module is used to calculate the comprehensive risk score of a case through a joint risk scoring function, and generate a corresponding structured risk label based on the comprehensive risk score. The structured risk label includes a risk level field and a risk type field.

Images