Disease cost optimization method and device, terminal equipment and storage medium

Abstract

The invention provides a disease cost optimization method and device, terminal equipment and a storage medium. The method comprises the following steps: acquiring current disease resource consumption information; the disease resource consumption information comprises department disease resource consumption information and department doctor disease resource consumption information; inputting the current disease resource consumption information into a pre-constructed disease standard cost model for cost calculation to obtain corresponding standard cost information; performing difference test on the department disease resource consumption information and the standard cost information, judging the disease to which the department disease resource consumption information whose difference exceeds a preset threshold value belongs as an abnormal disease, and obtaining disease resource consumption information of a first department doctor under the abnormal disease, generating a difference comparison result between the disease resource consumption information of the first department doctor and the corresponding standard cost information; and according to a difference comparison result, generating a disease cost optimization strategy of the next period. According to the invention, efficient and accurate disease cost optimization decision can be realized.

Description

Technical Field

[0001] This invention relates to the field of digital healthcare technology, and in particular to methods, devices, terminal equipment, and storage media for optimizing disease-specific costs. Background Technology

[0002] The current situation regarding the consumption of medical resources is complex. On the one hand, population aging and changes in the disease spectrum are driving a continuous increase in medical demand, leading to an overall increase in resource consumption; on the other hand, resources are unevenly distributed.

[0003] The current hospital cost structure for each disease covers medication costs, examination fees, treatment fees, nursing fees, and bed fees, with medication and examination fees accounting for a relatively high proportion. Its management primarily relies on statistical analysis of actual data incurred during the current period, allowing doctors and operations departments to understand the cost situation. Whether the cost structure is reasonable and where optimization should be pursued largely depends on experience. However, this management approach has significant flaws. It lacks scientific rigor and precision, and relying on experience makes it difficult to grasp the composition and changing patterns of costs, thus hindering the development of effective optimization plans. How to scientifically adjust the cost structure for each disease to achieve efficient and intelligent decision-making is a core requirement for the current refined management of hospital operations. Summary of the Invention

[0004] The present invention aims to provide a method, apparatus, terminal equipment and storage medium for optimizing disease-specific costs, so as to solve the above-mentioned technical problems and achieve efficient and accurate decision-making for optimizing disease-specific costs.

[0005] To address the aforementioned technical problems, this invention provides a method for optimizing disease-specific costs, comprising: Obtain the current period's disease resource consumption information; this includes departmental disease resource consumption information and departmental physician disease resource consumption information. The current period's disease-specific resource consumption information is input into a pre-built disease-specific standard cost model for cost calculation, resulting in corresponding standard cost information. The disease-specific standard cost model is trained based on a historical sample set of disease-specific resource consumption information. The departmental disease resource consumption information and standard cost information are compared. The disease to which the departmental disease resource consumption information belongs with the difference exceeding the preset threshold is identified as an abnormal disease. The disease resource consumption information of the first department doctor under the abnormal disease is obtained, and the difference comparison results between the disease resource consumption information of the first department doctor and the corresponding standard cost information are generated. Based on the comparison results, a disease-specific cost optimization strategy for the next period is generated.

[0006] In the above scheme, a standard cost model for each disease, pre-built and trained based on historical data, is used to calculate costs by combining current disease resource consumption information. This yields standard cost information that closely reflects the actual situation, providing a scientific and accurate reference for subsequent cost management. A difference test is used to compare departmental disease resource consumption information with standard cost information, accurately identifying abnormal diseases where the difference exceeds a preset threshold. This allows for focusing on key issues and concentrating efforts on resolving cost anomalies. After identifying abnormal diseases, the resource consumption information of the doctors in the first department under that abnormal disease is further obtained and compared, deepening the analysis to the doctor level and clarifying the specific source of cost anomalies, making subsequent optimization strategies more targeted. Based on the difference comparison results, a disease cost optimization strategy for the next period is generated. Cost management measures are adjusted in a timely manner according to the actual situation to reasonably optimize disease costs, improve resource utilization efficiency, and enhance the level of medical cost control.

[0007] In one implementation, obtaining the current period's disease-specific resource consumption information specifically includes: Obtain the inpatient medical record data for this period; Each inpatient's medical record data is categorized into its corresponding disease type, and medical resource consumption information is calculated separately for both the disease type and physician dimensions to obtain departmental disease type resource consumption information and departmental physician disease type resource consumption information. Among them, the departmental disease type resource consumption information represents the average medical cost of each case in the same disease type within the department, and the departmental physician disease type resource consumption information represents the average medical cost of different doctors in the same disease type within the department.

[0008] The above approach begins by obtaining information from the medical records of inpatients in the current period. These records are authentic records of the medical process, accurately reflecting the actual consumption of medical resources and providing a solid and reliable data foundation for subsequent analysis. The medical record data is divided by disease type and physician, and medical resource consumption information is calculated separately for each, yielding departmental disease-specific resource consumption information and departmental physician-specific disease-specific resource consumption information. This multi-dimensional analysis provides a comprehensive understanding of the distribution and use of medical expenses from both a macro-level (disease-specific) and micro-level (physician) perspective.

[0009] In one implementation, the disease-specific cost optimization method further includes: constructing a sample set of historical disease-specific resource consumption information, specifically: Retrieve the inpatient medical record data from the previous period; the inpatient medical record data includes basic patient information, patient disease characteristics, medical resource characteristics, and patient cost data; Preprocessing of inpatient medical record data; including removing cases with missing or abnormal data and removing cases with abnormal patient costs; Each inpatient's medical record data is classified into its corresponding disease category, and the standard deviation of medical expenses for each case under each disease category is calculated one by one. The standard deviation of medical expenses is based on the difference between the medical expenses of each case and the mean medical expenses of each case under the corresponding disease category. Inpatient medical records with medical cost standard deviations exceeding a preset standard deviation threshold are removed to obtain a historical disease resource consumption information sample set for each disease; among them, the number of inpatient medical records for each disease in the historical disease resource consumption information sample set exceeds a preset threshold.

[0010] In the above scheme, by acquiring and preprocessing the medical records of inpatients from the previous period, cases with missing data, anomalies, and abnormal patient costs are removed, ensuring the accuracy and reliability of the data used to construct the sample set and providing a high-quality data foundation for subsequent model training and analysis. Calculating the standard deviation of medical costs for each case under each disease category can accurately measure the difference between the case cost and the mean cost of the disease category, helping to identify abnormal cases with large cost fluctuations. The scheme requires that the number of inpatient medical records included under each disease category exceeds a preset threshold, ensuring that the sample set is of sufficient size to better reflect the actual resource consumption of the disease category and improve the scientific rigor and reliability of subsequent analysis and decision-making.

[0011] In one implementation, the standard cost model for diseases is trained based on a historical sample set of disease resource consumption information. The standard cost model for diseases is trained in the following way: Feature data and tag data are established for the data content of each historical disease resource consumption information sample set; The number of neurons, d, is set based on the amount of feature data, and the number of hidden layers is set to 2. Construct the network topology and set the standardized feature vectors as X∈R d (X=[x1,x2,…,x) d ] T ).

[0012] Forward propagation is used and the output of each layer is set, specifically: Output of the first hidden layer: H1 = σ(W1X + b1) (W1 ∈ R) n1×d Let b1 ∈ R be the weight matrix. n1 (for bias vectors) Output of the second hidden layer: H2 = σ(W2H1 + b2) (W2 ∈ R) n2×n1 Let b2 be the weight matrix. Rn2 (for bias vectors) Output layer predictions: =W3H2+b3(W3∈ R1×n2Let b3 be the weight matrix and b3 ∈ R be the bias vector. (Forecast costs); The historical disease resource consumption information sample set is divided into a training set and a validation set according to a preset ratio. The network topology is iteratively trained based on the training set, and the parameters of the network topology are updated until the optimal loss function is obtained. The loss function is obtained based on the difference between the predicted and actual values ​​of the labeled data, and its expression is: ; Where N is the number of samples, Yi is the actual cost of the i-th sample, and λ is the regularization coefficient; The average performance of the network topology is verified several times based on the verification set, and its coefficient of determination is calculated; wherein the expression for the coefficient of determination is: ; in As the coefficient of determination, This represents the average of the actual costs. When the determination coefficient exceeds a preset threshold, the training of the standard cost model for the current disease is deemed complete. In the above scheme, feature data and label data are established for the historical disease resource consumption information sample set, and the number of hidden layers is set according to the preset training batch to improve training efficiency and enable the model to learn data features more comprehensively. By iteratively training the model and continuously updating the parameters to minimize the loss function, the model can gradually adapt to the data features, improve the accuracy and reliability of prediction, and ensure that the model can learn the inherent patterns in the data.

[0013] In one implementation, feature data and label data are established for the data content of each historical disease resource consumption information sample set, specifically as follows: The characteristic data includes patient gender, age, medical insurance type, primary diagnosis, admission route, discharge method, treatment method, primary surgery, surgical level, number of procedures, whether the patient was admitted to the intensive care unit, and number of resuscitation attempts. The data includes total cost, drug cost, examination and testing cost, medical service cost, and consumable cost.

[0014] In the above scheme, the feature data covers various aspects such as patient gender, age, and medical insurance type, comprehensively reflecting the patient's basic situation, disease characteristics, and treatment process. This provides rich and accurate input for the disease-specific cost model, enabling the model to more accurately capture various factors affecting costs. The label data includes specific cost categories such as total cost and drug costs, clarifying the model's prediction target and closely related to the actual medical cost structure. This helps the model learn the patterns of cost generation, providing a direct and effective basis for cost prediction and management. The rational construction of feature data and label data allows the model to better learn the inherent patterns and rules in historical disease-specific resource consumption information sample sets, thereby improving the prediction accuracy and reliability of the disease-specific cost model and providing stronger support for disease-specific cost optimization.

[0015] In one implementation, the resource consumption information of the first department doctor under the abnormal disease category is obtained, and a comparison result of the difference between the resource consumption information of the first department doctor and the corresponding standard cost information is generated, specifically including: Obtain the disease resource consumption information of the first department doctor under the abnormal disease; among which, the disease resource consumption information of the first department doctor is the average medical expenses of different doctors in the department under the abnormal disease. The average medical expenses include the average total cost per case, the average drug cost per case, the average examination and testing cost per case, the average medical service cost per case, and the average consumable cost per case. Initial hypothesis tests are established based on the average medical expenses of each doctor; the initial hypothesis test is that the average medical expenses of doctors are no different from the average standard expenses. The average total cost per case for each doctor is compared with the total cost in the standard cost information. When the difference between the two exceeds a preset threshold, the initial hypothesis test is deemed invalid, and there is an anomaly in the current doctor's disease resource consumption information. The comparison results between the disease-specific resource consumption information and the standard cost information are generated based on the average medical expenses of doctors.

[0016] The above-mentioned scheme, by obtaining the detailed average medical expenses of different doctors in the department for abnormal diseases and comparing them with standard expense information, can accurately identify which doctors' resource consumption for certain diseases is abnormal, facilitating targeted management and intervention. Using hypothesis testing, with the initial assumption that the average medical expenses of doctors are no different from the average standard expenses, the scheme uses difference testing to determine whether the hypothesis is valid, providing a scientific and objective basis for judging whether doctors' expense consumption is abnormal and avoiding errors from subjective judgment. Furthermore, by generating difference comparison results based on the averages of various medical expenses, the scheme can analyze the reasons for cost differences in more detail, helping to gain a deeper understanding of the use of medical resources across different projects.

[0017] In one implementation, a disease-specific cost optimization strategy for the next period is generated based on the difference comparison results, specifically including: Based on the comparison results, cost control results are generated for each doctor. The cost control results are sorted, and a preset number of doctors with the highest rankings are selected to generate a doctor cost control focus list. Based on the difference comparison results and cost control results, cost control results for each abnormal disease are generated. The cost control results are sorted, and a preset number of diseases with the highest ranking are selected to generate a list of diseases for cost control attention. Based on the preset cost control time period and cost control results, generate phased cost optimization targets; The cost optimization strategy for the next period is generated based on the physician cost control focus list, the disease cost control focus list, and the phased cost optimization goals. The cost optimization strategy includes real-time monitoring of the disease resource consumption of each physician in the physician cost control focus list in the next period based on the phased cost optimization goals.

[0018] The above scheme generates lists of physicians and diseases requiring special attention for cost control, respectively identifying those that need focused attention. This allows management resources to be concentrated on key areas, improving the targeting and efficiency of cost control. Based on the cost control results and a pre-set timeframe, phased cost optimization targets are generated, providing clear direction and measurable standards for cost optimization, facilitating gradual progress and long-term cost control. Finally, the next phase of cost optimization strategies is formulated by integrating the physician and disease-specific cost control lists and the phased optimization targets. This comprehensive and systematic approach addresses cost control from both physician and disease perspectives simultaneously.

[0019] In one implementation, the method for optimizing disease-specific costs further includes: introducing an intelligent interaction step based on a large language model, used to receive natural language queries input by users, parse the elements such as disease, department, physician, time interval, and cost indicators, automatically identify the query intent and call the corresponding data analysis or model calculation function, and generate a natural language response result containing a three-layer structure of "conclusion layer, evidence layer, and suggestion layer", so as to realize intelligent interpretation and auxiliary decision-making for cost anomalies, cost control targets, and trend changes.

[0020] The above solution improves the ease of user interaction with the system, eliminating the need for complex queries; users can simply express their needs using natural language. Simultaneously, the structured response results present information more clearly, helping users quickly understand the cost details and make informed decisions.

[0021] In one implementation, the large language model is linked with the disease-specific cost model, cost control result database, and early warning knowledge base through a retrieval enhancement generation mechanism to achieve fusion reasoning of structured data and textual knowledge. The conclusion layer is used to output cost determination and trend results, the evidence layer is used to display the calculation basis and data references, and the suggestion layer is used to output cost control optimization prompts.

[0022] In the above approach, fusion reasoning enables large language models to comprehensively utilize multiple types of data, providing more comprehensive and accurate results. The structured output has a clear hierarchy, meeting the needs of users at different levels, from quickly understanding the results to delving into the underlying reasons and obtaining improvement suggestions.

[0023] In one implementation, the large language model is linked with the disease-specific standard cost model, cost control result database, and early warning knowledge base through a retrieval enhancement generation mechanism. This enables the fusion generation of natural language query and structured data analysis results, and can automatically match relevant calculation results and generate interpretive output based on disease, physician, and time interval.

[0024] In the above approach, adding an early warning knowledge base enriches the model's information sources, enabling it to provide more comprehensive explanations and early warning information. Personalized query responses can better meet users' specific needs, further improving the user experience.

[0025] In one implementation, the disease cost optimization method includes a task orchestration module, which dynamically calls between the anomaly detection, cost control calculation, trend analysis and report generation modules based on intent recognition results, and maintains the continuity and consistency of multi-turn question-and-answer semantics through a session-level context tracking mechanism.

[0026] In the above solution, the task orchestration module enables flexible invocation of various functional modules of the system, improving system efficiency and response speed. The session-level context tracking mechanism supports multi-turn question-and-answer sessions, facilitating in-depth user exploration of questions. The structured card-based output and explicit triggering mechanism make system interaction more intuitive and convenient.

[0027] Secondly, this application also provides a disease-specific cost optimization device, including: an information acquisition module, a cost calculation module, a difference comparison module, and a strategy generation module; The information acquisition module is used to acquire the current period's disease resource consumption information; among which, the disease resource consumption information includes the department's disease resource consumption information and the department's doctors' disease resource consumption information; The cost calculation module is used to input the current period's disease resource consumption information into a pre-built disease standard cost model to calculate the cost and obtain the corresponding standard cost information; the disease standard cost model is trained based on a historical disease resource consumption information sample set; The difference comparison module is used to test the difference between the department's disease resource consumption information and the standard cost information. The disease to which the department's disease resource consumption information belongs with the difference exceeding the preset threshold is determined as an abnormal disease. The module obtains the disease resource consumption information of the first department doctor under the abnormal disease and generates the difference comparison result between the disease resource consumption information of the first department doctor and the corresponding standard cost information. The strategy generation module is used to generate the next phase of disease cost optimization strategy based on the difference comparison results.

[0028] In the above scheme, a standard cost model for each disease, pre-built and trained based on historical data, is used to calculate costs by combining current disease resource consumption information. This yields standard cost information that closely reflects the actual situation, providing a scientific and accurate reference for subsequent cost management. A difference test is used to compare departmental disease resource consumption information with standard cost information, accurately identifying abnormal diseases where the difference exceeds a preset threshold. This allows for focusing on key issues and concentrating efforts on resolving cost anomalies. After identifying abnormal diseases, the resource consumption information of the doctors in the first department under that abnormal disease is further obtained and compared, deepening the analysis to the doctor level and clarifying the specific source of cost anomalies, making subsequent optimization strategies more targeted. Based on the difference comparison results, a disease cost optimization strategy for the next period is generated. Cost management measures are adjusted in a timely manner according to the actual situation to reasonably optimize disease costs, improve resource utilization efficiency, and enhance the level of medical cost control.

[0029] Thirdly, this application also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the above-mentioned disease cost optimization method.

[0030] Fourthly, this application also provides a computer-readable storage medium, which includes a stored computer program, wherein the computer program controls the device where the computer-readable storage medium is located to execute the above-mentioned disease cost optimization method when it is running. Attached Figure Description

[0031] Figure 1 This is a flowchart illustrating a method for optimizing disease-specific costs according to one embodiment of the present invention; Figure 2 This is a schematic diagram of a disease cost optimization device provided in one embodiment of the present invention. Detailed Implementation

[0032] The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and are not intended to limit the scope of the invention.

[0033] The terms "first" and "second," etc., in the specification, claims, and drawings of this application are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to such processes, methods, products, or apparatus.

[0034] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0035] Example 1 See Figure 1 , Figure 1 This is a flowchart illustrating a method for optimizing disease-specific costs according to an embodiment of the present invention. The embodiment includes steps 101 to 104, each step of which is detailed below: Step 101: Obtain the disease resource consumption information for this period; the disease resource consumption information includes the department's disease resource consumption information and the department's doctor's disease resource consumption information.

[0036] It should be noted that the specific time frame of "this period" can be one month, one quarter, or one year, determined according to the actual needs of hospital operation and management, and is not limited here. Meanwhile, the data collection sources mainly include the hospital's information management system, electronic medical record system, financial system, and material management system. In this embodiment of the invention, by obtaining detailed information on resource consumption by department and physician for each disease, the hospital can achieve refined management of various departments and physicians. It can analyze the differences in resource consumption among different departments when handling the same disease, identify departments with high resource utilization efficiency for experience promotion, and simultaneously optimize and improve departments with excessively high resource consumption. For individual physicians, it can assess the rationality of their treatment behavior and the efficiency of resource utilization, incentivizing physicians to use medical resources rationally and improve the quality of medical services.

[0037] In one embodiment, obtaining the current period's disease-specific resource consumption information specifically includes: obtaining the current period's inpatient medical record data; classifying each inpatient medical record data into its corresponding disease category, and calculating the medical resource consumption information under the disease category dimension and the physician dimension respectively, to obtain the department's disease-specific resource consumption information and the department's physician's disease-specific resource consumption information; wherein, the department's disease-specific resource consumption information represents the average medical cost of each case in the same disease category within the department, and the department's physician's disease-specific resource consumption information represents the average medical cost of different physicians in the same disease category within the department.

[0038] In this embodiment of the invention, after obtaining the inpatient medical record data for the current period, the data must first be preprocessed. This includes checking the completeness of the data and removing cases with missing key information (such as cost information, diagnostic information, etc.). Simultaneously, outliers in the data are processed; for example, by setting a reasonable cost threshold, cost data exceeding the normal range is corrected or removed to ensure the accuracy of subsequent calculations. To accurately classify each inpatient medical record data into its corresponding disease category, a unified disease classification standard is required. The International Classification of Diseases (ICD) coding can be referenced to standardize the diagnostic information in the medical records, ensuring that cases of the same disease can be accurately classified. Departmental disease resource consumption information can be calculated using the following formula: The cost indicators for the i-th disease in the department are calculated as follows (i=1, 2...n, where n is the number of diseases, the same below): (1) Total cost per case for disease i = ∑ cost of cases under disease i / number of cases under disease i; (2) Average drug cost per case of disease i = ∑ Drug cost per case of disease i / Number of cases of disease i; (3) Average consumable cost per case of disease i = ∑ Consumable cost per case of disease i / Number of cases of disease i; (4) Average examination and testing cost per case of disease i = ∑ Examination and testing cost of cases under disease i / Number of cases under disease i; (5) Average medical service cost per case of disease i = ∑ Medical service cost of cases under disease i / Number of cases under disease i; The resource consumption information for each disease by which doctors in a department are treated can be calculated using the following formula: The cost indicators for the j-th physician under the i-th disease in the department are calculated as follows (j=1, 2...m, where m is the number of physicians, the same below); (1) Average total cost per case for physician j = ∑ cost of cases under physician j / number of cases under physician j; (2) Average drug cost per case for physician j = ∑ Drug cost per case under physician j / Number of cases under physician j; (3) Average cost of consumables per case for physician j = ∑ cost of consumables for cases under physician j / number of cases under physician j; (4) Average examination and testing cost per case for physician j = ∑ Examination and testing cost of cases under physician j / Number of cases under physician j; (5) Average medical service fee per case for physician j = ∑ Medical service fee for cases under physician j / Number of cases under physician j

[0039] Preferably, in addition to calculating the average total cost per case, average drug cost per case, average consumable cost per case, average examination and testing cost per case, and average medical service cost per case, the cost data can be further refined. For example, in drug costs, the cost of Western medicine and traditional Chinese medicine can be distinguished; in examination and testing costs, the cost of different types of examination items can be subdivided (such as imaging examination costs, laboratory test costs, etc.).

[0040] Furthermore, the calculated information on resource consumption for different diseases and doctors within the department can be visualized, for example, using bar charts, line graphs, and other charts to intuitively present the cost differences between different diseases and different doctors, making it easier for hospital administrators to analyze and make decisions.

[0041] Example of departmental disease resource consumption information for this period:

[0042] Example of departmental physician disease resource consumption information for this period:

[0043] For example, suppose we want to calculate the resource consumption of a hospital for a specific disease in the first half of this year. We obtain the inpatient medical records for the first half of this year from the hospital's information management system, including basic patient information, diagnostic information, and cost information. Taking "coronary heart disease" as an example, suppose Department 1 admitted 100 patients with "coronary heart disease" in the first half of this year, with a total medical record cost of 2 million yuan, total drug costs of 500,000 yuan, total consumable costs of 300,000 yuan, total examination and testing costs of 600,000 yuan, and total medical service costs of 600,000 yuan. Then the department's resource consumption information for this disease would be: The average total cost per case for the disease "coronary heart disease" is 2 million yuan / 100 = 20,000 yuan; The average drug cost per case for the disease "coronary heart disease" is 500,000 yuan / 100 = 5,000 yuan. The average cost of consumables per case for the disease "coronary heart disease" is 300,000 yuan / 100 = 3,000 yuan; The average examination and testing cost per case for the disease "coronary heart disease" is 600,000 yuan / 100 = 6,000 yuan. The average medical service cost per case for the disease "coronary heart disease" is 600,000 yuan / 100 = 6,000 yuan; Assuming that among these 100 patients with coronary heart disease, Doctor 1 treated 30 cases, with a total case cost of 630,000 yuan, total medication cost of 160,000 yuan, total consumable cost of 90,000 yuan, total examination and testing cost of 180,000 yuan, and total medical service cost of 200,000 yuan, then the department's doctor's resource consumption information for each disease is as follows: The average total cost per doctor per case = 630,000 yuan / 30 = 21,000 yuan; The average cost of medication per doctor per case is approximately 160,000 yuan / 30 ≈ 5,300 yuan. The average cost of consumables per doctor per case = 90,000 yuan / 30 = 3,000 yuan; The average cost of examinations and tests per doctor per case is 180,000 yuan / 30 = 6,000 yuan. The average medical service fee per doctor per case is approximately 200,000 yuan / 30 = 6,700 yuan.

[0044] Step 102: Input the current period's disease resource consumption information into the pre-built disease standard cost model to calculate the cost and obtain the corresponding standard cost information; wherein, the disease standard cost model is trained based on a historical disease resource consumption information sample set.

[0045] In this embodiment of the invention, after obtaining the current period's disease resource consumption information, this information is input into a disease-specific standard cost model pre-trained using a historical disease resource consumption information sample set for cost calculation, ultimately yielding the corresponding standard cost information. The standard cost can serve as a reference for evaluating the quality of medical services. When the actual cost differs significantly from the standard cost, it can reflect issues such as over-treatment, resource waste, or substandard medical service quality, facilitating timely improvements.

[0046] In one embodiment, the method for optimizing disease-specific costs further includes: constructing a historical sample set of disease-specific resource consumption information, specifically: acquiring the previous period's inpatient medical record data; wherein, the inpatient medical record data includes basic patient information, patient disease characteristics, medical resource characteristics, and patient cost data; preprocessing the inpatient medical record data; wherein, the preprocessing includes removing cases with missing or abnormal data and removing cases with abnormal patient costs; classifying each inpatient medical record data into its corresponding disease category, and calculating the standard deviation of medical costs for each case under each disease category; wherein, the standard deviation of medical costs is based on the difference between the medical cost of each case and the mean medical cost of each case under its corresponding disease category; removing inpatient medical record data whose standard deviation of medical costs exceeds a preset standard deviation threshold, thereby obtaining a historical sample set of disease-specific resource consumption information for each disease category; wherein, the number of inpatient medical record data under each disease category in the historical sample set of disease-specific resource consumption information exceeds a preset threshold.

[0047] In this embodiment of the invention, inpatient medical record data from the previous period is acquired. This data includes basic patient information, patient disease characteristics, medical resource characteristics, and patient cost data. Basic patient information includes patient ID (unique patient identifier), medical record number (unique identifier for inpatient medical records), discharge date, gender, age, and medical insurance type. Patient disease characteristics include primary diagnosis, secondary diagnosis, primary procedure, secondary procedure, and admission route. Medical resource characteristics include discharge department, attending physician, discharge method, length of stay, primary surgery, other surgeries, high / low magnification case identifiers (high magnification case, low magnification case, normal case), whether admitted to the intensive care unit, and number of resuscitation attempts. Patient cost data includes total cost, medication costs, examination and testing costs, medical service costs, and consumable costs. The collected inpatient medical record data undergoes certain preprocessing. When removing cases, cases with missing fields can be simply removed initially. For missing important fields, more complex methods can be used for processing. For example, when patient age is missing, patients can be grouped according to their medical insurance type and the department they were admitted to, and then the average age of each group can be used to fill in the gaps. When the primary diagnosis is missing, it can be inferred and supplemented by combining information such as secondary diagnoses and major procedures. Then, the patient's expenses on the medical record front page are compared with the total expenses in the patient's inpatient billing details. Cases with a discrepancy greater than 500 yuan are removed, and the reasons for the discrepancy can be further analyzed. If the discrepancy is due to different billing item classifications, a unified adjustment can be made; if it is a data entry error, it should be corrected promptly. Simultaneously, for the removal of high and low multiplier cases, different multiplier thresholds can be set according to the characteristics of different diseases. After data preprocessing, disease grouping is defined using the primary diagnosis category, and each case is assigned to the corresponding disease group. Preferably, more detailed grouping can be performed by combining information such as the patient's disease characteristics and treatment methods. For example, for patients with the same primary diagnosis, the disease grouping can be further subdivided according to the disease stage and treatment method (surgery, conservative treatment, etc.) to make the grouping more reasonable and accurate. After classifying each case into its corresponding disease group, the Z-score method is used to calculate the standard deviation of medical costs for each case within each disease group. Specifically, z = (x - μ) / σ, where x is the cost of the case, μ is the mean cost of the disease to which the case belongs, and σ is the standard deviation of costs for the disease to which the case belongs. Cases with |z| > 3 are removed. Preferably, for some diseases with large cost fluctuations, the threshold can be appropriately increased; for diseases with relatively stable costs, the threshold can be decreased. In addition, other anomaly detection methods, such as machine learning-based anomaly detection algorithms, can be used to further improve the accuracy of identifying abnormal cases. It should be noted that, in order to ensure that there is sufficient case data for each disease group for subsequent data processing, the historical disease resource consumption information sample set for each disease group needs to retain at least 100 cases of case data.Preferably, the threshold for the number of cases can be adjusted based on factors such as the importance and incidence of the disease. For some rare diseases, the requirement for the number of cases can be appropriately lowered.

[0048] In one embodiment, the standard cost model for diseases is trained based on a sample set of historical disease resource consumption information. The standard cost model for diseases is trained in the following way: feature data and label data are established for the data content of each sample set of historical disease resource consumption information. The number of neurons, d, is set based on the amount of feature data, and the number of hidden layers is set to 2. Construct the network topology and set the standardized feature vectors as X∈R d (X=[x1,x2,…,x) d ] T ).

[0049] Forward propagation is used and the output of each layer is set, specifically: Output of the first hidden layer: H1 = σ(W1X + b1) (W1 ∈ R) n1×d Let b1 ∈ R be the weight matrix. n1 (for bias vectors) Output of the second hidden layer: H2 = σ(W2H1 + b2) (W2 ∈ R) n2×n1 Let b2 be the weight matrix. Rn2 (for bias vectors) Output layer predictions: =W3H2+b3(W3∈ R1×n2 Let b3 be the weight matrix and b3 ∈ R be the bias vector. (Forecast costs); The historical disease resource consumption information sample set is divided into a training set and a validation set according to a preset ratio. The network topology is iteratively trained based on the training set, and the parameters of the network topology are updated until the optimal loss function is obtained. The loss function is obtained based on the difference between the predicted and actual values ​​of the labeled data, and its expression is: ; Where N is the number of samples, Yi is the actual cost of the i-th sample, and λ is the regularization coefficient; The average performance of the network topology is verified several times based on the verification set, and its coefficient of determination is calculated; wherein the expression for the coefficient of determination is: ; in As the coefficient of determination, This represents the average of the actual costs. When the determination coefficient exceeds a preset threshold, the training of the standard cost model for the current disease is considered complete.

[0050] In this embodiment of the invention, for each historical disease resource consumption information sample set, relevant features need to be extracted from the data content as feature data (e.g., patient age, gender, disease type, treatment duration, etc.), and a target value needs to be determined as label data (i.e., disease cost). This transforms the original data into a format suitable for model training. Preferably, during feature data extraction, certain feature processing can be performed, such as encoding categorical features (e.g., one-hot encoding) and normalizing numerical features to improve the model's training effect. Then, the data index is randomly rearranged to avoid the data order affecting model training. The data is then divided into a training set and a test set in a 7:3 ratio. The training set is used for model training, and the test set is used to evaluate the model's generalization ability. Stratified sampling can be used to ensure that the proportion of each category (different diseases) in the training set and the test set is consistent with the proportion in the original dataset. This avoids poor prediction performance for certain diseases due to uneven data distribution. Before starting iterative training, parameters are randomly generated using a standard normal distribution. This allows the parameters to have a certain degree of randomness at the initial stage, preventing the model from getting trapped in local optima. Furthermore, besides the standard normal distribution, other initialization methods can be used, such as Xavier initialization and He initialization. These methods allow for the selection of appropriate initialization methods based on different activation functions, which helps with model convergence. Alternatively, mean squared error (MSE) and loss functions can be used. The formula for calculating mean squared error (MSE) is... .in, This represents the actual cost of the disease (label data). The model predicts the cost of the disease (predicted value). By minimizing the loss function, the model's predicted value can be as close as possible to the true value. Alternatively, other loss functions can be considered, such as Mean Absolute Error (MAE) and Huber loss. These loss functions have different sensitivities to outliers, and the appropriate loss function can be selected based on the characteristics of the data. Then, gradient descent is used to update the model parameters. Gradient descent is an iterative optimization algorithm that calculates the gradient of the loss function with respect to parameters a and b, and then updates the parameters in the opposite direction of the gradient, causing the loss function to gradually decrease. Then, when training the model, the number of iterations is set to epoch=50. One epoch represents training the entire training dataset once. Setting epoch=50 means the model will be trained on the training dataset for 50 iterations. During training, the number of epochs can be dynamically adjusted based on the model's convergence. For example, during training, the change in the loss function can be monitored. If the loss function no longer decreases or decreases very little after a certain epoch, training can be stopped early to avoid overfitting. For the inner loop, the learning rate is set to Ir=0.001. The learning rate controls the step size of parameter updates; lr=0.001 indicates that the parameter update increment is 0.001. During forward computation, the network is initialized with 12 features, meaning there are 12 input features. Furthermore, a learning rate decay strategy can be employed, gradually decreasing the learning rate during training. This avoids excessively large parameter update increments when the model approaches its optimal solution, improving model stability. Simultaneously, different feature numbers can be experimented with to select the optimal feature combination. After each epoch of training, the loss function values ​​for the training and test sets are calculated to evaluate the model's performance on those sets. By iterating through different learning rates, the optimal learning rate and loss function that minimize and approximate the loss function values ​​for the training and test sets are found. For example, grid search or random search methods can be used to find the optimal learning rate more efficiently. The training process and evaluation metrics for each learning rate can be recorded for subsequent analysis. The updated spatial function parameters are validated using the test set, and the coefficient of determination is used to determine the optimal learning rate. A score ≥0.8 indicates that the standard cost model for the current disease has completed training. Validation aims to ensure the model has good generalization ability on unseen data. Furthermore, cross-validation can be used to further divide the training set into multiple subsets for multiple training and validation runs, improving the reliability of the validation results. The optimal model for each disease group is trained and stored as the standard cost model. The stored model can be directly loaded for subsequent use to predict costs for new diseases. Preferably, when storing the model, in addition to saving the model parameters, information such as the model structure, hyperparameter settings during training, and evaluation metrics can also be saved for subsequent model analysis and optimization.

[0051] In one embodiment, feature data and tag data are established for the data content of each historical disease resource consumption information sample set. Specifically, the feature data includes: patient gender, age, medical insurance type, primary diagnosis, admission route, discharge method, treatment method, primary surgery, surgical level, number of operations, whether admitted to the intensive care unit, and number of resuscitations; the tag data includes: total cost, drug cost, examination and testing cost, medical service cost, and consumable cost.

[0052] In this embodiment of the invention, a feature data and tag data system is constructed based on a sample set of historical disease resource consumption information. The feature data covers basic patient characteristics (gender, age), medical insurance related information (medical insurance type), and disease diagnosis and treatment process information (primary diagnosis, admission route, discharge method, treatment method, primary surgery, surgical level, number of procedures, whether admitted to the intensive care unit, number of resuscitations). These features can characterize the patient's medical treatment from different dimensions. The tag data focuses on different components of medical expenses (total cost, drug cost, examination and testing cost, medical service cost, and consumable cost), used to measure the economic cost of medical resource consumption. Comprehensive feature data allows the model to more accurately capture factors affecting medical expenses, thereby accurately predicting various costs and helping hospitals and medical insurance departments plan resources and funds in advance.

[0053] Step 103: Perform a difference test between the departmental disease resource consumption information and the standard cost information. Determine the disease to which the departmental disease resource consumption information with a difference exceeding a preset threshold belongs to an abnormal disease. Obtain the disease resource consumption information of the first department doctor under the abnormal disease and generate a difference comparison result between the disease resource consumption information of the first department doctor and the corresponding standard cost information.

[0054] In this embodiment of the invention, a discrepancy test is first performed on the resource consumption information and standard cost information for each department's disease category. Diseases with discrepancies exceeding a preset threshold are identified as abnormal diseases. Next, the resource consumption information of the doctor in the first department under the abnormal disease category is obtained. Finally, a comparison result of these discrepancies with the corresponding standard cost information is generated. Through discrepancy testing and threshold judgment, abnormal diseases with significant discrepancies between resource consumption and standard costs in a department can be identified in a timely manner, helping hospitals to focus on potentially problematic medical service items, such as over-treatment, resource waste, or unreasonable charges. Furthermore, focusing on the resource consumption information of the doctor in the first department under the abnormal disease category allows the problem to be refined to specific doctors and diseases, facilitating hospital management to accurately identify the responsible party and deeply analyze the reasons for the abnormal costs.

[0055] The departmental disease resource consumption information is compared with the standard cost information. Diseases to which the departmental disease resource consumption information with discrepancies exceeding a preset threshold are identified as abnormal diseases. The specific implementation process is as follows: Hypothesis is established. The first hypothesis is that the departmental resource consumption information for each disease type means that the average medical cost per case within the department for the same disease type is not significantly different from the total cost predicted by the standard cost model. This implies that, if the null hypothesis holds, the actual resource consumption costs for the current departmental diseases are consistent with the costs predicted by the standard cost model, and there are no significant anomalies. Alternative hypothesis The hypothesis is that the actual cost of a disease within the department differs from the total cost predicted by the standard cost model. If the alternative hypothesis is true, it indicates a significant deviation between the actual cost and the standard cost for that disease, potentially leading to over-treatment, resource waste, or unreasonable charges. A two-sample t-test is then used to compare the current cost with the total cost calculated by the standard cost model. The t-test is a commonly used statistical method to determine if there is a significant difference between the means of two sets of data. In this scenario, one set of data represents the actual cost of the disease incurred by the department during the current period, and the other set represents the cost calculated by the standard cost model. The significance level α = 0.5 is a pre-set significance level, representing the upper probability of incorrectly rejecting the null hypothesis if it is true; that is, allowing a 5% chance of incorrectly concluding that the two sets of data are different. The p-value is an important output of the t-test, representing the probability of obtaining the current sample data or more extreme data if the null hypothesis is true. When p < .5, it means that, assuming the null hypothesis is true, the probability of obtaining the current sample data is extremely small, lower than our pre-set error tolerance of 5%. Therefore, we have sufficient reason to reject the null hypothesis, considering that the mean cost of this disease differs significantly from the standard cost model, and classify it as an abnormal disease, including it in subsequent in-depth analysis. When p ≥ .5, it indicates that, assuming the null hypothesis is true, obtaining the current sample data is reasonable. We do not have sufficient evidence to reject the null hypothesis, considering that the mean cost of this disease does not differ significantly from the standard cost model. This disease does not need to be included in the abnormal disease analysis, and its cost structure is considered normal.

[0056] In one embodiment, the method involves obtaining the resource consumption information of the first department doctor under the abnormal disease category and generating a comparison result between the resource consumption information of the first department doctor and the corresponding standard cost information. Specifically, this includes: obtaining the resource consumption information of the first department doctor under the abnormal disease category; wherein, the resource consumption information of the first department doctor is the average medical cost of different doctors in the department under the abnormal disease category, and the average medical cost includes the average total cost per case, the average drug cost per case, the average examination and testing cost per case, the average medical service cost per case, and the average consumable cost per case; establishing an initial hypothesis test based on the average medical cost of each doctor; wherein, the initial hypothesis test is that there is no difference between the average medical cost of the doctor and the average standard cost; performing a difference test on the average total cost per case of each doctor and the total cost in the standard cost information, and when the difference between the two exceeds a preset threshold, determining that the initial hypothesis test is not valid and that the current doctor's resource consumption information for the disease category is abnormal; and generating a comparison result between the resource consumption information for the disease category and the standard cost information based on the average medical cost information of the doctor.

[0057] In this embodiment of the invention, the average medical expenses of different doctors in the department under abnormal diseases are obtained. These averages include the average total cost per case, the average drug cost per case, the average examination and testing cost per case, the average medical service cost per case, and the average consumable cost per case. These average costs can comprehensively reflect the resource consumption of each doctor when treating abnormal diseases. An initial hypothesis test is established for each doctor, with the null hypothesis set as no difference between the average medical expenses of doctors and the average standard cost. The average total cost per case of each doctor is compared with the total cost in the standard cost information for a difference test. In specific implementation, a two-independent-samples t-test is used, with a test significance level of α=0.05. The t-test statistic and p-value results are observed to make a judgment. When p<0.05, it means that if the null hypothesis is true, the probability of obtaining the current sample data or more extreme data is very small, lower than our pre-set error tolerance of 5%, so the null hypothesis is rejected, the initial hypothesis test is determined to be invalid, that is, the current doctor's disease resource consumption information is considered abnormal. Conversely, when p ≥ 0.05, it indicates insufficient evidence to reject the null hypothesis, concluding that the doctor's mean cost is not significantly different from the standard cost model. An example, the resulting comparison table is shown below:

[0058] Based on the average medical expenses of doctors, a comparison result is generated between the resource consumption information for each disease and the standard cost information. This result is usually presented in tabular form, such as the given output difference comparison result table, which includes information such as disease, department, doctor, average cost per case, p-value test result, and whether it is abnormal. Through this table, you can intuitively see the difference between each doctor's cost item and the standard cost, and whether it is marked as abnormal.

[0059] Step 104: Generate the next phase of disease cost optimization strategy based on the difference comparison results.

[0060] In one embodiment, generating a disease-specific cost optimization strategy for the next period based on the difference comparison results specifically includes: generating cost control results for each doctor based on the difference comparison results, sorting the cost control results, and selecting a predetermined number of doctors with the highest rankings to generate a doctor cost control focus list; generating cost control results for each abnormal disease based on the difference comparison results and the cost control results, sorting the cost control results, and selecting a predetermined number of diseases with the highest rankings to generate a disease cost control focus list; generating a phased cost optimization target based on a predetermined cost control time period and cost control results; and generating a cost optimization strategy for the next period based on the doctor cost control focus list, the disease cost control focus list, and the phased cost optimization target; wherein, the cost optimization strategy includes real-time monitoring of the disease-specific resource consumption of each doctor in the doctor cost control focus list in the next period based on the phased cost optimization target.

[0061] In this embodiment of the invention, for physicians marked as abnormal, the cost control result for each physician is calculated according to the formula "Physician Cost Control Space = (Physician's Average Total Cost per Case in the Current Period - Standard Average Total Cost per Case) * Number of Physician Cases". This indicator measures the potential cost control space of each abnormal physician, that is, the total amount of cost that can be saved if the physician's cost can reach the standard level. The cost control spaces of all abnormal physicians are sorted in descending order, and a predetermined number of physicians with the highest rankings are selected. These physicians are the key targets for cost control and are included in the physician cost control focus list, thereby prioritizing the management and supervision of physicians with large cost control spaces to achieve more significant cost optimization effects. For diseases marked as abnormal, their cost control space is equal to the sum of the cost control spaces of all abnormal physicians under that disease, that is, "Disease Cost Control Space = ∑ Physician Cost Control Space". This reflects the potential of each abnormal disease in terms of overall cost control. For example, the cost control space (results) table for abnormal physicians is shown below:

[0062] All abnormal disease categories are sorted by cost control potential in descending order. A predetermined number of diseases at the top of the list are selected as the key diseases requiring focused cost control and added to the disease cost control focus list. By focusing on these diseases with large cost control potential, targeted optimizations can be made to related diagnosis and treatment processes and resource utilization. Then, an optimization period M is set, representing the plan to steadily control the costs of abnormal physicians within the standard cost range over the next M months. This sets a timeframe for cost optimization, ensuring that cost adjustments are implemented gradually and avoiding excessive impact on the quality of medical services. The formula "Physician's target average cost per case in month N = Physician's average cost per case - (Physician's average cost per case - Standard average cost per case) / Number of months to achieve the target" is used to calculate the target cost per case. Calculate the target average cost per case for each physician with abnormal costs each month. This provides physicians with clear cost control targets, guiding them gradually towards standard costs. An example of a phased cost optimization target is as follows:

[0063] Based on the physician cost control focus list, the disease-specific cost control focus list, and the phased cost optimization goals, a cost optimization strategy for the next period is formulated. The core of this strategy is to monitor the disease-specific resource consumption of each physician on the physician cost control focus list in real time, based on the phased cost optimization goals. Furthermore, to promptly identify anomalies in the cost control process, certain early warning rules can be set. For example: if the physician's average cost per case is not higher than the target average cost per case, the target is met; if the physician's average cost per case is within 5% of the target average cost per case, it is within the normal fluctuation range; if it is 5%-10% higher than the target average cost per case, a mild warning is issued, indicating the need to monitor cost changes; if it is more than 10% higher than the target average cost per case, a severe warning is issued, indicating a significant deviation in cost control, requiring stricter intervention measures.

[0064] Continuous monitoring of physicians' achievement of various cost indicators is conducted, and the slope of cost trends during the dynamic monitoring period is used to assess the effectiveness of cost control improvements. For example, the Theil-Senestimator method is used to observe the slope of the fitted curve for the average cost per physician during the period. The following judgments are made based on the slope: a slope between -0.3 and 0.3 indicates relatively stable cost changes; a slope less than -0.3 indicates a downward trend in costs, signifying improved cost control; and a slope greater than 0.3 indicates an upward trend in costs, indicating poor cost control effectiveness. The physician cost control dynamic monitoring table is shown below:

[0065] The dynamic monitoring table above provides a clear understanding of each physician's cost control results, allowing for assessment of the effectiveness of cost control improvements and timely adjustments to cost optimization strategies.

[0066] In this embodiment of the invention, a disease-specific cost optimization device is also provided, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the above-described disease-specific cost optimization method.

[0067] In one embodiment, the method for optimizing disease-specific costs further includes: introducing an intelligent interaction step based on a large language model, which is used to receive natural language queries input by users, parse the elements such as disease, department, physician, time interval and cost indicators, automatically identify the query intent and call the corresponding data analysis or model calculation function, and generate a natural language response result containing a three-layer structure of "conclusion layer, evidence layer and suggestion layer", so as to realize intelligent interpretation and auxiliary decision-making for cost anomalies, cost control targets and trend changes.

[0068] In this embodiment of the invention, based on the disease cost optimization method described above, an intelligent interactive function based on a large language model is introduced. Users can use natural language to query, and the large language model will parse the key elements in the query, such as disease, department, physician, time interval and cost indicators. Then, according to the user's intent category (query, alarm explanation, optimization plan, trend analysis), the corresponding abnormal result prompts (used to answer questions about abnormal cost situations), cost control target calculation (providing users with relevant information on physician cost control targets) and trend quality judgment (evaluating and judging the trend of physician cost changes) are fed back, generating output content with a three-layer structure of "conclusion layer - evidence layer - suggestion layer": (1) Conclusion information: clearly give key conclusions such as whether physician costs are abnormal, the size of the difference from the standard cost, whether the standard is met, and the quality of the trend, so that users can quickly understand the core situation; (2) Evidence information: provide detailed data such as comparison mean, P-value test, target threshold, trend slope as evidence to support the conclusion, and enhance the credibility of the results; (3) give stage goals and improvement suggestions based on the analysis results, and provide users with practical optimization directions. Furthermore, when parsing natural language queries, more advanced natural language processing techniques can be employed, such as fine-tuning pre-trained language models, to improve the accuracy of feature extraction. For query intent identification, an intent classification database can be established, and classification rules can be continuously updated and optimized to adapt to more diverse query needs.

[0069] In one embodiment, the large language model is linked with the disease standard cost model, cost control result database and early warning knowledge base through a retrieval enhancement generation mechanism to achieve the fusion reasoning of structured data and text knowledge; wherein, the conclusion layer is used to output cost determination and trend results, the evidence layer is used to display the calculation basis and data references, and the suggestion layer is used to output cost control optimization prompts.

[0070] In this embodiment of the invention, a retrieval enhancement generation mechanism is used to link with the disease-specific standard cost model and the cost control result database. This linkage enables the fusion reasoning of structured data (such as cost data in the database) and textual knowledge (such as cost-related rules and explanations). The conclusion layer outputs cost determination and trend results, allowing users to quickly understand the core information; the evidence layer displays the calculation basis and data references, increasing the credibility of the results; and the suggestion layer provides cost control optimization tips, offering users specific directions for improvement.

[0071] In one embodiment, the large language model, through a retrieval enhancement generation mechanism, links with the disease-specific standard cost model, cost control result database, and early warning knowledge base to achieve the fusion generation of natural language query and structured data analysis results. It can automatically match relevant calculation results and generate interpretive output based on disease, physician, and time interval.

[0072] In this embodiment of the invention, the scope of the large language model's linkage is further expanded by adding an early warning knowledge base. Through a Retrieval Enhanced Generation (RAG) mechanism, the fusion of natural language queries and structured data analysis results is achieved. The model can automatically match relevant calculation results based on disease type, physician, and time interval, and generate interpretive output to provide users with more personalized query responses. Furthermore, the early warning knowledge base can be regularly updated and maintained to ensure the accuracy and effectiveness of its information. Machine learning algorithms can be introduced during the automatic matching of calculation results to improve the accuracy and efficiency of the matching. In one embodiment, the disease cost optimization method includes a task orchestration module, which is used to dynamically call between the anomaly detection, cost control calculation, trend analysis and report generation modules based on intent recognition results, and maintain the continuity and consistency of multi-turn question-and-answer semantics through a session-level context tracking mechanism.

[0073] In this embodiment of the invention, the task orchestration module can dynamically call upon the anomaly detection, cost control calculation, trend analysis, and report generation modules based on the intent recognition results. Simultaneously, a conversation-level context tracking mechanism maintains the continuity and consistency of multi-turn question-and-answer semantics. The process of establishing the AI ​​intelligent question-and-answer assistant details each step of the algorithm, including intent recognition and parameter extraction, task orchestration, result fusion and generation, and dialogue tracking. The system returns results in a structured card format, containing conclusion information, evidence information, and optimization suggestions. Furthermore, when a user clicks on an anomaly alert or actively inquires about cost control progress, the system responds accordingly and automatically pushes anomaly alert cards based on the warning results. Further, a priority mechanism can be introduced in the task orchestration module to sort tasks according to their urgency and importance. For the dialogue tracking mechanism, more advanced memory techniques, such as Long Short-Term Memory (LSTM) networks, can be employed to improve the accuracy and persistence of contextual memory.

[0074] Label 1, methods for optimizing disease-specific costs, including: Obtain the current period's disease resource consumption information; wherein, the disease resource consumption information includes departmental disease resource consumption information and departmental physician disease resource consumption information; input the current period's disease resource consumption information into a pre-constructed disease standard cost model for cost calculation to obtain the corresponding standard cost information; wherein, the disease standard cost model is trained based on a historical disease resource consumption information sample set; perform a difference test between the departmental disease resource consumption information and the standard cost information, and determine the disease to which the departmental disease resource consumption information with a difference exceeding a preset threshold belongs to an abnormal disease, obtain the first departmental physician disease resource consumption information under the abnormal disease, and generate a difference comparison result between the first departmental physician disease resource consumption information and the corresponding standard cost information; generate a disease cost optimization strategy for the next period based on the difference comparison result.

[0075] According to label 2, based on label 1, obtaining the current period's disease resource consumption information specifically includes: obtaining the current period's inpatient medical record data; classifying each inpatient medical record data into its corresponding disease category, and calculating the medical resource consumption information under the disease category dimension and the physician dimension respectively, to obtain the department's disease resource consumption information and the department's physician's disease resource consumption information; wherein, the department's disease resource consumption information represents the average medical cost of each case in the same disease category within the department, and the department's physician's disease resource consumption information represents the average medical cost of different physicians in the same disease category within the department.

[0076] Based on label 1, the method for optimizing disease-specific costs further includes: constructing a historical disease-specific resource consumption information sample set, specifically: acquiring the previous period's inpatient medical record data; wherein the inpatient medical record data includes basic patient information, patient disease characteristics, medical resource characteristics, and patient cost data; preprocessing the inpatient medical record data; wherein the preprocessing includes removing cases with missing or abnormal data and removing cases with abnormal patient costs; classifying each inpatient medical record data into its corresponding disease category, and calculating the standard deviation of medical costs for each case under each disease category; wherein the standard deviation of medical costs is based on the difference between the medical cost of each case and the mean medical cost of each case under the corresponding disease category; removing inpatient medical record data whose standard deviation of medical costs exceeds a preset standard deviation threshold, thereby obtaining a historical disease-specific resource consumption information sample set corresponding to each disease category; wherein the number of inpatient medical record data under each disease category in the historical disease-specific resource consumption information sample set exceeds a preset threshold.

[0077] Label 4, based on label 1, describes a standard cost model for specific diseases that is trained using a historical sample set of disease resource consumption information. This standard cost model is trained in the following manner: Feature data and label data are established for the data content of each historical disease resource consumption information sample set; the number of neurons d is set based on the number of feature data, and the number of hidden layers is set to 2; the network topology is constructed, and the standardized feature vector is set as X∈R d (X=[x1,x2,…,x) d ] T ).

[0078] Forward propagation is used and the output of each layer is set, specifically: Output of the first hidden layer: H1 = σ(W1X + b1) (W1 ∈ R) n1×d Let b1 ∈ R be the weight matrix. n1 (for bias vectors) Output of the second hidden layer: H2 = σ(W2H1 + b2) (W2 ∈ R) n2×n1 Let b2 be the weight matrix. Rn2 (for bias vectors) Output layer predictions: =W3H2+b3(W3∈ R1×n2 Let b3 be the weight matrix and b3 ∈ R be the bias vector. (Forecast costs); The historical disease resource consumption information sample set is divided into a training set and a validation set according to a preset ratio. The network topology is iteratively trained based on the training set, and the parameters of the network topology are updated until the optimal loss function is obtained. The loss function is obtained based on the difference between the predicted and actual values ​​of the labeled data, and its expression is: ; Where N is the number of samples, Yi is the actual cost of the i-th sample, and λ is the regularization coefficient; The average performance of the network topology is verified several times based on the verification set, and its coefficient of determination is calculated; wherein the expression for the coefficient of determination is: ; in As the coefficient of determination, This represents the average of the actual costs. When the determination coefficient exceeds a preset threshold, the training of the standard cost model for the current disease is considered complete.

[0079] Based on label 4, the establishment of feature data and label data for the data content of each historical disease resource consumption information sample set is as follows: The feature data includes patient gender, age, medical insurance type, primary diagnosis, admission route, discharge method, treatment method, primary surgery, surgical level, number of operations, whether admitted to the intensive care unit, and number of resuscitations; The label data includes total cost, drug cost, examination and testing cost, medical service cost, and consumable cost.

[0080] According to label 6, based on label 5, the step of obtaining the disease resource consumption information of the first department doctor under the abnormal disease and generating a comparison result between the disease resource consumption information of the first department doctor and the corresponding standard cost information specifically includes: obtaining the disease resource consumption information of the first department doctor under the abnormal disease; wherein, the disease resource consumption information of the first department doctor is the average medical cost of different doctors in the department under the abnormal disease, and the average medical cost includes the average total cost per case, the average drug cost per case, the average examination and testing cost per case, the average medical service cost per case, and the average consumable cost per case; establishing an initial hypothesis test based on the average medical cost of each doctor; wherein, the initial hypothesis test is that the average medical cost of the doctor is no different from the average standard cost; performing a difference test on the average total cost per case of each doctor and the total cost in the standard cost information, and when the difference between the two exceeds a preset threshold, determining that the initial hypothesis test is not valid and that the disease resource consumption information of the current doctor is abnormal; generating a comparison result between the disease resource consumption information and the standard cost information based on the average medical cost information of the doctor.

[0081] According to label 7, based on label 6, the step of generating the next period's disease-specific cost optimization strategy based on the difference comparison results specifically includes: generating cost control results for each doctor based on the difference comparison results, sorting the cost control results, and selecting a predetermined number of doctors with the highest rankings to generate a doctor cost control focus list; generating cost control results for each abnormal disease based on the difference comparison results and the cost control results, sorting the cost control results, and selecting a predetermined number of diseases with the highest rankings to generate a disease-specific cost control focus list; generating a phased cost optimization target based on a predetermined cost control time period and the cost control results; and generating the next period's cost optimization strategy based on the doctor cost control focus list, the disease-specific cost control focus list, and the phased cost optimization target; wherein, the cost optimization strategy includes real-time monitoring of the disease-specific resource consumption of each doctor in the doctor cost control focus list in the next period based on the phased cost optimization target.

[0082] Label 8, in addition to any one of labels 1 to 7, also includes: An intelligent interaction step based on a large language model is introduced to receive natural language queries input by users, parse elements such as disease type, department, physician, time interval, and cost indicators, automatically identify the query intent and call the corresponding data analysis or model calculation function to generate a natural language response result with a three-layer structure of "conclusion layer, evidence layer, and suggestion layer" to achieve intelligent interpretation and decision support for cost anomalies, cost control targets, and trend changes.

[0083] Based on No. 8, the large language model, through the retrieval enhancement generation mechanism, links with the disease standard cost model, cost control result database, and early warning knowledge base to achieve the fusion reasoning of structured data and text knowledge; wherein, the conclusion layer is used to output cost determination and trend results, the evidence layer is used to display the calculation basis and data references, and the suggestion layer is used to output cost control optimization prompts.

[0084] Based on No. 8, the large language model described in No. 10, through the linkage of the retrieval enhancement generation mechanism with the disease standard cost model, cost control result database and early warning knowledge base, realizes the fusion generation of natural language query and structured data analysis results, and can automatically match relevant calculation results and generate interpretive output according to disease, physician and time interval.

[0085] Based on 8 or 10, the disease cost optimization method includes a task orchestration module, which is used to dynamically call between the anomaly detection, cost control calculation, trend analysis and report generation modules based on intent recognition results, and maintain the continuity and consistency of multi-turn question-and-answer semantics through a session-level context tracking mechanism.

[0086] In this embodiment of the invention, a computer-readable storage medium is also provided, which includes a stored computer program, wherein the computer program controls the device where the computer-readable storage medium is located to execute the above-described disease cost optimization method when it is running.

[0087] For example, a computer program can be divided into one or more modules, one or more of which are stored in memory and executed by a processor to perform the present invention. One or more modules can be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the disease-specific cost optimization device.

[0088] The device for optimizing costs based on disease type can be a desktop computer, laptop, handheld computer, or cloud server, among other computing devices. This device may include, but is not limited to, processors, memory, and displays. Those skilled in the art will understand that the above components are merely examples of devices for optimizing costs based on disease type and do not constitute a limitation on such devices. The device may include more or fewer components, combinations of certain components, or different components. For example, the device may also include input / output devices, network access devices, buses, etc.

[0089] The processor referred to can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the disease-specific cost optimization device, connecting all parts of the device through various interfaces and lines.

[0090] The memory can be used to store computer programs and / or modules. The processor implements various functions of the disease cost optimization device by running or executing the computer programs and / or modules stored in the memory, and by calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area. The program storage area can store the operating system, at least one application program required for a function (such as sound playback function, text conversion function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, text message data, etc.). In addition, the memory can include high-speed random access memory, and can also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0091] Among them, if the module based on optimizing the cost of specific diseases is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. Those skilled in the art can understand and implement this without any inventive effort.

[0092] This invention provides a method for optimizing disease-specific costs. By pre-constructing a standard cost model for diseases trained on historical data and combining it with current disease-specific resource consumption information, costs are calculated to obtain standard cost information that aligns with actual conditions, providing a scientific and accurate reference for subsequent cost management. A difference test is used to compare departmental disease-specific resource consumption information with standard cost information, thereby accurately identifying abnormal diseases where the difference exceeds a preset threshold, focusing on key issues, and facilitating concentrated efforts to resolve cost anomalies. After identifying abnormal diseases, further information on disease-specific resource consumption by the doctors in the first department under that abnormal disease is obtained and the differences are compared, deepening the analysis to the doctor level, clarifying the specific source of cost anomalies, and making subsequent optimization strategies more targeted. Based on the difference comparison results, a disease-specific cost optimization strategy for the next period is generated. Cost management measures are adjusted in a timely manner according to the actual situation to reasonably optimize disease-specific costs, improve resource utilization efficiency, and enhance the level of medical cost control.

[0093] Example 2 See Figure 2 , Figure 2 This is a schematic diagram of a disease-specific cost optimization device provided in one embodiment of the present invention. The embodiment of the present invention provides a disease-specific cost optimization device, including: an information acquisition module 201, a cost calculation module 202, a difference comparison module 203, and a strategy generation module 204; The information acquisition module 201 is used to acquire the disease resource consumption information for the current period; among which, the disease resource consumption information includes the department's disease resource consumption information and the department's doctor's disease resource consumption information; The cost calculation module 202 is used to input the current period's disease resource consumption information into the pre-built disease standard cost model to calculate the cost and obtain the corresponding standard cost information; wherein, the disease standard cost model is trained based on a historical disease resource consumption information sample set; The difference comparison module 203 is used to perform difference testing between the departmental disease resource consumption information and the standard cost information. The disease to which the departmental disease resource consumption information with a difference exceeding a preset threshold belongs is determined as an abnormal disease. The resource consumption information of the first department doctor under the abnormal disease is obtained, and the difference comparison result between the resource consumption information of the first department doctor and the corresponding standard cost information is generated. The strategy generation module 204 is used to generate the next period's disease cost optimization strategy based on the difference comparison results.

[0094] In one embodiment, the information acquisition module is used to acquire the current period's disease-specific resource consumption information, specifically including: acquiring the current period's inpatient medical record data; classifying each inpatient medical record data into its corresponding disease category, calculating the medical resource consumption information under the disease category dimension and the physician dimension respectively, to obtain the department's disease-specific resource consumption information and the department's physician's disease-specific resource consumption information; wherein, the department's disease-specific resource consumption information represents the average medical cost of each case in the same disease category within the department, and the department's physician's disease-specific resource consumption information represents the average medical cost of different physicians in the same disease category within the department.

[0095] In one embodiment, the disease-specific cost optimization device further includes: an information set construction module for constructing a historical disease-specific resource consumption information sample set, specifically: acquiring the previous period's inpatient medical record data; wherein, the inpatient medical record data includes patient basic information, patient disease characteristics, medical resource characteristics, and patient cost data; preprocessing the inpatient medical record data; wherein, the preprocessing includes removing cases with missing or abnormal data and removing cases with abnormal patient costs; classifying each inpatient medical record data into its corresponding disease category, and calculating the standard deviation of medical costs for each case under each disease category; wherein, the standard deviation of medical costs is based on the difference between the medical cost of each case and the mean medical cost of each case under its corresponding disease category; removing inpatient medical record data whose standard deviation of medical costs exceeds a preset standard deviation threshold, thereby obtaining a historical disease-specific resource consumption information sample set corresponding to each disease category; wherein, the number of inpatient medical record data under each disease category in the historical disease-specific resource consumption information sample set exceeds a preset threshold.

[0096] In one embodiment, the standard cost model for diseases is trained based on a historical sample set of resource consumption information for each disease. The standard cost model is trained as follows: feature data and label data are established for the data content of each historical sample set of resource consumption information for each disease; the number of neurons, d, is set based on the number of feature data, and the number of hidden layers is set to 2; a network topology is constructed, and the standardized feature vector is set as X∈R. d (X=[x1,x2,…,x) d ] T The forward propagation is used, and the output of each layer is set. Specifically, the output of the first hidden layer is: H1 = σ(W1X + b1) (W1 ∈ R). n1×d Let b1 ∈ R be the weight matrix. n1 (W2 is the bias vector); the output of the second hidden layer: H2=σ(W2H1+b2) (W2∈R) n2×n1 Let b2 be the weight matrix. Rn2 (for bias vectors); output layer predictions: =W3H2+b3(W3∈ R1×n2 Let b3 be the weight matrix and b3 ∈ R be the bias vector. (To predict costs), a historical disease resource consumption information sample set is divided into a training set and a validation set according to a preset ratio. The network topology is iteratively trained based on the training set, updating the parameters of the network topology until the optimal loss function is obtained. The loss function is obtained based on the difference between the predicted and actual values ​​of the labeled data, and its expression is: ; Where N is the number of samples, Yi is the actual cost of the i-th sample, and λ is the regularization coefficient; The average performance of the network topology is verified several times based on the verification set, and its coefficient of determination is calculated; wherein the expression for the coefficient of determination is: ; in As the coefficient of determination, This represents the average of the actual costs. When the determination coefficient exceeds a preset threshold, the training of the standard cost model for the current disease is considered complete.

[0097] In one embodiment, feature data and label data are established for the data content of each historical disease resource consumption information sample set, specifically as follows: Feature data includes patient gender, age, medical insurance type, primary diagnosis, admission route, discharge method, treatment method, primary surgery, surgical level, number of procedures, whether admitted to the intensive care unit, and number of resuscitation attempts; tag data includes total cost, drug cost, examination and testing cost, medical service cost, and consumable cost.

[0098] In one embodiment, the method involves obtaining the resource consumption information of the first department doctor under the abnormal disease category and generating a comparison result between the resource consumption information of the first department doctor and the corresponding standard cost information. Specifically, this includes: obtaining the resource consumption information of the first department doctor under the abnormal disease category; wherein, the resource consumption information of the first department doctor is the average medical cost of different doctors in the department under the abnormal disease category, and the average medical cost includes the average total cost per case, the average drug cost per case, the average examination and testing cost per case, the average medical service cost per case, and the average consumable cost per case; establishing an initial hypothesis test based on the average medical cost of each doctor; wherein, the initial hypothesis test is that there is no difference between the average medical cost of the doctor and the average standard cost; performing a difference test on the average total cost per case of each doctor and the total cost in the standard cost information, and when the difference between the two exceeds a preset threshold, determining that the initial hypothesis test is not valid and that the current doctor's resource consumption information for the disease category is abnormal; and generating a comparison result between the resource consumption information for the disease category and the standard cost information based on the average medical cost information of the doctor.

[0099] In one embodiment, the strategy generation module is used to generate a disease-specific cost optimization strategy for the next period based on the difference comparison results. Specifically, this includes: generating cost control results for each doctor based on the difference comparison results; sorting the cost control results; selecting a preset number of doctors with the highest rankings to generate a doctor cost control watchlist; generating cost control results for each abnormal disease based on the difference comparison results and the cost control results; sorting the cost control results; selecting a preset number of diseases with the highest rankings to generate a disease cost control watchlist; generating a phased cost optimization target based on a preset cost control time period and the cost control results; and generating a cost optimization strategy for the next period based on the doctor cost control watchlist, the disease cost control watchlist, and the phased cost optimization target. The cost optimization strategy includes real-time monitoring of the disease-specific resource consumption of each doctor in the doctor cost control watchlist in the next period based on the phased cost optimization target.

[0100] In one embodiment, the disease cost optimization device further includes: introducing an intelligent interaction step based on a large language model, which is used to receive natural language queries input by users, parse the elements such as disease, department, physician, time interval and cost indicators, automatically identify the query intent and call the corresponding data analysis or model calculation function, and generate a natural language response result containing a three-layer structure of "conclusion layer, evidence layer and suggestion layer", so as to realize intelligent interpretation and auxiliary decision-making for cost anomalies, cost control targets and trend changes.

[0101] In one embodiment, the large language model is linked with the disease standard cost model, cost control result database and early warning knowledge base through a retrieval enhancement generation mechanism to achieve the fusion reasoning of structured data and text knowledge; wherein, the conclusion layer is used to output cost determination and trend results, the evidence layer is used to display the calculation basis and data references, and the suggestion layer is used to output cost control optimization prompts.

[0102] In one embodiment, the large language model, through a retrieval enhancement generation mechanism, links with the disease-specific standard cost model, cost control result database, and early warning knowledge base to achieve the fusion generation of natural language query and structured data analysis results. It can automatically match relevant calculation results and generate interpretive output based on disease, physician, and time interval.

[0103] In one embodiment, the disease cost optimization device includes a task orchestration module, which is used to dynamically call between the anomaly detection, cost control calculation, trend analysis and report generation modules based on intent recognition results, and maintain the continuity and consistency of multi-turn question-and-answer semantics through a session-level context tracking mechanism.

[0104] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0105] This invention provides a disease-specific cost optimization device. It uses a pre-built standard cost model for diseases, trained on historical data, combined with current disease resource consumption information to calculate costs, resulting in standard cost information that aligns with actual conditions. This provides a scientific and accurate reference for subsequent cost management. A difference test is used to compare departmental disease resource consumption information with standard cost information, accurately identifying abnormal diseases where the difference exceeds a preset threshold. This allows for focusing on key issues and concentrating efforts on resolving cost anomalies. After identifying abnormal diseases, the device further obtains and compares the disease resource consumption information of the doctors in the first department under that abnormal disease, deepening the analysis to the doctor level and clarifying the specific source of the cost anomaly, making subsequent optimization strategies more targeted. Based on the difference comparison results, a disease-specific cost optimization strategy for the next period is generated. Cost management measures are adjusted promptly according to actual conditions to reasonably optimize disease costs, improve resource utilization efficiency, and enhance the level of medical cost control.

[0106] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make several improvements and substitutions without departing from the technical principles of the present invention, and these improvements and substitutions should also be considered within the scope of protection of the present invention.

Claims

1. A method for optimizing disease-specific costs, characterized in that, include: Obtain the current period's disease resource consumption information; wherein, the disease resource consumption information includes departmental disease resource consumption information and departmental physician disease resource consumption information; The current period's disease resource consumption information is input into a pre-constructed disease standard cost model for cost calculation to obtain the corresponding standard cost information; wherein, the disease standard cost model is trained based on a historical disease resource consumption information sample set; The departmental disease resource consumption information and the standard cost information are compared to perform a difference test. The disease to which the departmental disease resource consumption information belongs with a difference exceeding a preset threshold is determined as an abnormal disease. The first departmental doctor's disease resource consumption information under the abnormal disease is obtained, and a difference comparison result between the first departmental doctor's disease resource consumption information and the corresponding standard cost information is generated. Based on the comparison results, a disease cost optimization strategy for the next period will be generated.

2. The method for optimizing disease-specific costs as described in claim 1, characterized in that, The acquisition of information on the consumption of disease resources in the current period specifically includes: Obtain the inpatient medical record data for this period; Each inpatient's medical record data is categorized into its corresponding disease type, and medical resource consumption information is calculated at both the disease type and physician dimensions to obtain the department's disease type resource consumption information and the department's physician's disease type resource consumption information. The department's disease type resource consumption information represents the average medical cost per case within the same disease type, while the department's physician's disease type resource consumption information represents the average medical cost per physician within the same disease type.

3. The method for optimizing disease-specific costs as described in claim 1, characterized in that, The method for optimizing disease-specific costs further includes: constructing a sample set of historical disease-specific resource consumption information, specifically: Obtain the inpatient medical record data from the previous period; wherein, the inpatient medical record data includes basic patient information, patient disease characteristics, medical resource characteristics, and patient cost data; The inpatient medical record data is preprocessed; wherein, the preprocessing includes removing cases with missing or abnormal data and removing cases with abnormal patient costs; Each inpatient's medical record data is classified into its corresponding disease category, and the standard deviation of medical expenses for each case under each disease category is calculated one by one; wherein, the standard deviation of medical expenses is based on the difference between the medical expenses of each case and the mean medical expenses of each case under the corresponding disease category. Inpatient medical records with medical cost standard deviations exceeding a preset standard deviation threshold are removed to obtain a historical disease resource consumption information sample set for each disease; wherein, the number of inpatient medical records for each disease in the historical disease resource consumption information sample set exceeds a preset threshold.

4. The method for optimizing disease-specific costs as described in claim 1, characterized in that, The standard cost model for the disease is trained based on a historical sample set of disease resource consumption information. The standard cost model for the disease is trained in the following way: Feature data and tag data are established for the data content of each historical disease resource consumption information sample set; The number of neurons, d, is set based on the amount of feature data, and the number of hidden layers is set to 2. Construct the network topology and set the standardized feature vectors as X∈R d (X=[x1,x2,…,x) d ] T ). Forward propagation is used and the output of each layer is set, specifically: Output of the first hidden layer: H1 = σ(W1X + b1) (W1 ∈ R) n1×d Let b1 ∈ R be the weight matrix. n1 (for bias vectors) Output of the second hidden layer: H2 = σ(W2H1 + b2) (W2 ∈ R) n2×n1 Let b2 be the weight matrix. Rn2 (for bias vectors) Output layer predictions: =W3H2+b3(W3∈ R1×n2 Let b3 be the weight matrix and b3 ∈ R be the bias vector. (Forecast costs); The historical disease resource consumption information sample set is divided into a training set and a validation set according to a preset ratio. The network topology is iteratively trained based on the training set, and the parameters of the network topology are updated until the optimal loss function is obtained. The loss function is obtained based on the difference between the predicted and actual values ​​of the labeled data, and its expression is: ； Where N is the number of samples, Yi is the actual cost of the i-th sample, and λ is the regularization coefficient; The average performance of the network topology is verified several times based on the verification set, and its coefficient of determination is calculated; wherein the expression for the coefficient of determination is: ； in As the coefficient of determination, This represents the average of the actual costs. When the determination coefficient exceeds a preset threshold, the training of the standard cost model for the current disease is considered complete.

5. The method for optimizing disease-specific costs as described in claim 4, characterized in that, The specific steps for establishing feature data and label data for the data content of each historical disease resource consumption information sample set are as follows: The characteristic data includes patient gender, age, medical insurance type, primary diagnosis, admission route, discharge method, treatment method, primary surgery, surgical level, number of operations, whether admitted to the intensive care unit, and number of resuscitation attempts; The label data includes total cost, drug cost, examination and testing cost, medical service cost, and consumable cost.

6. The method for optimizing disease-specific costs as described in claim 5, characterized in that, The step of obtaining the resource consumption information of the first department doctor under the abnormal disease category and generating a comparison result of the difference between the resource consumption information of the first department doctor and the corresponding standard cost information specifically includes: Obtain the disease resource consumption information of the first department doctor under the abnormal disease; wherein, the disease resource consumption information of the first department doctor is the average medical expenses of different doctors in the department under the abnormal disease, and the average medical expenses include the average total cost per case, the average drug cost per case, the average examination and testing cost per case, the average medical service cost per case, and the average consumable cost per case; An initial hypothesis test is established based on the average medical cost for each doctor; wherein the initial hypothesis test is that the average medical cost of doctors is no different from the average standard cost. The average total cost per case for each doctor is compared with the total cost in the standard cost information. When the difference between the two exceeds a preset threshold, it is determined that the initial hypothesis test is invalid and there is an anomaly in the current doctor's disease resource consumption information. The comparison results between the disease-specific resource consumption information and the standard cost information are generated based on the average medical expenses of doctors.

7. The method for optimizing disease-specific costs as described in claim 6, characterized in that, The step of generating the next phase's disease cost optimization strategy based on the difference comparison results specifically includes: Based on the comparison results, cost control results are generated for each doctor. The cost control results are sorted, and a preset number of doctors with the highest rankings are selected to generate a doctor cost control focus list. Based on the difference comparison results, cost control results for each abnormal disease are generated and the cost control results are generated. The cost control results are sorted, and a preset number of diseases with the highest ranking are selected to generate a cost control focus list for each disease. A phased cost optimization target is generated based on the preset cost control time period and the cost control results; The cost optimization strategy for the next period is generated based on the physician cost control focus list, the disease cost control focus list, and the phased cost optimization target; wherein, the cost optimization strategy includes real-time monitoring of the disease resource consumption of each physician in the physician cost control focus list in the next period based on the phased cost optimization target.

8. A disease-specific cost optimization device, characterized in that, include: The module includes an information acquisition module, a cost calculation module, a difference comparison module, and a strategy generation module. The information acquisition module is used to acquire the current period's disease resource consumption information; wherein, the disease resource consumption information includes departmental disease resource consumption information and departmental doctor's disease resource consumption information; The cost calculation module is used to input the current period's disease resource consumption information into a pre-built disease standard cost model for cost calculation, and obtain the corresponding standard cost information; wherein, the disease standard cost model is trained based on a historical disease resource consumption information sample set; The difference comparison module is used to perform difference detection between the departmental disease resource consumption information and the standard cost information, determine the disease to which the departmental disease resource consumption information with a difference exceeding a preset threshold belongs as an abnormal disease, obtain the disease resource consumption information of the first department doctor under the abnormal disease, and generate a difference comparison result between the disease resource consumption information of the first department doctor and the corresponding standard cost information. The strategy generation module is used to generate a disease cost optimization strategy for the next period based on the difference comparison results.

9. A terminal device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the disease cost optimization method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform the disease cost optimization method as described in any one of claims 1 to 7.

Images