Clinical pathway cost real-time monitoring method and system based on DIP payment standard

Abstract

The present application relates to the technical field of medical informatization and clinical pathway management, and discloses a clinical pathway cost real-time monitoring method and system based on DIP payment standards, which comprises the following steps: obtaining DIP grouping and payment standard information; collecting diagnosis and treatment cost details in real time and accumulating and calculating the current total cost; dynamically calculating the remaining payable amount; when the optimization starting threshold in the multi-level early warning threshold is triggered, extracting historical cost and efficacy data from a cost-path-efficacy ternary association database; under the premise of meeting the cost standard constraint, performing multi-objective optimization solution with the goal of minimizing efficacy loss, generating a medical order adjustment recommendation list and pushing it to the doctor workstation. Through the above technical solution, dynamic balance of the dual goals of cost control and efficacy can be achieved during diagnosis and treatment.

Description

Technical Field

[0001] This invention relates to the field of medical informatics and clinical pathway management technology, and in particular to a method and system for real-time monitoring of clinical pathway costs based on the DIP payment standard. Background Technology

[0002] Disease-based value (DIP) payment is a core system in my country's medical insurance payment reform. It achieves refined allocation of medical insurance funds by grouping discharged cases and assigning corresponding values. Clinical pathways, as a standardized treatment management tool, are widely used to unify treatment behaviors and control medical costs. In recent years, the industry has begun to explore combining DIP payment standards with clinical pathway management. This involves comparing the cost of a pathway template with the DIP benchmark cost during the design phase, selecting the optimal template that meets the cost target, and statistically analyzing indicators such as pathway completion rate and average cost per case after implementation. For example, Chinese patent CN115346647A discloses an intelligent DIP clinical pathway planning and management information method and system, which selects a pathway template by comparing the cost of a clinical pathway template with the DIP medical insurance benchmark cost and prompts for exit processing when the implemented path does not match the template.

[0003] Existing technical solutions suffer from fundamental technical flaws: their cost control models are essentially based on "pre-treatment comparison during the pathway design phase" and "post-treatment analysis after pathway execution," failing to intervene during the real-time execution phase of the treatment process. When actual costs deviate from expectations due to factors such as adjustments to medical orders or changes in the patient's condition during pathway execution, existing technologies can only perform statistical analysis after the patient leaves the pathway or is discharged. By this time, the costs have already been incurred and cannot be reversed. More importantly, existing technologies focus only on the single dimension of cost, neglecting the varying impacts of different treatment plans on patient efficacy. When costs face the risk of exceeding the limit, there is a lack of technical means to simultaneously balance cost control objectives and efficacy assurance. This "reactive post-treatment" model leads to delayed cost control effects and easily results in the risk of sacrificing efficacy for cost control, failing to meet the requirements of precise real-time management of clinical pathways under the DIP payment reform.

[0004] Therefore, this invention proposes a method and system for real-time monitoring of clinical pathway costs based on the DIP payment standard. Summary of the Invention

[0005] This invention provides a method and system for real-time monitoring of clinical pathway costs based on the DIP payment standard. By monitoring costs in real time and automatically recommending the medical order adjustment plan with the least loss of efficacy based on cost-efficacy multi-objective optimization before the cost exceeds the limit, it solves the technical defects of existing technologies that cannot achieve a dynamic balance between cost control and efficacy during the diagnosis and treatment process, and achieves the beneficial effect of avoiding cost exceeding the limit while maximizing the maintenance of the patient's treatment effect.

[0006] This invention provides a method for real-time monitoring of clinical pathway costs based on the DIP payment standard, comprising: Obtain the DIP grouping information and corresponding DIP payment standard information for the target patient's current admission; The system collects detailed data on the medical expenses already incurred by the target patient in real time from the hospital information system and calculates the current total cost. Based on the standard payment limit in the DIP payment standard information and the current total cost, the remaining payable amount is dynamically calculated; When the remaining payable amount triggers the optimized activation threshold among the preset multi-level warning thresholds, obtain the clinical pathway node information of the target patient; From the cost-pathway-efficacy ternary correlation database, extract historical cost data and historical efficacy index data corresponding to different combinations of medical orders for historical patients at the current clinical pathway node; Under the premise of meeting the cost target constraint, with the goal of minimizing the loss of efficacy, the optional medical order adjustment plan at the current clinical pathway node is optimized by multi-objective solution to generate a recommended list of medical order adjustments. The cost target constraint is that the adjusted expected total cost is less than or equal to the standard payment amount in the DIP payment standard information. The list of recommended adjustments to medical orders will be pushed to the doctor's workstation.

[0007] Furthermore, the multi-level early warning thresholds include an alert threshold, a warning threshold, and an optimization activation threshold. The alert threshold is higher than the warning threshold, and the warning threshold is higher than the optimization activation threshold. When the remaining payable amount falls below the alert threshold, a fee reminder notification will be issued; A fee warning notification is issued when the remaining payable amount falls below the warning threshold.

[0008] Furthermore, the historical efficacy indicators consist of one or more of the following: length of hospital stay, complication rate, and readmission rate; the cost target constraint is that the adjusted estimated total cost is less than or equal to the product of the standard payment amount in the DIP payment standard information and the preset target coefficient.

[0009] Furthermore, the Pareto front algorithm is used to solve the multi-objective optimization problem, specifically as follows: Map each historical medical order combination at the current clinical pathway node to a point in a two-dimensional space, with the horizontal axis representing the cost value and the vertical axis representing the loss of efficacy. Select the Pareto front point set from the mapping points that is not jointly dominated by other points in both the x-axis and y-axis; From the Pareto frontier cluster, further screening was conducted to identify the corresponding medical order adjustment plans for points that met the cost target constraints, and these were then sorted in ascending order of efficacy loss.

[0010] Furthermore, the construction process of the cost-pathway-efficacy ternary association database includes: Extract clinical pathway execution records of discharged cases under the DIP group from the hospital's historical medical record system. Each record contains details of the medical orders actually executed by the patient at the clinical pathway node, corresponding cost details, length of hospital stay, complication occurrence markers, and readmission markers. Each record's medical order items are grouped and aggregated according to three dimensions: DIP grouping, clinical pathway node, and medical order combination. The medical order combination is formed by the aggregation of medical order items, forming a mapping relationship of DIP grouping - pathway node - medical order combination - cost. The length of hospital stay, the occurrence of complications, and the readmission criteria are normalized and weighted to generate a comprehensive efficacy score. The comprehensive efficacy score is linked to the cost value of the corresponding DIP group-path node-medical order combination to form a cost-path-efficacy ternary association database.

[0011] Furthermore, it also includes a hybrid architecture that combines offline training with online inference: In the offline phase, historical data from the cost-path-efficacy ternary association database is used, with medical order combination features and individual feature data as inputs and comprehensive efficacy scores as outputs, to train an efficacy prediction model. Individual feature data includes age and comorbidity codes. During the online phase, the system acquires the target patient's current medical order combination, candidate medical order adjustment plans, and individualized characteristic data of the target patient. It then calls the trained efficacy prediction model and outputs the predicted efficacy score of each candidate plan for the current target patient. The predicted efficacy score for the current target patient is substituted into the multi-objective optimization solution process to generate a list of recommended adjustments to medical orders.

[0012] Furthermore, it also includes: When a doctor selects a recommended option from the list of recommended adjustments to medical orders, the doctor records a comparison between the adopted option and the non-adopted option. When a doctor submits a custom prescription combination instead of the recommended plan, record the custom prescription combination information and the reason for not adopting it; The adoption and non-adoption records will be written back to the cost-path-efficacy ternary association database.

[0013] Furthermore, it also includes: At key diagnosis and treatment nodes in the clinical pathway, the core medical system compliance verification is automatically executed. The verification rules include verification of transfer time limits, verification of blood transfusion effect evaluation writing, and verification of discharge criteria for Level 1 nursing care. If the verification fails, a task to be done is generated and pushed to the doctor's workstation, and the system compliance status is marked in the path execution record; The hospital information system collects the actual number of days of hospitalization for target patients in real time. When the remaining payable amount has not triggered the optimization start threshold, but the actual number of days of hospitalization exceeds the standard number of days of hospitalization set by the clinical pathway and the average daily cost is higher than the preset average daily cost warning line, an overdue hospitalization cost warning is triggered.

[0014] Furthermore, it also includes: When a target patient triggers a branch switch within the clinical pathway or a cross-pathway conversion during the execution of the clinical pathway, the path identifier before conversion, the path identifier after conversion, the reason for conversion, and the conversion timestamp are recorded. The new DIP grouping and new DIP payment standard information corresponding to the converted path are obtained, and the calculation basis of the remaining payable amount is reset based on the new DIP payment standard information. When the DIP payment standard information changes due to policy adjustments, update the DIP payment standard information, re-execute the multi-objective optimization solution, and mark the schemes in the medical order adjustment recommendation list that no longer meet the cost target constraint due to the change in payment standard as invalid.

[0015] This invention provides a real-time monitoring system for clinical pathway costs based on the DIP payment standard, comprising: The disease-payment standard matching module is used to store the DIP grouping codes and DIP payment standard information associated with the clinical pathway forms; A cost-pathway-efficacy ternary association database is used to store historical patient medical order combinations, cost data, and efficacy indicator data by DIP grouping and clinical pathway node aggregation; The offline training module is used to train the efficacy prediction model using historical data from the cost-path-efficacy ternary association database, with medical order combination features and individual feature data as input and comprehensive efficacy score as output. The real-time cost collection module is used to establish a data interface with the hospital information system, and to obtain detailed data on the treatment costs of the target patients in real time after the clinical pathway is started, and to accumulate and generate the current total cost. The dynamic settlement and multi-level early warning module is used to dynamically calculate the remaining payable amount based on the standard payment amount in the DIP payment standard information and the current total cost generated by the real-time cost collection module. It compares the remaining payable amount with the preset multi-level early warning threshold in real time and issues corresponding early warning notifications. When the remaining payable amount triggers the optimization start threshold in the multi-level early warning threshold, it obtains the clinical pathway node information of the target patient and outputs the optimization signal. The multi-objective optimization solution module is used to receive optimization signals, extract historical cost data and historical efficacy index data corresponding to the current clinical path node from the cost-path-efficacy ternary association database, call the efficacy prediction model generated by the offline training module, and generate a list of recommended medical order adjustments with the Pareto front algorithm under the premise of meeting the cost target constraint and minimizing efficacy loss. The recommendation push and adoption module is used to push the list of recommended medical order adjustments to the doctor's workstation, receive the doctor's adoption selection or custom medical order combination, and write the adoption record back to the cost-path-efficacy three-element association database; The path conversion and policy adaptation module is used to detect branch switching or cross-path conversion events during the execution of clinical pathways, record the path identifiers and reasons for conversion before and after conversion, obtain the new DIP grouping and new DIP payment standard information corresponding to the converted path, and notify the dynamic settlement and multi-level early warning module to reset the calculation benchmark. When the DIP payment standard information changes due to policy adjustments, the DIP payment standard information is updated, triggering the multi-objective optimization solution module to resolve and mark the failed solution. The core system compliance and inpatient monitoring module is used to perform core medical system compliance verification at key diagnosis and treatment nodes of the clinical pathway. The verification rules include verification of transfer time limits, verification of blood transfusion effect evaluation writing, and verification of discharge criteria for Level 1 nursing care. When the verification fails, a task is generated and pushed to the doctor's workstation. The module collects the actual number of days of hospitalization of the target patient in real time from the hospital information system. When the remaining payable amount has not triggered the optimization start threshold, but the actual number of days of hospitalization exceeds the standard number of days of hospitalization set by the clinical pathway and the average daily cost is higher than the preset average daily cost warning line, an overdue hospitalization cost warning is triggered.

[0016] The beneficial effects of this invention compared to existing technologies are as follows: It overcomes the fundamental technical deficiency of existing DIP clinical pathway management systems, which can only perform pre-treatment cost comparisons during the pathway design phase or post-treatment statistical analysis after pathway execution, and cannot intervene in cost deviations in real time during the treatment process. It also solves the problem that existing technologies only focus on the single dimension of cost while ignoring the differentiated impact of different treatment plans on patient efficacy. By constructing a complete technical closed loop encompassing real-time cost collection, dynamic remaining budget calculation, multi-level early warning triggering, extraction of cost-pathway-efficacy three-dimensional data, multi-objective optimization solution, and prescription adjustment recommendation, a dynamic balance between cost control and efficacy assurance is achieved in the field of DIP clinical pathway management. When costs approach the payment standard, the system does not simply intercept or alert for exceeding the limit, but automatically calculates and recommends prescription adjustment plans that minimize efficacy loss while meeting cost constraints based on historical real-case cost and efficacy data. This ensures the rational use of medical insurance funds while maximizing patient treatment outcomes, promoting the upgrade of clinical pathway management from passive outflow processing to proactive intelligent optimization.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a technical flowchart of the real-time monitoring method for clinical pathway costs based on the DIP payment standard in an embodiment of the present invention; Figure 2 This is a system architecture diagram of a real-time monitoring system for clinical pathway costs based on the DIP payment standard, as described in this embodiment of the invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] refer to Figure 1 and Figure 2 This invention provides an embodiment of a method for real-time monitoring of clinical pathway costs based on the DIP payment standard, comprising: Obtain the DIP grouping information and corresponding DIP payment standard information for the target patient's current admission; The system collects detailed data on the medical expenses already incurred by the target patient in real time from the hospital information system and calculates the current total cost. Based on the standard payment limit in the DIP payment standard information and the current total cost, the remaining payable amount is dynamically calculated; When the remaining payable amount triggers the optimized activation threshold among the preset multi-level warning thresholds, obtain the clinical pathway node information of the target patient; From the cost-pathway-efficacy ternary correlation database, extract historical cost data and historical efficacy index data corresponding to different combinations of medical orders for historical patients at the current clinical pathway node; Under the premise of meeting the cost target constraint, with the goal of minimizing the loss of efficacy, the optional medical order adjustment plan at the current clinical pathway node is optimized by multi-objective solution to generate a recommended list of medical order adjustments. The cost target constraint is that the adjusted expected total cost is less than or equal to the standard payment amount in the DIP payment standard information. The list of recommended adjustments to medical orders will be pushed to the doctor's workstation.

[0022] In this embodiment, the DIP grouping information and corresponding DIP payment standard information for the target patient's current admission are obtained in the following way: The system sends the target patient's discharge diagnosis code and treatment method code to the DIP grouper interface deployed in the hospital, receives the DIP grouping code and group name returned by the DIP grouper, and queries the preset score-based unit price mapping table according to the DIP grouping code to obtain the corresponding DIP payment standard information. The DIP payment standard information includes the standard payment amount and payment multiplier range. Taking a hospitalized patient diagnosed with acute appendicitis and undergoing laparoscopic appendectomy as an example, after the patient completes the diagnosis code entry upon admission, the system automatically calls the DIP grouper interface to obtain the patient's corresponding DIP grouping code and queries the mapping table to obtain the payment standard amount for the group, providing a cost benchmark for subsequent real-time cost monitoring.

[0023] In this embodiment, when collecting detailed medical expense data of the target patient from the hospital information system in real time, the system establishes a data connection with the hospital information system through the HL7 standard interface or WebService interface. The hospital information system can be a system provided by mainstream vendors such as Neusoft, Winning, and Chuangye Huikang. The collected detailed medical expense data includes itemized data such as drug costs, examination costs, consumable costs, treatment costs, and nursing costs. The system adds up the above detailed data item by item to calculate the current total cost. Whenever a new cost record is generated in the hospital information system, the system obtains the record in real time through the interface and updates the current total cost.

[0024] In this embodiment, the remaining payable amount is dynamically calculated based on the standard payment amount in the DIP payment standard information and the current total cost. Specifically, the remaining payable amount equals the standard payment amount minus the current total cost. Simultaneously, the system also calculates the cost consumption percentage, which is equal to the current total cost divided by the standard payment amount and then multiplied by 100%. Each time the system receives new detailed medical expense data, it automatically triggers a recalculation of the remaining payable amount and the cost consumption percentage to ensure that the cost monitoring data remains synchronized with the patient's actual cost status.

[0025] In this embodiment, the multi-level early warning thresholds include an alert threshold, a warning threshold, and an optimization activation threshold, all three thresholds using the remaining payable amount as a comparison benchmark. The alert threshold is higher than the warning threshold, and the warning threshold is higher than the optimization activation threshold. An optional threshold configuration is as follows: the alert threshold is set to 30% of the standard payment amount, the warning threshold is set to 10% of the standard payment amount, and the optimization activation threshold is set to 5% of the standard payment amount. When the remaining payable amount falls below the alert threshold, the system issues a cost alert notification on the doctor's workstation interface, including the current amount of cost consumed, the remaining payable amount, and the percentage of cost consumed. When the remaining payable amount falls below the warning threshold, the system issues a cost warning notification. The visual intensity of the warning notification is higher than that of the alert notification, for example, by using a different color or by displaying a warning window requiring the doctor's confirmation to close it. When the remaining payable amount falls below the optimization activation threshold, the system automatically triggers the subsequent multi-objective optimization solution process. The threshold values ​​and hierarchical relationships corresponding to the above three levels of early warning enable the system to take intervention measures of different intensities at different stages of cost risk, achieving progressive control from light to heavy and from reminders to proactive optimization.

[0026] In this embodiment, the construction process of the cost-path-efficacy ternary association database is as follows: Step 1: Extract clinical pathway execution records of discharged cases under each DIP group from the hospital's historical medical record system. The extracted data covers discharged cases within the past 3 years. Each record contains details of the medical orders actually executed by the patient at the clinical pathway node, corresponding cost details, length of hospital stay, complication occurrence markers, and readmission status markers. Step 2: Group and aggregate the medical orders of each record according to three dimensions: DIP group, clinical pathway node, and medical order combination. Medical order combinations are formed by aggregating medical order items. The same medical order combination executed by different patients at the same clinical pathway node is merged into one medical order combination record. After aggregation, a mapping relationship of DIP group-path node-medical order combination-cost is formed. Step 3: Normalize and weight the length of hospital stay, complication occurrence markers, and readmission status markers to generate a comprehensive efficacy score. The normalization adopts the Min-Max normalization method, mapping each indicator to the range of 0 to 1, where 1 represents the best efficacy and 0 represents the worst efficacy. The normalized calculation method for length of stay is as follows: Take the maximum and minimum length of stay for all historical cases under the specified DIP group in the dataset. The normalized length of stay is equal to the maximum value minus the actual length of stay, divided by the difference between the maximum and minimum values. The complication occurrence indicator is set as follows: 0 for complications, 1 for no complications. The readmission indicator is set as follows: 0 for readmission within 30 days, 1 for no readmission. The weighting coefficients satisfy the condition that the sum of the three weights equals 1. One optional weighting configuration is: length of stay weight 0.4, complication occurrence indicator weight 0.3, and readmission indicator weight 0.3. The comprehensive efficacy score is equal to the normalized length of stay multiplied by its weight, plus the normalized complication occurrence indicator multiplied by its weight, plus the normalized readmission indicator multiplied by its weight. Step 4: Establish a correlation between the comprehensive efficacy score and the cost value of the corresponding DIP group-path node-medical order combination, completing the construction of the cost-path-efficacy ternary association database.

[0027] In this embodiment, the Pareto front algorithm is used for multi-objective optimization, and the specific execution process is as follows: Step 1: Map each historical medical order combination on the current clinical pathway node to a point in a two-dimensional space, with the horizontal axis representing the cost value and the vertical axis representing the efficacy loss value. The efficacy loss value is equal to 1 minus the comprehensive efficacy score. Step 2: Select the Pareto front point set from all mapped points. The Pareto front point set consists of all points that meet the following conditions: no other point exists in the set, its horizontal axis cost value is smaller, and its vertical axis efficacy loss value is also smaller, that is, the point is not jointly dominated by any other point in both cost and efficacy dimensions. Step 3: From the Pareto front point set, further select the medical order adjustment schemes corresponding to points that meet the cost target constraint. The cost target constraint is that the adjusted expected total cost is less than or equal to the product of the standard payment amount in the DIP payment standard information and the preset target coefficient, which can be a value between 0.95 and 1.00. Step 4: Sort the selected medical order adjustment plans in ascending order of efficacy loss value, generating a recommended medical order adjustment list. After the recommended list is pushed to the doctor's workstation, it displays the specific medical order content, estimated cost, estimated comprehensive efficacy score, and estimated efficacy loss value of each recommended plan in list format. Doctors can select the appropriate plan from the list based on the patient's specific condition and clinical experience.

[0028] In this invention, efficacy loss is used to quantify the difference between a given prescription adjustment plan and the optimal efficacy plan. In a direct calculation method based on historical data, efficacy loss equals 1 minus the overall efficacy score. In an online inference method using an efficacy prediction model, efficacy loss equals 1 minus the predicted efficacy score. The smaller the efficacy loss value, the closer the expected efficacy of the plan is to the optimal level.

[0029] In this embodiment, after the recommended list of medical order adjustments is generated, the system pushes the list to the doctor's workstation terminal via the hospital's internal network. The doctor's workstation interface displays the recommended list in the form of a pop-up window or sidebar. Each recommended option in the list includes detailed information on the specific medical order items, the corresponding estimated total cost, the estimated comprehensive efficacy score, and the efficacy loss value, facilitating doctors to compare and select among multiple recommended options. Doctors can choose one recommended option from the list to adopt directly, make minor adjustments to the recommended option before submitting, or submit a custom combination of medical orders without adopting any recommended options.

[0030] Furthermore, the multi-level early warning thresholds include an alert threshold, a warning threshold, and an optimization activation threshold. The alert threshold is higher than the warning threshold, and the warning threshold is higher than the optimization activation threshold. When the remaining payable amount falls below the alert threshold, a fee reminder notification will be issued; A fee warning notification is issued when the remaining payable amount falls below the warning threshold.

[0031] In this embodiment, the multi-level early warning thresholds include an alert threshold, a warning threshold, and an optimization activation threshold, all three thresholds using the remaining payable amount as a comparison benchmark. The alert threshold is higher than the warning threshold, and the warning threshold is higher than the optimization activation threshold. Specifically, a larger remaining payable amount indicates greater safety in terms of expenses; therefore, the thresholds, from high to low, correspond to a progressive relationship of low to high risk. One optional threshold configuration scheme is as follows: the alert threshold is set to 30% of the standard payment amount, the warning threshold is set to 10% of the standard payment amount, and the optimization activation threshold is set to 5% of the standard payment amount. Taking a patient with a standard payment amount of 15,000 yuan as an example, when the remaining payable amount drops below 4,500 yuan, an expense alert notification is triggered; when the remaining payable amount drops below 1,500 yuan, an expense warning notification is triggered; and when the remaining payable amount drops below 750 yuan, optimization activation is triggered.

[0032] When the remaining payable amount falls below the alert threshold, the system sends a payment reminder notification on the doctor's workstation interface. The notification appears as a prominent payment indicator bar at the top or side of the medical order entry interface, displaying the current amount spent, the remaining payable amount, and the percentage of payment consumed. The notification does not interfere with the doctor's normal medical order processing; it serves only to inform and remind the doctor.

[0033] When the remaining payable amount further decreases below the warning threshold, the system issues a cost warning notification. The visual cues for warning notifications are stronger than those for alert notifications; for example, they may use a different red or orange label, or be presented as a pop-up window requiring the doctor's confirmation to close. In addition to displaying the same cost information as the alert notification, the warning notification also alerts the doctor that the current cost is approaching the standard payment limit, and that the necessity of subsequent prescriptions should be carefully assessed.

[0034] The aforementioned reminder and warning thresholds enable the system to draw doctors' attention with mild reminders when the cost risk is low, to prompt doctors to make prudent decisions with stronger warnings when the cost risk increases, and to automatically initiate a multi-objective optimization process when the cost risk reaches a high level, mining the optimal medical order adjustment plan that balances cost control and efficacy assurance from historical data, thus forming a complete early warning system from passive prompts to proactive optimization.

[0035] Furthermore, the historical efficacy indicators consist of one or more of the following: length of hospital stay, complication rate, and readmission rate; the cost target constraint is that the adjusted estimated total cost is less than or equal to the product of the standard payment amount in the DIP payment standard information and the preset target coefficient.

[0036] In this embodiment, historical efficacy index data is used to quantitatively evaluate the actual treatment effects of different medication order combinations in historical cases. Historical efficacy index data consists of one or more of the following: length of hospital stay, complication rate, and readmission rate. When length of hospital stay is selected as the efficacy index, a shorter hospital stay indicates better efficacy. When complication rate is selected as the efficacy index, a lower complication rate indicates better efficacy. When readmission rate is selected as the efficacy index, a lower readmission rate indicates better efficacy. When multiple indicators are selected simultaneously, each indicator needs to be normalized and weighted to generate a comprehensive efficacy score to comprehensively reflect the overall efficacy level corresponding to the medication order combination. Taking a hospitalized patient diagnosed with acute exacerbation of chronic obstructive pulmonary disease as an example, one of the patient's clinical pathway nodes is the anti-infective treatment stage, at which three commonly used antibiotic medication order combinations have historically existed. The historical average length of hospital stay for cases corresponding to the first combination of medical orders was 8 days, with a complication rate of 5% and a readmission rate of 3% within 30 days. The historical average length of hospital stay for cases corresponding to the second combination of medical orders was 10 days, with a complication rate of 8% and a readmission rate of 5% within 30 days. The historical average length of hospital stay for cases corresponding to the third combination of medical orders was 7 days, with a complication rate of 4% and a readmission rate of 2% within 30 days. These three indicators were normalized and weighted to generate a comprehensive efficacy score for each combination of medical orders. A higher score indicates a better overall efficacy of the treatment plan.

[0037] The cost compliance constraint is used to limit the feasible domain of prescription adjustment schemes during multi-objective optimization. Specifically, the adjusted total estimated cost must be less than or equal to the product of the standard payment amount in the DIP payment standard information and the preset compliance coefficient. The preset compliance coefficient ranges from 0 to 1, with an optional value of 0.95. For example, with a standard payment amount of 15,000 yuan in the DIP payment standard information and a preset compliance coefficient of 0.95, the cost compliance constraint is that the adjusted total estimated cost must be less than or equal to 14,250 yuan. After solving the Pareto front, the system filters prescription adjustment schemes that meet this cost constraint from the Pareto front set, including only schemes with costs not exceeding 14,250 yuan in the recommendation list. When the preset compliance coefficient is 1, the cost compliance constraint is that the adjusted total estimated cost must be less than or equal to the standard payment amount. By adjusting the preset target achievement coefficient, hospital administrators can flexibly control the strictness of cost constraints and achieve a balance between cost control rigidity and clinical flexibility.

[0038] Furthermore, the Pareto front algorithm is used to solve the multi-objective optimization problem, specifically as follows: Map each historical medical order combination at the current clinical pathway node to a point in a two-dimensional space, with the horizontal axis representing the cost value and the vertical axis representing the loss of efficacy. Select the Pareto front point set from the mapping points that is not jointly dominated by other points in both the x-axis and y-axis; From the Pareto frontier cluster, further screening was conducted to identify the corresponding medical order adjustment plans for points that met the cost target constraints, and these were then sorted in ascending order of efficacy loss.

[0039] In this embodiment, the Pareto front algorithm is used for multi-objective optimization, which can find the optimal set of balanced solutions between the two mutually constraining objectives of cost and efficacy. The specific execution process consists of four steps.

[0040] Step 1: Map each historical prescription combination at the current clinical pathway node to a point in a two-dimensional space. The horizontal axis represents the cost value, indicating the historical average or estimated cost of that prescription combination; the vertical axis represents the efficacy loss value, which equals 1 minus the overall efficacy score corresponding to that prescription combination. The overall efficacy score is a value between 0 and 1, with a higher score indicating better efficacy. Therefore, a lower efficacy loss value indicates that the efficacy is closer to the optimal level. Taking the anti-infective treatment phase of a patient with acute exacerbation of chronic obstructive pulmonary disease as an example, there are 3 historical prescription combinations at this clinical pathway node. The historical average cost of prescription combination A is 3200 yuan, and the comprehensive efficacy score is 0.85. It is mapped to point A, with an x-axis of 3200 and a y-axis of 0.15. The historical average cost of prescription combination B is 2800 yuan, and the comprehensive efficacy score is 0.72. It is mapped to point B, with an x-axis of 2800 and a y-axis of 0.28. The historical average cost of prescription combination C is 3500 yuan, and the comprehensive efficacy score is 0.90. It is mapped to point C, with an x-axis of 3500 and a y-axis of 0.10.

[0041] Step 2: Select the Pareto front set from all mapped points. The Pareto front set consists of points that meet the following conditions: no other point in the set has a smaller cost value on the x-axis and a smaller efficacy loss value on the y-axis; that is, the point is not jointly dominated by any other point in both cost and efficacy dimensions. Continuing the example above, compare points A, B, and C. Point A's cost of 3200 yuan is lower than point C's cost of 3500 yuan, but its efficacy loss of 0.15 is higher than point C's 0.10. Therefore, A is not dominated by C, and C is not dominated by A. Point B's cost of 2800 yuan is lower than point A's cost of 3200 yuan, but its efficacy loss of 0.28 is higher than point A's 0.15. Therefore, B is not dominated by A, and A is not dominated by B. Point B's cost of 2800 yuan is lower than point C's cost of 3500 yuan, but its efficacy loss of 0.28 is higher than point C's 0.10. Therefore, B is not dominated by C, and C is not dominated by B. All three points are not surpassed by any other point in both dimensions and belong to the Pareto front set. If there exists a point with higher cost and greater loss of efficacy, it is jointly dominated by some points in the Pareto front and will be excluded.

[0042] Step 3 involves further filtering the Pareto front set to identify adjustment plans that meet the cost compliance constraint. The cost compliance constraint is that the adjusted total estimated cost is less than or equal to the product of the standard payment amount in the DIP payment standard information and the preset compliance coefficient. Assuming the standard payment amount is 15,000 yuan and the preset compliance coefficient is 0.95, the cost limit is 14,250 yuan. In the example above, the estimated costs of all three plans are significantly lower than 14,250 yuan, thus all meeting the cost compliance constraint. If the cost value of a Pareto front exceeds 14,250 yuan, the corresponding adjustment plan for that point will be excluded from the recommendation list.

[0043] Step 4 involves sorting the selected prescription adjustment plans in ascending order of efficacy loss value to generate a recommended prescription adjustment list. Ascending order means that the plan with the smallest efficacy loss, i.e., the highest overall efficacy score, is placed at the top of the list. In the example above, point C, with an efficacy loss of 0.10, is ranked first, point A, with an efficacy loss of 0.15, is ranked second, and point B, with an efficacy loss of 0.28, is ranked third. After the recommended list is pushed to the doctor's workstation, it is displayed in list format. Each item shows the details of the specific prescription items included in the plan, the estimated cost, the overall efficacy score, and the efficacy loss value. Doctors can choose the most suitable plan from the list based on the patient's specific condition and clinical experience.

[0044] Furthermore, the construction process of the cost-pathway-efficacy ternary association database includes: Extract clinical pathway execution records of discharged cases under the DIP group from the hospital's historical medical record system. Each record contains details of the medical orders actually executed by the patient at the clinical pathway node, corresponding cost details, length of hospital stay, complication occurrence markers, and readmission markers. Each record's medical order items are grouped and aggregated according to three dimensions: DIP grouping, clinical pathway node, and medical order combination. The medical order combination is formed by the aggregation of medical order items, forming a mapping relationship of DIP grouping - pathway node - medical order combination - cost. The length of hospital stay, the occurrence of complications, and the readmission criteria are normalized and weighted to generate a comprehensive efficacy score. The comprehensive efficacy score is linked to the cost value of the corresponding DIP group-path node-medical order combination to form a cost-path-efficacy ternary association database.

[0045] In this embodiment, the construction of the cost-path-efficacy ternary association database is completed in four steps.

[0046] Step 1 involves extracting clinical pathway execution records for discharged patients under the DIP group from the hospital's historical medical record system. The extracted data covers discharged cases within the past three years to ensure a sufficient sample size for statistical analysis. Each record includes details of the medical orders actually executed by the patient at a specific clinical pathway node, corresponding cost details, length of hospital stay, complication occurrence markers, and readmission status. The detailed medical order list specifies all medical orders actually issued to the patient at a particular clinical pathway node, such as the specific drug name and dosage in medication orders, the specific examination items and frequency in examination orders, and the specific treatment procedure names in treatment orders. The corresponding cost details record the actual cost incurred for each medical order item. The length of hospital stay records the total number of days the patient was hospitalized from admission to discharge. The complication occurrence marker records whether the patient experienced any complications during hospitalization; if complications occurred, it is marked as "yes," and if not, it is marked as "no." The readmission indicator records whether the patient was readmitted within 30 days of discharge for the same diagnosis. If readmitted, it is marked as yes; if not readmitted, it is marked as no.

[0047] Step 2 involves grouping and aggregating the medical orders for each record according to three dimensions: DIP grouping, clinical pathway node, and medical order combination. The DIP grouping dimension distinguishes different disease groups, the clinical pathway node dimension distinguishes different treatment stages within the same clinical pathway, and the medical order combination dimension distinguishes different sets of medical orders actually executed at the same clinical pathway node. Medical order combinations are formed by aggregating medical order items. Specifically, the aggregation method involves concatenating the codes of all medical order items executed by the same patient at the same clinical pathway node in a fixed order into a single combination identifier. Records with the same combination identifier are grouped into one medical order combination. After aggregation, a mapping relationship is formed between DIP grouping, pathway node, medical order combination, and cost. This mapping relationship allows the system to quickly query the historical average cost of any medical order combination at any clinical pathway node under any DIP group.

[0048] Step 3 involves normalizing and weighting the length of hospital stay, complication occurrence markers, and readmission status markers to generate a comprehensive efficacy score. Normalization uses a Min-Max normalization method, mapping each indicator to a range of 0 to 1, where 1 represents optimal efficacy and 0 represents worst efficacy. The normalization calculation for length of hospital stay is as follows: the maximum and minimum length of hospital stay for all historical cases under the DIP group in the dataset are taken. The normalized value of length of hospital stay equals the maximum value minus the actual length of hospital stay, divided by the difference between the maximum and minimum values. The shorter the length of hospital stay, the closer the normalized value is to 1. The complication occurrence marker is set as follows: 0 for complications and 1 for no complications. The readmission status marker is set as follows: 0 for readmission within 30 days and 1 for no readmission. The weight coefficients satisfy the condition that the sum of the three weights equals 1. One optional weight configuration is: length of hospital stay weight 0.4, complication occurrence marker weight 0.3, and readmission status marker weight 0.3. The overall efficacy score is equal to the normalized value of the length of hospital stay multiplied by its weight, plus the value of the complication occurrence marker multiplied by its weight, plus the value of the readmission marker multiplied by its weight. For example, consider a patient with a 10-day hospital stay, no complications, and no readmissions. Assuming the maximum historical length of hospital stay in this DIP group is 30 days and the minimum is 3 days, the normalized value of the length of hospital stay is (30 - 10) divided by (30 - 3), approximately equal to 0.74. The complication occurrence marker is set to 1, the readmission marker is set to 1, and the overall efficacy score is equal to 0.74 multiplied by 0.4 plus 1 multiplied by 0.3 plus 1 multiplied by 0.3, equaling 0.896.

[0049] Step 4 involves establishing a correlation between the overall efficacy score and the cost values ​​of the corresponding DIP group-pathway node-prescription combination, thus completing the construction of a cost-pathway-efficacy ternary association database. Each record in the database contains the following fields: DIP group code, clinical pathway node identifier, prescription combination identifier, historical average cost, and overall efficacy score. Through this ternary database, the system can quickly extract the cost values ​​and efficacy scores of each historical prescription combination at the current clinical pathway node during optimization, providing data support for the execution of the Pareto frontier algorithm.

[0050] Furthermore, it also includes a hybrid architecture that combines offline training with online inference: In the offline phase, historical data from the cost-path-efficacy ternary association database is used, with medical order combination features and individual feature data as inputs and comprehensive efficacy scores as outputs, to train an efficacy prediction model. Individual feature data includes age and comorbidity codes. During the online phase, the system acquires the target patient's current medical order combination, candidate medical order adjustment plans, and individualized characteristic data of the target patient. It then calls the trained efficacy prediction model and outputs the predicted efficacy score of each candidate plan for the current target patient. The predicted efficacy score for the current target patient is substituted into the multi-objective optimization solution process to generate a list of recommended adjustments to medical orders.

[0051] In this embodiment, the hybrid architecture of offline training and online inference is used to solve the technical problem that some medical order combinations in historical data cannot be directly calculated with stable efficacy scores due to insufficient sample size, while realizing personalized efficacy prediction based on the individual characteristics of the current target patient.

[0052] The offline workflow is as follows. Training data is extracted from the cost-path-efficacy ternary association database. Each training sample contains two parts: input features and output labels. Input features are composed of combined medical order features and individualized feature data. The specific construction method of the combined medical order features is as follows: medication category, examination items, and treatment operations are individually encoded using one-hot encoding to generate three binary vectors, which are then concatenated into a fixed-length combined feature vector. Individualized feature data includes age and comorbidity codes. Age is normalized to the 0-1 range using Min-Max, and comorbidity codes are generated into binary vectors using one-hot encoding. The output label is the comprehensive efficacy score corresponding to the historical record. The comprehensive efficacy score has been pre-calculated using a normalized weighted fusion method during the construction of the cost-path-efficacy ternary association database. The efficacy prediction model adopts a gradient boosting tree model. This model iteratively constructs multiple decision trees. Each new tree fits the residual of the previous round along the negative gradient direction of the loss function, gradually approximating the true comprehensive efficacy score value. The gradient boosting tree model is configured with the following parameters: 150 decision trees, a maximum tree depth of 6 layers, a learning rate of 0.05, and a loss function of mean squared error. Training data is divided into a training set (80%) and a test set (20%). Training is considered complete when the model's mean squared error on the test set is below 0.05 and its coefficient of determination is above 0.80. The trained model is serialized and stored on the server disk for loading and use during online inference. Offline training is performed at fixed intervals, such as monthly, using newly added historical case data for incremental updates or full retraining to ensure the model continuously adapts to changes in treatment patterns.

[0053] The workflow in the online phase is as follows. When the remaining payable amount for the target patient triggers the optimization initiation threshold, the system obtains the target patient's current medical order combination, a list of candidate medical order adjustment schemes, and the target patient's individualized characteristic data. The current medical order combination refers to the set of medical orders that the target patient has already initiated or plans to initiate at the current clinical pathway node. Candidate medical order adjustment schemes refer to all medical order combinations with historical data at the current clinical pathway node recorded in the cost-pathway-efficacy ternary association database. The system converts the current medical order combination and each candidate medical order adjustment scheme into a medical order combination feature vector using the same encoding method as in the offline phase. Then, it uses the target patient's normalized age value and the one-hot encoded vector of comorbidities as individualized feature vectors, concatenating them with the medical order combination feature vectors of each candidate scheme to form the complete input feature vector corresponding to each candidate scheme. The system loads the trained gradient boosting tree model and sequentially feeds the input feature vectors of each candidate scheme into the model for forward inference. The model outputs a predicted efficacy score for each candidate scheme for the current target patient. The predicted efficacy score is a continuous value between 0 and 1; a higher value indicates a better expected efficacy for the current target patient. The system substitutes the predicted efficacy scores of each candidate treatment into the multi-objective optimization solution process, converts the predicted efficacy scores into predicted efficacy loss values ​​and uses them as the vertical axis input of the Pareto front algorithm, and uses the cost value as the horizontal axis input. It then performs Pareto front solution and cost target constraint screening, and finally generates a personalized medical order adjustment recommendation list based on the individual characteristics of the current target patient.

[0054] Taking a hospitalized patient diagnosed with acute exacerbation of chronic obstructive pulmonary disease (COPD), aged 68, with comorbidities including hypertension and diabetes, as an example. At the clinical pathway node of anti-infective therapy, the cost-pathway-efficacy ternary association database records five historical medical order combinations. During the online inference phase, the system normalizes the patient's age (68 years) and uses the one-hot codes for hypertension and diabetes as individualized feature vectors. These are concatenated with the feature vectors of the five candidate medical order combinations and input into a gradient boosting tree model. The model outputs five predicted efficacy scores. Specifically, medical order combination A has a historical comprehensive efficacy score of 0.88 for younger patients without comorbidities, but considering the current patient's age of 68 and comorbidities (hypertension and diabetes), the model lowers its predicted efficacy score to 0.72. Medical order combination B has a historical comprehensive efficacy score of 0.80 for elderly patients with multiple underlying diseases, and the model's predicted efficacy score for the current patient is 0.78. After converting all five predicted efficacy scores into predicted efficacy loss values, the system inputs them, along with the estimated cost values ​​of each candidate treatment, into the Pareto frontier solution process to generate a personalized recommendation list that reflects the individual characteristics of the current patient.

[0055] Furthermore, it also includes: When a doctor selects a recommended option from the list of recommended adjustments to medical orders, the doctor records a comparison between the adopted option and the non-adopted option. When a doctor submits a custom prescription combination instead of the recommended plan, record the custom prescription combination information and the reason for not adopting it; The adoption and non-adoption records will be written back to the cost-path-efficacy ternary association database.

[0056] In this embodiment, after the list of recommended medical orders is pushed to the doctor's workstation, the system records the doctor's adoption and non-adoption behaviors respectively, forming a complete data loop from recommendation to feedback.

[0057] When a doctor selects a recommended option from the list of recommended prescription adjustments, the system automatically records complete information about this adoption event. The recorded information includes: the target patient's visit identifier, the current clinical pathway node identifier, the prescription combination identifier of the adopted option, the estimated cost of the adopted option, the predicted efficacy score of the adopted option, the prescription combination identifiers and predicted efficacy scores of each option in the list of non-adopted options, and the adoption timestamp. By recording the comparison information between adopted and non-adopted options, the system accumulates real decision-making data from doctors regarding the trade-offs between cost and efficacy.

[0058] When a doctor submits a custom-defined combination of prescriptions instead of any of the recommended options, the system displays a window prompting the doctor to select or fill in a reason for non-adoption. Options for non-adoption include: the recommended option is not suitable for the patient's specific condition; a prescription in the recommended option has a contraindication; the medication in the recommended option is currently unavailable in the hospital; or the doctor believes, based on clinical experience, that the option needs adjustment. The system records the following: the target patient's visit identifier, the current clinical pathway node identifier, the detailed list of prescription items in the custom-defined combination, the estimated cost of the custom-defined combination, the prescription combination identifiers for all options in the recommended list, the reason for non-adoption entered by the doctor, and the submission timestamp.

[0059] The system periodically writes both adopted and unadopted records back to the cost-pathway-efficacy ternary association database. During the write-back, the usage frequency count of adopted medical order combinations in the database increases by one, and the corresponding historical case sample size increases accordingly. This further enhances the statistical stability of the scheme when recalculating the comprehensive efficacy score during subsequent offline training. Custom medical order combinations in unadopted records are added as new medical order combination records to the cost-pathway-efficacy ternary association database. If the custom medical order combination does not yet exist in the database, a new record is created, with the initial comprehensive efficacy score temporarily left blank. The efficacy data will be filled in after the patient is discharged based on the actual length of hospital stay, the occurrence of complications, and whether readmission occurs. If the custom medical order combination already exists, its usage frequency count is updated. Through this write-back mechanism, the cost-pathway-efficacy ternary association database continuously accumulates new clinical practice data as the system continues to operate, continuously expanding the training data foundation for the efficacy prediction model. The subsequently generated medical order adjustment recommendation list becomes increasingly closer to real clinical decision-making needs.

[0060] Furthermore, it also includes: At key diagnosis and treatment nodes in the clinical pathway, the core medical system compliance verification is automatically executed. The verification rules include verification of transfer time limits, verification of blood transfusion effect evaluation writing, and verification of discharge criteria for Level 1 nursing care. If the verification fails, a task to be done is generated and pushed to the doctor's workstation, and the system compliance status is marked in the path execution record; The hospital information system collects the actual number of days of hospitalization for target patients in real time. When the remaining payable amount has not triggered the optimization start threshold, but the actual number of days of hospitalization exceeds the standard number of days of hospitalization set by the clinical pathway and the average daily cost is higher than the preset average daily cost warning line, an overdue hospitalization cost warning is triggered.

[0061] In this embodiment, in addition to cost monitoring, the system introduces two auxiliary monitoring dimensions: core medical system compliance verification and overdue hospitalization cost early warning, forming a comprehensive management and control system covering cost, quality, and efficiency.

[0062] The core medical system compliance verification is automatically triggered at key treatment nodes in the clinical pathway. Key treatment nodes refer to stages in the clinical pathway involving patient transfer or important treatment decisions. The transfer time limit verification is executed when the patient has been admitted for 48 hours or when the doctor submits a transfer application. The system checks whether the patient has stayed in the current department for more than 48 hours and whether the transfer application process has been completed. If a transfer application has not been submitted or the transfer handover has not been completed within 48 hours, the verification fails. The transfusion effect evaluation writing verification is executed 24 hours after the transfusion order is executed. The system checks whether a transfusion effect evaluation record has been entered in the medical record. If no such record is found, the verification fails. The Level 1 nursing discharge criteria verification is executed when the doctor submits a discharge application for a Level 1 nursing patient. The system checks whether the patient meets the preset discharge criteria, such as stable vital signs and satisfactory daily living ability scores. If these criteria are not met, the verification fails. When the verification fails, the system automatically generates a task to be pushed to the to-do list of the responsible doctor and nurse's workstation. The task includes a detailed explanation of the reason for the verification failure, reference to relevant policy clauses, and a deadline for rectification. At the same time, the system marks the compliance status of this node as non-compliant in the clinical pathway execution record, and records the specific verification rule name and trigger time of the non-compliance. This marking information is synchronized to the quality control backend of the hospital management department, which can help the management department to keep track of the implementation status of core policies in each department in real time.

[0063] The overstay-of-hospitalization cost warning is a supplementary monitoring mechanism operating independently, separate from the cost dimension. The system collects the actual number of days of hospitalization for target patients in real time from the hospital information system. The actual number of days of hospitalization is calculated from the patient's admission date and automatically updated at midnight every day. The system triggers an overstay-of-hospitalization cost warning when the following three conditions are met simultaneously: 1. The remaining payable amount has not triggered the optimization activation threshold, meaning the current cost has not yet reached the cost warning line; 2. The actual number of days of hospitalization exceeds the standard number of days of hospitalization set by the clinical pathway. For example, the standard number of days of hospitalization set by the clinical pathway for a certain disease is 7 days, but the patient has actually been hospitalized for 9 days; 3. The average daily cost is higher than the preset average daily cost warning line. The average daily cost equals the current total cost divided by the actual number of days of hospitalization. The average daily cost warning line is set by hospital administrators based on the historical average daily cost data for each DIP group. Meeting all three conditions simultaneously means that although the total cost has not yet exceeded the standard, the patient's hospitalization time has exceeded the standard and the cost is still consuming at a high rate every day. Based on the current trend of the average daily cost, the total cost is highly likely to exceed the payment standard limit during subsequent hospitalizations. At this point, the system triggers a warning for excessive hospital stay costs, issuing a notification on the doctor's workstation interface. The notification includes the actual number of days hospitalized, the standard number of days of hospitalization, the current average daily cost, the daily cost warning line, and the estimated number of days and amount of overspending calculated based on the current average daily cost. This warning mechanism allows doctors to recognize the risk of cost overruns due to abnormally prolonged hospital stays even before the actual cost exceeds the limit, enabling them to promptly assess whether the patient meets discharge criteria or whether the treatment plan needs to be adjusted to shorten the hospital stay.

[0064] The triggering conditions for the above-mentioned core medical system compliance verification and inpatient overdue cost warning functions are independent of each other. Failure to pass the verification or triggering the warning will not affect the operation of the main cost monitoring process, but will generate corresponding prompt information and push it to the doctor's workstation, forming a multi-dimensional and multi-level clinical pathway execution monitoring network.

[0065] Furthermore, it also includes: When a target patient triggers a branch switch within the clinical pathway or a cross-pathway conversion during the execution of the clinical pathway, the path identifier before conversion, the path identifier after conversion, the reason for conversion, and the conversion timestamp are recorded. The new DIP grouping and new DIP payment standard information corresponding to the converted path are obtained, and the calculation basis of the remaining payable amount is reset based on the new DIP payment standard information. When the DIP payment standard information changes due to policy adjustments, update the DIP payment standard information, re-execute the multi-objective optimization solution, and mark the schemes in the medical order adjustment recommendation list that no longer meet the cost target constraint due to the change in payment standard as invalid.

[0066] In this embodiment, the system sets up an adaptive mechanism for path conversion and an adaptive mechanism for policy adjustment for two types of change scenarios that may occur during the execution of clinical pathways, to ensure that cost monitoring and optimization recommendations continue to operate effectively in a dynamically changing clinical and management environment.

[0067] The adaptive pathway transition mechanism handles scenarios where target patients trigger intra-pathway branch switching or cross-pathway transitions during clinical pathway execution. Intra-pathway branch switching refers to a patient changing from their current branch to another within the same clinical pathway due to a change in their condition. For example, in a community-acquired pneumonia clinical pathway, a patient initially enters the general treatment branch but is later transferred to the intensive care branch due to respiratory failure. Cross-pathway transition refers to a patient completely exiting their current clinical pathway and entering another. For example, a patient admitted for acute myocardial infarction and entering the corresponding clinical pathway, then discovering concurrent acute ischemic stroke during treatment, transitions from the acute myocardial infarction pathway to the acute ischemic stroke pathway. When the aforementioned conversion event occurs, the system automatically performs the following operations: records the path identifier before conversion, the path identifier after conversion, the reason for conversion, and the conversion timestamp, forming a complete conversion trajectory log in the path execution record. The reason for conversion is selected by the doctor from preset options or manually entered; obtains the new DIP group and new DIP payment standard information corresponding to the converted path. If the DIP group associated with the converted path is different from that before conversion, the standard payment amount and payment multiplier range in the new DIP payment standard information may change; resets the calculation benchmark of the remaining payable amount based on the standard payment amount in the new DIP payment standard information, recalculating the remaining payable amount to be equal to the new standard payment amount minus the current total cost. Subsequent cost monitoring and early warning judgments are all performed based on the reset remaining payable amount. Taking a patient initially admitted to the general treatment branch for community-acquired pneumonia as an example, the patient's initial DIP group corresponding to the standard payment amount is 8,000 yuan, and 5,000 yuan has been incurred on the 3rd day of hospitalization, leaving a remaining payable amount of 3,000 yuan. On the 4th day, the patient was transferred to the intensive care unit due to respiratory failure. The standard payment limit for the severe pneumonia DIP group associated with the new pathway is 18,000 yuan. The remaining payable limit after the reset is 18,000 yuan minus the 5,000 yuan already incurred, which equals 13,000 yuan. The system uses 13,000 yuan as the new benchmark to re-execute the multi-level early warning threshold comparison and optimization trigger judgment.

[0068] The policy adjustment adaptive mechanism handles scenarios where DIP payment standard information changes due to annual policy adjustments by healthcare security departments. Healthcare security departments in various regions typically release an updated version of the DIP disease catalog and unit price adjustment plan annually, leading to increases or decreases in the standard payment amount for some diseases. When the DIP payment standard information in the hospital information system changes, the system automatically performs the following operations: updates the stored DIP payment standard information, synchronizing the new standard payment amount to the dynamic settlement and multi-level early warning module; for all inpatients currently following clinical pathways and with generated prescription adjustment recommendation lists, re-executes the multi-objective optimization solution process, using the updated standard payment amount as the calculation parameter in the cost compliance constraint; after re-solving, schemes in the original prescription adjustment recommendation list that no longer meet the cost compliance constraint due to the payment standard change are marked as invalid. Invalid schemes are displayed in gray font with an explanation of the invalidation in the recommendation list on the doctor's workstation interface, and doctors can no longer select invalid schemes. Taking a patient whose standard payment limit is reduced from 15,000 yuan to 13,000 yuan as an example, the estimated total cost of a certain plan in the original recommended list was 14,000 yuan. Under the original standard, this met the cost compliance constraint. However, after the standard reduction, 14,000 yuan is greater than 13,000 yuan, so this plan is marked as invalid. At the same time, the system may replace another plan in the original recommended list with an estimated total cost of 12,500 yuan as the preferred plan in the recommendation ranking. The above mechanism ensures that after the DIP payment policy changes, the system's cost monitoring benchmarks and medical order recommendations for inpatients remain consistent with the latest policy, avoiding the risk of cost overruns due to policy lag.

[0069] This invention provides an embodiment of a real-time monitoring system for clinical pathway costs based on the DIP payment standard, comprising: The disease-payment standard matching module is used to store the DIP grouping codes and DIP payment standard information associated with the clinical pathway forms; A cost-pathway-efficacy ternary association database is used to store historical patient medical order combinations, cost data, and efficacy indicator data by DIP grouping and clinical pathway node aggregation; The offline training module is used to train the efficacy prediction model using historical data from the cost-path-efficacy ternary association database, with medical order combination features and individual feature data as input and comprehensive efficacy score as output. The real-time cost collection module is used to establish a data interface with the hospital information system, and to obtain detailed data on the treatment costs of the target patients in real time after the clinical pathway is started, and to accumulate and generate the current total cost. The dynamic settlement and multi-level early warning module is used to dynamically calculate the remaining payable amount based on the standard payment amount in the DIP payment standard information and the current total cost generated by the real-time cost collection module. It compares the remaining payable amount with the preset multi-level early warning threshold in real time and issues corresponding early warning notifications. When the remaining payable amount triggers the optimization start threshold in the multi-level early warning threshold, it obtains the clinical pathway node information of the target patient and outputs the optimization signal. The multi-objective optimization solution module is used to receive optimization signals, extract historical cost data and historical efficacy index data corresponding to the current clinical path node from the cost-path-efficacy ternary association database, call the efficacy prediction model generated by the offline training module, and generate a list of recommended medical order adjustments with the Pareto front algorithm under the premise of meeting the cost target constraint and minimizing efficacy loss. The recommendation push and adoption module is used to push the list of recommended medical order adjustments to the doctor's workstation, receive the doctor's adoption selection or custom medical order combination, and write the adoption record back to the cost-path-efficacy three-element association database; The path conversion and policy adaptation module is used to detect branch switching or cross-path conversion events during the execution of clinical pathways, record the path identifiers and reasons for conversion before and after conversion, obtain the new DIP grouping and new DIP payment standard information corresponding to the converted path, and notify the dynamic settlement and multi-level early warning module to reset the calculation benchmark. When the DIP payment standard information changes due to policy adjustments, the DIP payment standard information is updated, triggering the multi-objective optimization solution module to resolve and mark the failed solution. The core system compliance and inpatient monitoring module is used to perform core medical system compliance verification at key diagnosis and treatment nodes of the clinical pathway. The verification rules include verification of transfer time limits, verification of blood transfusion effect evaluation writing, and verification of discharge criteria for Level 1 nursing care. When the verification fails, a task is generated and pushed to the doctor's workstation. The module collects the actual number of days of hospitalization of the target patient in real time from the hospital information system. When the remaining payable amount has not triggered the optimization start threshold, but the actual number of days of hospitalization exceeds the standard number of days of hospitalization set by the clinical pathway and the average daily cost is higher than the preset average daily cost warning line, an overdue hospitalization cost warning is triggered.

[0070] In this embodiment, the system consists of eight functional modules, each deployed on a hospital server or cloud platform. Communication between modules is achieved through message queues and RESTful interfaces, and data exchange with the hospital information system and electronic medical record system is conducted via HL7 standard interfaces or WebService interfaces. The specific implementation methods of each module are described below.

[0071] The disease-payment standard matching module stores the DIP group codes and DIP payment standard information associated with clinical pathway forms. This module maintains a disease-pathway-payment standard mapping table, using the DIP group code as the primary key, and associates the clinical pathway form number, path name, standard payment amount, payment multiplier range, and version validity period under that group. When the healthcare security department releases the annual DIP disease catalog and score-based unit price adjustment plan, the hospital's medical insurance management department imports the updated payment standard data in batches through the module's management backend. The system automatically compares the differences between the old and new versions and generates a change log. When a doctor initiates a clinical pathway for a patient, this module queries the corresponding DIP group code based on the patient's admission diagnosis code and returns the clinical pathway forms associated with that group for the doctor to choose from.

[0072] The cost-pathway-efficacy ternary association database stores historical patient prescription combinations, cost data, and efficacy index data aggregated by DIP grouping and clinical pathway nodes. Deployed on the hospital's data server, the database uses a relational database engine with DIP grouping codes as the first-level index key, clinical pathway node identifiers as the second-level index key, and prescription combination hash values ​​as the third-level index key, enabling fast retrieval through a three-level composite index. Each record in the database contains the following fields: DIP grouping code, clinical pathway node identifier, prescription combination identifier, detailed list of prescription items, historical average cost, number of sample cases, and overall efficacy score. The number of sample cases field indicates the number of historical cases corresponding to the prescription combination. When the number of sample cases is lower than a preset minimum sample size threshold, the overall efficacy score of the prescription combination is marked as statistically insufficient, and the predicted value from the efficacy prediction model is preferentially used instead of the directly calculated value during multi-objective optimization.

[0073] The offline training module is used to train the efficacy prediction model using historical data from the cost-path-efficacy ternary association database. This module runs as a scheduled task in the background, performing a full training run monthly by default, but can also be manually triggered for immediate training by hospital administrators. Upon training startup, the module reads all records from the cost-path-efficacy ternary association database within the past three years with a sample number not less than a preset minimum threshold, dividing the data into an 80% training set and a 20% test set. The model uses a gradient boosting tree model with parameters set to 150 decision trees, a maximum tree depth of 6 layers, a learning rate of 0.05, and a mean squared error loss function. The input feature dimension is determined by adding the one-hot encoding dimension of the medical order combination features and the encoding dimension of the individualized feature data. The output is a continuous comprehensive efficacy score between 0 and 1. The module calculates the mean squared error and coefficient of determination on the test set. Training is considered converged when the mean squared error is below 0.05 and the coefficient of determination is above 0.80. The model parameters are then serialized and saved to the file system, and the model version number is updated. After training is completed, the module sends a model update notification to the dynamic settlement and multi-level early warning module and the multi-objective optimization solution module. Upon receiving the notification, each module reloads the latest version of the model file.

[0074] The real-time cost collection module establishes a data interface with the hospital information system to acquire detailed treatment cost data for target patients in real time after the clinical pathway is initiated. This module deploys a data collection agent on the hospital information system side. The agent monitors the cost transaction table of the hospital information system, and when a new cost record is detected, it pushes the detailed cost data to the real-time cost collection module in real time via a message queue. The pushed data includes the cost generation time, cost category code, cost item name, cost amount, department code, and executing department code. After receiving the data, the module verifies whether the target patient for the cost record is in a clinical pathway execution state. If so, it adds the cost amount to the patient's current total cost and pushes the current total cost to the dynamic settlement and multi-level early warning module.

[0075] The dynamic settlement and multi-level early warning module is used to perform the core calculation logic for cost monitoring. Each time the module receives a new push notification of the current total cost, it recalculates the remaining payable amount based on the standard payment amount in the current patient's DIP payment standard information. The remaining payable amount equals the standard payment amount minus the current total cost. The module compares the remaining payable amount sequentially with preset reminder thresholds, warning thresholds, and optimization activation thresholds. When the remaining payable amount is lower than the reminder threshold but higher than the warning threshold, a reminder notification is generated and sent to the message component on the doctor's workstation via push notification. When the remaining payable amount is lower than the warning threshold but higher than the optimization activation threshold, a warning notification is generated and sent to the doctor's workstation with higher priority push notification. When the remaining payable amount is lower than the optimization activation threshold, the module obtains the target patient's current clinical pathway node information, encapsulates it into an optimization signal, and includes the patient's visit identifier, current clinical pathway node identifier, current total cost, remaining payable amount, and DIP payment standard information. This signal is sent to the multi-objective optimization solution module via a RESTful interface.

[0076] The multi-objective optimization solution module receives optimization signals and generates a list of recommended medical order adjustments. Upon receiving the optimization signal, the module extracts cost data and comprehensive efficacy scores for all historical medical order combinations at the current clinical pathway node from the cost-pathway-efficacy ternary association database, using the current clinical pathway node identifier as the query condition. The module then calls the latest version of the efficacy prediction model generated by the offline training module to obtain individualized characteristic data of the target patient, including age and comorbidity codes, and calculates the predicted efficacy score for each candidate medical order combination for the current target patient. Using the cost value of each candidate medical order combination as the x-axis and 1 minus the predicted efficacy score as the y-axis, the module executes the Pareto front algorithm to filter the set of non-dominated points, and then filters out schemes exceeding the product of the standard payment amount and the preset compliance coefficient under cost compliance constraints. Finally, it generates a list of recommended medical order adjustments, sorted in ascending order of efficacy loss. The recommended list is sent to the recommendation push and adoption module via a message queue.

[0077] The recommendation push and adoption module is used to push the list of recommended medical orders to the doctor's workstation and collect the doctor's feedback. The module pushes the recommendation list to the front-end interface of the doctor's workstation in real time via a WebSocket long connection, which is displayed as a pop-up window or sidebar. The module continuously monitors doctor actions on the recommendation list. When it detects that a doctor clicks the "adopt" button to select a recommended plan, it records the adopted plan information; when it detects that a doctor closes the recommendation window and submits the medical order themselves, a pop-up window for collecting reasons for non-adoption appears, recording the custom medical order combination information and the reason for non-adoption. Both adoption and non-adoption records are written back to the cost-path-efficacy ternary relational database through the module.

[0078] The path conversion and policy adaptation module handles two types of dynamic change scenarios. The module detects path conversions by subscribing to patient transfer events and diagnosis change events in the hospital information system. When a branch switch within a path or a cross-path conversion is detected, it automatically records the path identifier before and after the conversion, the reason for the conversion, and the conversion timestamp. The conversion event is written to the path execution log. The module retrieves the payment standard information corresponding to the new path by querying the disease-payment standard matching module and sends the new standard payment amount to the dynamic settlement and multi-level early warning module to reset the calculation benchmark. Simultaneously, the module monitors payment standard update events from the disease-payment standard matching module. Upon detecting an update, it iterates through all inpatients, synchronizes the updated standard payment amount to the dynamic settlement and multi-level early warning module, sends a re-solution instruction to the multi-objective optimization solution module, and notifies the recommendation push and adoption module of solutions marked as invalid after re-solution. The recommendation push and adoption module then updates the display status of the recommendation list on the doctor's workstation.

[0079] The core compliance and inpatient monitoring module provides auxiliary monitoring functions beyond the cost dimension. The module pre-configures verification rules and trigger conditions for each key treatment node within the clinical pathway template. For example, the transfer time limit verification is triggered when a patient has been admitted for 48 hours, checking for a valid transfer request record. The transfusion effect evaluation documentation verification is triggered 24 hours after the transfusion order is executed, checking for the existence of a transfusion effect evaluation document in the medical record. The Level 1 nursing discharge criteria verification is triggered when the doctor submits a Level 1 nursing discharge request, checking whether vital signs data meet the discharge criteria. Tasks that fail verification are generated and sent to the doctor's workstation via push notification, while a non-compliance mark is added to the clinical pathway execution record. The 48-hour time limit used in the above-mentioned department transfer time limit verification is an optional configuration example. The specific time limit value for department transfer time limit verification can be configured according to the medical system requirements of each department. For example, it can be configured as 48 hours in the emergency and critical care department and 72 hours in the general ward department. The system reads the preset time limit configuration parameters of each clinical pathway form from the disease-payment standard matching module and dynamically determines the time limit threshold for verification during runtime. The module automatically collects the actual number of hospitalization days of all inpatients at midnight every day and compares it with the standard number of hospitalization days set for each patient's current clinical pathway. When it detects that the actual number of hospitalization days exceeds the standard number of hospitalization days, the remaining payable amount has not triggered the optimization start threshold, and the average daily cost is higher than the preset average daily cost warning line, a hospitalization overdue cost warning is generated and sent to the doctor's workstation via message push.

[0080] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for real-time monitoring of clinical pathway costs based on DIP payment standards, characterized by, include: Obtain the DIP grouping information and corresponding DIP payment standard information for the target patient's current admission; The system collects detailed data on the medical expenses already incurred by the target patient in real time from the hospital information system and calculates the current total cost. Based on the standard payment limit in the DIP payment standard information and the current total cost, the remaining payable amount is dynamically calculated; When the remaining payable amount triggers the optimized activation threshold among the preset multi-level warning thresholds, obtain the clinical pathway node information of the target patient; From the cost-pathway-efficacy ternary correlation database, extract historical cost data and historical efficacy index data corresponding to different combinations of medical orders for historical patients at the current clinical pathway node; Under the premise of meeting the cost target constraint, with the goal of minimizing the loss of efficacy, the optional medical order adjustment plan at the current clinical pathway node is optimized by multi-objective solution to generate a recommended list of medical order adjustments. The cost target constraint is that the adjusted expected total cost is less than or equal to the standard payment amount in the DIP payment standard information. The list of recommended adjustments to medical orders will be pushed to the doctor's workstation.

2. The method for real-time monitoring of clinical pathway costs based on the DIP payment standard according to claim 1, characterized in that, The multi-level early warning threshold includes an alert threshold, a warning threshold, and an optimization activation threshold. The alert threshold is higher than the warning threshold, and the warning threshold is higher than the optimization activation threshold. When the remaining available credit falls below the alert threshold, a fee reminder notification will be sent. A fee warning notification is issued when the remaining payable amount falls below the warning threshold.

3. The method for real-time monitoring of clinical pathway costs based on the DIP payment standard according to claim 1, characterized in that, Historical efficacy indicators consist of one or more of the following: length of hospital stay, complication rate, and readmission rate; the cost target constraint is that the adjusted total estimated cost is less than or equal to the product of the standard payment amount in the DIP payment standard information and the preset target coefficient.

4. The method for real-time monitoring of clinical pathway costs based on the DIP payment standard according to claim 1, characterized in that, The Pareto front algorithm is used to solve the multi-objective optimization problem. Specifically: Map each historical medical order combination at the current clinical pathway node to a point in a two-dimensional space, with the horizontal axis representing the cost value and the vertical axis representing the loss of efficacy. Select the Pareto front point set from the mapping points that is not jointly dominated by other points in both the x-axis and y-axis; From the Pareto frontier cluster, further screening was conducted to identify the corresponding medical order adjustment plans for points that met the cost target constraints, and these were then sorted in ascending order of efficacy loss.

5. The method for real-time monitoring of clinical pathway costs based on the DIP payment standard according to claim 1, characterized in that, The process of constructing the cost-pathway-efficacy ternary association database includes: Extract clinical pathway execution records of discharged cases under the DIP group from the hospital's historical medical record system. Each record contains details of the medical orders actually executed by the patient at the clinical pathway node, corresponding cost details, length of hospital stay, complication occurrence markers, and readmission markers. Each record's medical order items are grouped and aggregated according to three dimensions: DIP grouping, clinical pathway node, and medical order combination. The medical order combination is formed by the aggregation of medical order items, forming a mapping relationship of DIP grouping - pathway node - medical order combination - cost. The length of hospital stay, the occurrence of complications, and the readmission criteria are normalized and weighted to generate a comprehensive efficacy score. The comprehensive efficacy score is linked to the cost value of the corresponding DIP group-path node-medical order combination to form a cost-path-efficacy ternary association database.

6. The method for real-time monitoring of clinical pathway costs based on the DIP payment standard according to claim 5, characterized in that, It also includes a hybrid architecture that combines offline training with online inference: In the offline phase, historical data from the cost-path-efficacy ternary association database is used, with medical order combination features and individual feature data as inputs and comprehensive efficacy scores as outputs, to train an efficacy prediction model. Individual feature data includes age and comorbidity codes. During the online phase, the system acquires the target patient's current medical order combination, candidate medical order adjustment plans, and individualized characteristic data of the target patient. It then calls the trained efficacy prediction model and outputs the predicted efficacy score of each candidate plan for the current target patient. The predicted efficacy score for the current target patient is substituted into the multi-objective optimization solution process to generate a list of recommended adjustments to medical orders.

7. The method for real-time monitoring of clinical pathway costs based on the DIP payment standard according to claim 1, characterized in that, Also includes: When a doctor selects a recommended option from the list of recommended adjustments to medical orders, the doctor records a comparison between the adopted option and the non-adopted option. When a doctor submits a custom prescription combination instead of the recommended plan, record the custom prescription combination information and the reason for not adopting it; The adoption and non-adoption records will be written back to the cost-path-efficacy ternary association database.

8. The method for real-time monitoring of clinical pathway costs based on the DIP payment standard according to claim 1, characterized in that, Also includes: At key diagnosis and treatment nodes in the clinical pathway, the core medical system compliance verification is automatically executed. The verification rules include verification of transfer time limits, verification of blood transfusion effect evaluation writing, and verification of discharge criteria for Level 1 nursing care. If the verification fails, a task to be done is generated and pushed to the doctor's workstation, and the system compliance status is marked in the path execution record; The hospital information system collects the actual number of days of hospitalization for target patients in real time. When the remaining payable amount has not triggered the optimization start threshold, but the actual number of days of hospitalization exceeds the standard number of days of hospitalization set by the clinical pathway and the average daily cost is higher than the preset average daily cost warning line, an overdue hospitalization cost warning is triggered.

9. The method for real-time monitoring of clinical pathway costs based on the DIP payment standard according to claim 1, characterized in that, Also includes: When a target patient triggers a branch switch within the clinical pathway or a cross-pathway conversion during the execution of the clinical pathway, the path identifier before conversion, the path identifier after conversion, the reason for conversion, and the conversion timestamp are recorded. The new DIP grouping and new DIP payment standard information corresponding to the converted path are obtained, and the calculation basis of the remaining payable amount is reset based on the new DIP payment standard information. When the DIP payment standard information changes due to policy adjustments, update the DIP payment standard information, re-execute the multi-objective optimization solution, and mark the schemes in the medical order adjustment recommendation list that no longer meet the cost target constraint due to the change in payment standard as invalid.

10. A real-time monitoring system for clinical pathway costs based on the DIP payment standard, characterized in that, include: The disease-payment standard matching module is used to store the DIP grouping codes and DIP payment standard information associated with the clinical pathway forms; A cost-pathway-efficacy ternary association database is used to store historical patient medical order combinations, cost data, and efficacy indicator data by DIP grouping and clinical pathway node aggregation; The offline training module is used to train the efficacy prediction model using historical data from the cost-path-efficacy ternary association database, with medical order combination features and individual feature data as input and comprehensive efficacy score as output. The real-time cost collection module is used to establish a data interface with the hospital information system, and to obtain detailed data on the treatment costs of the target patients in real time after the clinical pathway is started, and to accumulate and generate the current total cost. The dynamic settlement and multi-level early warning module is used to dynamically calculate the remaining payable amount based on the standard payment amount in the DIP payment standard information and the current total cost generated by the real-time cost collection module. It compares the remaining payable amount with the preset multi-level early warning threshold in real time and issues corresponding early warning notifications. When the remaining payable amount triggers the optimization start threshold in the multi-level early warning threshold, it obtains the clinical pathway node information of the target patient and outputs the optimization signal. The multi-objective optimization solution module is used to receive optimization signals, extract historical cost data and historical efficacy index data corresponding to the current clinical path node from the cost-path-efficacy ternary association database, call the efficacy prediction model generated by the offline training module, and generate a list of recommended medical order adjustments with the Pareto front algorithm under the premise of meeting the cost target constraint and minimizing efficacy loss. The recommendation push and adoption module is used to push the list of recommended medical order adjustments to the doctor's workstation, receive the doctor's adoption selection or custom medical order combination, and write the adoption record back to the cost-path-efficacy three-element association database; The path conversion and policy adaptation module is used to detect branch switching or cross-path conversion events during the execution of clinical pathways, record the path identifiers and reasons for conversion before and after conversion, obtain the new DIP grouping and new DIP payment standard information corresponding to the converted path, and notify the dynamic settlement and multi-level early warning module to reset the calculation benchmark. When the DIP payment standard information changes due to policy adjustments, the DIP payment standard information is updated, triggering the multi-objective optimization solution module to resolve and mark the failed solution. The core system compliance and inpatient monitoring module is used to perform core medical system compliance verification at key diagnosis and treatment nodes in the clinical pathway. The verification rules include verification of transfer time limits, verification of blood transfusion effect evaluation writing, and verification of discharge criteria for Level 1 nursing care. When the verification fails, a task to be done is generated and pushed to the doctor's workstation. The hospital information system collects the actual number of days of hospitalization for target patients in real time. When the remaining payable amount has not triggered the optimization start threshold, but the actual number of days of hospitalization exceeds the standard number of days of hospitalization set by the clinical pathway and the average daily cost is higher than the preset average daily cost warning line, an overdue hospitalization cost warning is triggered.

Images