A multi-hospital intelligent medical resource scheduling method and system

Abstract

The application discloses a kind of multi-institutional medical resource intelligent scheduling method and scheduling system.The method first acquires multi-institutional medical resource data and historical scheduling records, excavates diagnosis and treatment collaborative relationship, and constructs resource dependency graph with dependency strength.Second, analyze patient diagnosis and treatment orders to generate structured demand list, and construct directed acyclic appointment chain after loop verification.Combined with patient characteristics and clinical pathway constraints, predict bed availability, and build a hierarchical appointment resource pool.Through multi-dimensional comprehensive score combined with dynamic weight mechanism, complete preliminary resource scheduling, and for capacity, space and time sequence link conflicts, rely on diagnosis and treatment urgency game equilibrium algorithm to complete conflict resolution and appointment chain correction.Finally, execute the appointment instruction, monitor resource abnormalities in real time, and trigger dynamic rescheduling based on the threshold of transmission influence.Six functional modules are provided in the supporting system to realize the above processes, and precise, collaborative and dynamic intelligent scheduling of multi-institutional medical resources can be achieved.

Description

Technical Field

[0001] This invention belongs to the field of medical information technology, specifically relating to a method and system for intelligent scheduling of medical resources across multiple hospital areas. Background Technology

[0002] With the rapid development of medical group systems, large hospitals generally adopt a "one hospital, multiple campuses" management model, with multiple campuses under the same medical group. Each campus differs in geographical location, specialty characteristics, and equipment configuration. Patients typically need to traverse multiple types of medical resources during the diagnosis and treatment process, such as from outpatient visits to medical technology examinations, from diagnosis to bed admission, and from bed admission to surgery, forming a complete disease treatment chain with strict time-dependent relationships.

[0003] Existing medical resource allocation systems generally adopt an independent management model for each resource type. Outpatient appointment systems, medical technology appointment systems, bed management systems, and surgical scheduling systems are each managed by different subsystems, resulting in data incompatibility and asynchronous status between them. This independent management model leads to the following prominent problems: When a doctor issues an admission certificate to a patient in the outpatient system, the outpatient system cannot automatically detect the bed occupancy status of the target department in the bed management system, and therefore cannot reserve a bed for the patient at the same time as issuing the admission certificate; when the surgical scheduling system determines the surgery time, the bed management system cannot automatically reserve a bed for the day before the surgery, potentially leaving patients without a bed when they are admitted the day before surgery; when a patient needs to complete multiple medical technology examinations on the same day, the appointment systems for each examination are unaware of each other and cannot detect time conflicts, potentially resulting in the patient being scheduled for two different examinations at the same time.

[0004] The root cause of these problems lies in the lack of a unified modeling of the dependencies between four types of medical resources—outpatient services, medical technology, hospital beds, and surgeries—in existing systems. This prevents the coordinated scheduling of a patient's entire treatment process as a whole. During a single treatment session, patients need to use multiple types of medical resources sequentially or in parallel, and these resources have clear temporal dependencies. For example, surgery depends on hospital beds, hospital beds depend on pre-operative examinations, and pre-operative examinations depend on outpatient assessments. These dependencies form a directed acyclic chain of treatment. However, existing systems fragment this chain into multiple independent single-point appointments. Each single-point appointment only considers the availability of the current resource type, neglecting the impact of upstream resource appointment results on downstream resource availability, and ignoring the constraints of downstream resource demand on upstream resource appointment time slots. When a time shift or resource anomaly occurs at any stage, the system cannot automatically detect the transmission impact on other stages of the chain, nor can it trigger a global rescheduling. This ultimately leads to frequent problems such as patients repeatedly rescheduling, a coexistence of idle and strained resources, and delays in the treatment process. Summary of the Invention

[0005] In view of this, the present invention provides a method and system for intelligent scheduling of medical resources in multiple hospital areas, in order to solve the problem that existing systems lack a unified model of the dependencies between four types of medical resources: outpatient, medical technology, beds, and surgery, and are unable to coordinate and schedule the patient's full-course diagnosis and treatment needs as a whole.

[0006] The technical solution adopted in this invention is as follows:

[0007] A method for intelligent scheduling of medical resources across multiple hospital campuses includes the following steps:

[0008] Step 1: Collect medical resource data and historical scheduling records from multiple hospital campuses, abstract each medical resource into a resource unit, mine the diagnostic and treatment collaboration dependencies between resource units from the historical scheduling records, and construct a resource dependency relationship map with dependency strength;

[0009] Step 1 specifically includes:

[0010] By using the data interface of the hospital information system, we can obtain real-time status data of medical resources in each hospital area and the historical scheduling records of the entire disease course that have been completed.

[0011] Each available time slot for each type of resource is abstracted into a resource unit. Each resource unit includes the resource type, the hospital area, the available time window, the carrying capacity, and the current occupancy status.

[0012] Completed patient appointment sequences throughout their entire treatment course are extracted from historical scheduling records. The frequency of collaboration between different resource units as adjacent treatment steps within the same appointment sequence is calculated using the following formula:

[0013] in Let A be the frequency of coordination between resource unit A and resource unit B. This represents the total number of times 'a' appears immediately after 'b' in the historical sequence. This represents the total number of all resource units that appear immediately following 'a'.

[0014] when When the threshold is exceeded and the treatment guidelines are followed, a resource dependency edge is established from a to b.

[0015] A resource dependency graph is constructed using all resource units as nodes and the resource dependency edges as directed edges. Each dependency edge is labeled with its dependency type and normalized dependency strength.

[0016] Step 2: Obtain patient treatment data and parse it to generate a structured resource requirement list. Map each treatment requirement in the list to an appointment node. Combine the treatment time sequence constraints with the resource dependency graph to establish directed edges between appointment nodes with dependencies. After loop verification, construct a directed acyclic appointment chain.

[0017] Step 2 specifically includes:

[0018] Acquire patient medical data, which includes medical orders, diagnostic codes, and historical medical records;

[0019] Natural language processing is performed on the medical orders to extract treatment items, resource types, timing requirements and priorities, and a structured resource requirement list consisting of treatment needs is generated.

[0020] Each medical need is converted into an appointment node, which includes the type of resource needed, the expected time window, and the patient's attribute needs.

[0021] Extract the temporal constraints between medical treatment items from medical orders and historical medical records. For each pair of appointment nodes with a sequential relationship, query the resource dependency graph: if there are corresponding dependency edges in the graph and the directions are consistent, directly establish directed edges between appointment nodes; if there are no corresponding dependency edges, supplement them and mark them as manual rule dependencies.

[0022] A depth-first search is used to traverse the reservation node network to detect the existence of directed cycles. If a cycle exists, the lowest priority dependency edge in the cycle is deleted according to priority. This process is repeated until all directed cycles are eliminated, resulting in a directed acyclic reservation chain.

[0023] Step 3: For appointment nodes involving bed resources in the appointment chain, predict the discharge time and confidence level based on patient characteristics and clinical pathway constraints, and include the bed resources into the appointmentable resource pool according to the confidence level; and directly include the resource units with sufficient available capacity in outpatient, medical technology, and surgical resources into the appointmentable resource pool.

[0024] Step 3 specifically includes:

[0025] For appointment nodes involving bed resources, the patient's disease code, age, number of comorbidities, number of surgical procedures and number of medication regimens are extracted as static features, and the vital signs sequence of historical patients with the same disease during hospitalization is extracted as temporal features. The static features and temporal features are concatenated to form a discharge prediction feature vector.

[0026] The discharge prediction feature vector is input into a temporal neural network model embedded in the clinical pathway constraint layer. The constraint layer queries the standard hospitalization day range in the preset clinical pathway knowledge base according to the disease code, applies boundary constraints to the predicted hospitalization days output by the model, and outputs the constraint prediction value.

[0027] The predicted discharge time is calculated based on the admission date and the constraint prediction value; the baseline confidence level is calculated by the ratio of the prediction error of the model on the test set to the average length of hospital stay, and then the final confidence level is obtained by correcting the patient's age deviation and the number of comorbidities.

[0028] The final confidence level is compared with the classification threshold: bed resources above the high threshold are included in the bookable resource pool starting from the predicted discharge time; bed resources between the high and low thresholds are included and marked as conditionally bookable starting from the predicted discharge time and the date of admission; those below the low threshold are not included.

[0029] For appointment nodes of outpatient, medical technology, and surgical resources, check the current occupancy status of each resource unit within the demand time window, and directly include resource units with sufficient available capacity into the appointment-available resource pool.

[0030] Step 4: Following the chronological order of the reservation chain, select candidate resource units from the available resource pool for each reservation node to generate cross-campus candidate allocation schemes, and calculate a comprehensive score including attribute matching degree, load balancing degree, and cross-campus movement cost; dynamically adjust the score weights according to the actual time sequence deviation of the executed nodes, and select the highest-scoring scheme as the scheduling result for that node.

[0031] Step 4 specifically includes:

[0032] Following the chronological order of the appointment chain, each appointment node to be scheduled is selected one by one, and candidate resource units that meet the basic conditions are screened from the available appointment resource pool according to the type of resource demand, expected time window, and patient attribute requirements. Multiple sets of candidate resource allocation schemes are then generated.

[0033] Calculate the overall score for each group of candidate solutions using the following formula:

[0034] in The resource attribute matching score is calculated by weighting the degree of specialty matching, nursing matching, and equipment matching between the diagnosed disease and the candidate resources. The resource load balance score for each campus is obtained by normalizing the variance of the load rate of similar resources in each campus. For cross-hospital mobility cost scoring, the highest value is taken when the candidate option is in the same hospital as the previous node, and the calculation is based on deductions for cross-hospital mobility factors such as travel time, travel cost and patient mobility ability when the option is in a different hospital. , , Let be the initial weights for the corresponding rating items, and satisfy . ;

[0035] Obtain the deviation between the actual completion time and the expected time of the executed appointment node: if the deviation exceeds a preset threshold, reduce the attribute matching weight. Increase the weight of inter-campus mobility costs If the deviation does not exceed the threshold, the initial weights remain unchanged.

[0036] The overall score is recalculated based on the adjusted weights, and the highest-scoring scheme is selected as the scheduling result for the reservation node.

[0037] Step 5: Perform full-link conflict detection on the scheduling results of all nodes. When a conflict is detected, backtrack along the directed edges of the reservation chain to determine the conflict propagation mode and propagate backward to determine the range of affected successor nodes. Generate candidate resolution schemes for nodes within the affected range. When multiple schemes compete for scarce resources, perform game optimization based on the urgency of diagnosis and treatment. If there are multiple equivalent optimization results, use the arrival time of the reservation request as the bottom rule and select the optimal resolution scheme to correct the reservation chain.

[0038] Step 5 specifically includes:

[0039] The scheduling results of all appointment nodes are obtained, and conflict detection is performed, including: detecting whether the total number of appointment requests for the same resource unit within the same time window exceeds the carrying capacity; detecting whether the time interval between adjacent appointment nodes with sequential dependencies and belonging to different hospital areas satisfies the sum of inter-hospital traffic time and target hospital area waiting buffer time; and detecting whether the completion time of the predecessor node is later than the appointment start time of the successor node. The inter-hospital traffic time is dynamically calculated based on the inter-hospital distance in the preset hospital area connectivity graph, the patient's selected mode of transportation, and the traffic congestion coefficient of the current time period. The target hospital area waiting buffer time is determined statistically based on the average waiting time of the same type of medical resources in the hospital area during the same historical period.

[0040] Upon detecting a conflict, starting from the conflict node, the system traces back along the directed edges of the reservation chain to identify the predecessor dependent nodes to trace the root cause of the conflict. The conflict propagation mode is determined based on the dependency type of each node on the backtracking path. The system then propagates backward along the directed edges, including all subsequent nodes affected by the conflict within the scope of the conflict's influence, until the reservation chain terminates. Predecessor nodes that have already been executed are not included in the scope of the conflict's influence.

[0041] For all appointment nodes within the scope of the conflict, three candidate resolution solutions are generated: adjusting appointment time, changing alternative resources, and switching the hospital area.

[0042] The urgency of treatment for each appointment node is extracted. The urgency is calculated by weighting the severity of the illness indicated by the medical order, the timeliness requirements of the treatment items, and the patient's tolerance for waiting. When multiple appointment chains compete for the same scarce cross-hospital shared resource, a game payoff matrix is ​​constructed. The value of the element in the i-th row and j-th column of the matrix is ​​the degree to which the j-th candidate solution satisfies the i-th appointment node in terms of time matching and resource matching, multiplied by the urgency of treatment for the i-th appointment node. Using the urgency of treatment for each node as a priority weight, an iterative elimination of strictly inferior strategies is used to solve the Nash equilibrium. If there are multiple equivalent equilibrium solutions or if it cannot converge to a pure strategy Nash equilibrium, the arrival time of the appointment requests of each competing appointment node is used as a safety net rule, and the solution corresponding to the first arrival is selected to correct each appointment chain.

[0043] Step 6: Decompose the revised reservation chain into resource reservation instructions for each hospital area and distribute them for execution; monitor resource status changes in real time, calculate the degree of impact of resource anomalies on each reservation node, and trigger rescheduling only for nodes exceeding the threshold.

[0044] Step 6 specifically includes:

[0045] The revised directed acyclic reservation chain is broken down into independent resource reservation instructions according to the respective hospital area, and distributed to the resource management systems of the corresponding hospital area for reservation execution through the data interface.

[0046] The status of resource units in each hospital area is monitored periodically in real time. The monitored events include equipment failures that cause resources to be unavailable, medical staff stopping their services and releasing appointment slots, emergency admissions that cause changes in bed occupancy, and public emergencies that lead to resource requisition.

[0047] When an abnormal resource status event is detected, the directly affected reservation nodes are located and mapped to resource unit nodes in the resource dependency graph. Based on the resource dependency graph, the degree of transmission influence between the abnormal resource unit node and the corresponding resource unit nodes of each reservation node in the reservation chain is calculated. The formula for calculating the degree of transmission influence is as follows:

[0048]

[0049] in To transmit the degree of influence, This represents the dependency strength between two resource unit nodes in the resource dependency graph. This represents the shortest dependency path length between two resource unit nodes in the resource dependency graph.

[0050] The degree of transmission impact of each node is compared with the preset threshold. Only the reservation nodes whose degree of transmission impact exceeds the threshold are marked as pending rescheduling. Steps 4 and 5 are re-executed for the reservation nodes marked as pending rescheduling.

[0051] To further illustrate the versatility and flexibility of this scheduling method, the following example illustrates a real-world scenario where a patient may no longer need subsequent steps during their treatment due to the doctor's judgment or the patient's wishes. In actual treatment, not all patients are required to complete the entire "outpatient visit, examination, surgery, hospitalization" process. For example, if a patient undergoes pre-operative examinations after an outpatient visit, and the results indicate that surgery is not necessary, no surgery appointment will be made. If the patient decides to postpone surgery for personal reasons, the doctor or patient can proactively cancel subsequent surgery and bed reservations through the system. This method can adaptively handle such mid-process termination scenarios, rather than mandating the completion of all collaborative scheduling.

[0052] For example, if the test results are normal, the doctor or patient may voluntarily cancel subsequent surgeries and hospital stays.

[0053] Patient Qian sought medical attention for dull abdominal pain. The outpatient doctor initially diagnosed him with chronic cholecystitis. To determine whether surgery was necessary, the doctor's order stated: "General Surgery Outpatient Clinic. Preoperative examinations include complete blood count, coagulation function tests, and an abdominal CT scan. The decision on whether to perform laparoscopic cholecystectomy will be based on the results. If surgery is required, the patient will be admitted to the hospital the day before the surgery, and the surgery will be performed on Friday morning." This order contains a complete potential chain of "outpatient visit, examination, surgery, and hospitalization."

[0054] In step 2, the system parses the medical orders to generate a structured resource requirement list, containing five treatment requirements: general surgery outpatient visit, complete blood count, coagulation function test, abdominal CT scan, laparoscopic cholecystectomy (conditional trigger), and bed admission (conditional trigger). The system converts each requirement into appointment nodes and extracts the temporal constraints from the medical orders: three examinations after the outpatient visit, follow-up visit after the examinations, and a decision on whether to perform surgery and hospitalization based on the results and the patient's wishes after the follow-up visit. For the surgery and bed nodes, the system marks them as conditional trigger nodes. In step 2.4, a directed edge is established between the surgery node and the outpatient follow-up node and marked as conditional dependency, and a directed edge is established between the bed node and the surgery node and marked as sequential dependency. In step 2.5, after loop verification, a directed acyclic appointment chain containing five nodes is generated, where the surgery node and the bed node are in a conditional trigger state, awaiting confirmation from the doctor or patient.

[0055] In step 3, although the surgical node and bed node already exist in the appointment chain, they are in a condition-triggered state, so the system does not perform bed discharge prediction or surgical resource pool construction for them. The system only performs resource pool construction for outpatient and three examination nodes to complete the appointment scheduling for outpatient and examinations.

[0056] In steps 4 and 5, the system sequentially allocates resources, performs comprehensive scoring, and detects conflicts for the outpatient node and the three examination nodes, generates scheduling results, and corrects the appointment chain.

[0057] After the patient completed their scheduled outpatient visit and three examinations, the doctor reviewed the results during the follow-up outpatient visit and determined it to be mild inflammation, requiring no surgery. The doctor informed the patient of the diagnosis, and the patient indicated they were not considering surgery at this time. The doctor proactively changed the status of the surgery node from "conditionally triggered" to "cancelled" through the system interface. Upon receiving the cancellation command from the doctor, the system propagated it along the directed edges of the appointment chain: the bed node, which depends on the surgery node, automatically changed to the "cancelled" status because its predecessor node had been cancelled. The scheduling results of the outpatient visit and three examinations nodes in the appointment chain remained unchanged, and completed appointments were unaffected.

[0058] If the examination results indicate that surgery is required and the patient agrees to the surgery, the doctor changes the status of the surgical node from "conditionally triggered" to "confirmed" through the system interface. After receiving the confirmation command, the system performs step 3 of bed discharge prediction and surgical resource pool construction for the surgical node, and simultaneously performs bed prediction for the bed node. Then, it sequentially performs step 4 of comprehensive scoring scheduling and step 5 of full-link conflict detection to complete the coordinated scheduling of surgery and beds.

[0059] As can be seen from the above scenarios, when faced with situations where subsequent steps may be canceled due to the doctor's judgment or the patient's wishes during the diagnosis and treatment process, the activation or cancellation of subsequent nodes is initiated by the doctor or patient through the system. The system automatically handles the state changes of related nodes according to the operation instructions along the dependency relationship of the appointment chain, rather than forcing all patients to complete the collaborative scheduling of all resource types.

[0060] A multi-hospital-area intelligent medical resource scheduling system, the system comprising:

[0061] The resource modeling module collects medical resource data and historical scheduling records from multiple hospital areas, abstracts each medical resource into a resource unit, mines the diagnostic and treatment collaboration dependencies between resource units from the historical scheduling records, and constructs a resource dependency graph with dependency strength.

[0062] The appointment chain construction module acquires patient diagnosis and treatment data and parses it to generate a structured resource demand list. It maps each diagnosis and treatment demand in the list to an appointment node. Combining the diagnosis and treatment time sequence constraints with the resource dependency relationship graph, it establishes directed edges between appointment nodes with dependencies. After loop verification, it constructs a directed acyclic appointment chain.

[0063] The resource pool management module predicts discharge time and confidence level for appointment nodes involving bed resources in the appointment chain based on patient characteristics and clinical pathway constraints, and includes bed resources into the appointmentable resource pool according to confidence level; it also directly includes resource units with sufficient available capacity from outpatient, medical technology, and surgical resources into the appointmentable resource pool.

[0064] The scheduling decision module, following the chronological order of the reservation chain, selects candidate resource units from the available resource pool for each reservation node to generate cross-campus candidate allocation schemes, and calculates a comprehensive score including attribute matching degree, load balancing degree, and cross-campus movement cost; dynamically adjusts the score weights according to the actual time sequence deviation of the executed nodes, and selects the highest-scoring scheme as the scheduling result for that node;

[0065] The conflict resolution module performs full-link conflict detection on the scheduling results of all nodes. When a conflict is detected, it backtracks along the directed edges of the reservation chain to determine the conflict propagation mode and propagates backward to determine the range of affected successor nodes. For nodes within the affected range, it generates candidate resolution schemes. When multiple schemes compete for scarce resources, it performs game theory optimization based on the urgency of diagnosis and treatment. If there are multiple equivalent optimization results, it uses the arrival time of the reservation request as the bottom rule and selects the optimal resolution scheme to correct the reservation chain.

[0066] The dynamic scheduling module breaks down the corrected reservation chain into resource reservation instructions for each hospital area and distributes them for execution; it monitors changes in resource status in real time, calculates the impact of resource anomalies on each reservation node, and only triggers rescheduling for nodes that exceed the threshold.

[0067] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0068] 1. In this invention, by automatically mining the diagnostic and treatment collaboration dependencies between resource units from historical scheduling records and constructing a resource dependency graph, the patient's full-course diagnostic and treatment needs are mapped into a directed acyclic reservation chain. This achieves unified modeling and full-course collaborative scheduling of the dependencies of four types of resources: outpatient services, medical technology, beds, and surgery. This solves the problem of fragmented diagnostic and treatment chains throughout the entire course caused by the independent management of each resource type in existing systems. By using a temporal neural network model embedded in the clinical pathway constraint layer to predict the bed discharge time and calculate the confidence level based on the prediction error and individual patient characteristics, bed resources are included in the reservationable resource pool according to the confidence level, so that the timing of bed resource release takes into account both prediction accuracy and individual patient differences.

[0069] 2. In this invention, a comprehensive score is calculated, including attribute matching degree, load balancing degree, and cross-hospital movement cost. The score weight is dynamically adjusted according to the actual time-series deviation of the executed nodes, enabling scheduling decisions to adapt to delays or advancements in treatment progress, thus achieving time-series deviation-driven feedback scheduling optimization. By backtracking along the directed edges of the appointment chain to determine the conflict propagation mode and propagating backward to determine the range of affected successor nodes, the resolution of single-point conflicts is extended to collaborative resolution within the entire link's influence range. Furthermore, when multiple resolution schemes face competition for scarce resources, a game equilibrium is solved based on the urgency of treatment, avoiding the generation of new resource competition conflicts during the conflict resolution process.

[0070] 3. In this invention, by calculating the degree of transmission impact of abnormal resource events on each reservation node, local rescheduling is triggered only for nodes exceeding the threshold, rather than rescheduling all nodes on the reservation chain, thereby reducing the disturbance range of rescheduling and achieving accurate abnormal response. Attached Figure Description

[0071] The present invention will be described by way of example and with reference to the accompanying drawings, wherein:

[0072] Figure 1 This is a schematic diagram of the overall process structure of Embodiment 1 of the present invention;

[0073] Figure 2 This is a schematic diagram of the process structure of step 1 in Embodiment 1 of the present invention;

[0074] Figure 3 This is a schematic diagram of the process structure of step 2 in Embodiment 1 of the present invention. Detailed Implementation

[0075] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0076] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0077] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.

[0078] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0079] In this invention, unless otherwise explicitly specified and limited, "above" or "below" the second feature can include direct contact between the first and second features, or contact between the first and second features through another feature between them. Furthermore, "above," "over," and "on top" of the second feature includes the first feature directly above or diagonally above the second feature, or simply indicates that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature includes the first feature directly below or diagonally below the second feature, or simply indicates that the first feature is at a lower horizontal level than the second feature.

[0080] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.

[0081] Example 1

[0082] like Figures 1-3 As shown in the figure, this embodiment of the invention discloses an intelligent scheduling method for medical resources in multiple hospital areas, including the following steps:

[0083] Step 1: Collect medical resource data and historical scheduling records from multiple hospital campuses, abstract each medical resource into a resource unit, mine the diagnostic and treatment collaboration dependencies between resource units from the historical scheduling records, and construct a resource dependency graph with dependency strength.

[0084] like Figure 2 As shown, step 1 specifically includes:

[0085] Step 1.1: Obtain real-time status data of medical resources in each hospital area and completed historical scheduling records for the entire treatment process through the hospital information system data interface. In actual implementation, the resource status snapshots of each hospital area are periodically retrieved through the data interface in a polling manner, with the polling period set to 1 minute. The historical scheduling records contain all appointment records of previous patients during a single treatment process, from the first appointment to the last appointment. Each record includes the patient identifier, resource type, hospital area identifier, appointment time window, and actual execution time.

[0086] Step 1.2: Abstract each available appointment time slot for each type of resource into a resource unit. Each resource unit includes the resource type, the hospital campus, the available appointment time window, the capacity, and the current occupancy status. For example, the appointment slots for the General Surgery Department on Wednesday mornings from 8:00 to 8:30 are considered an outpatient resource unit with a capacity of 1 appointment slot; the examination slots for the CT Room on Thursday mornings from 9:00 to 9:30 are considered a medical technology resource unit with a capacity of 2 examination slots; Bed No. 5 in the Cardiology Department of the East Campus is considered a bed resource unit with a capacity of 1 bed; and Operating Room No. 3 on Friday mornings from 8:00 to 12:00 is considered a surgical resource unit with a capacity of 1 surgery.

[0087] Step 1.3: Extract completed patient appointment sequences from historical scheduling records, and calculate the frequency of collaboration between different resource units as adjacent treatment steps within the same appointment sequence. The formula for calculating the collaboration frequency is:

[0088]

[0089] in, For resource units Pointing to resource unit The frequency of collaboration, with a value ranging from 0 to 1; In the historical sequence Adjacent Total number of occurrences; denominator For resource units The total number of times all resource units appeared immediately after it. for The total number of resource units that immediately follow the current one. The preset threshold is set to 0.3. When the value exceeds 0.3 and conforms to the logic of the diagnostic and treatment guidelines, establish a path from point to Resource dependency edges. For example, in historical data, there were 100 adjacent appointments after the main campus thoracic surgery outpatient resource unit, of which 65 were adjacent appointments to the main campus blood routine examination resource unit. The collaboration frequency is 100%. If the threshold of 0.3 is exceeded and the subsequent examinations after the outpatient visit comply with the diagnostic and treatment guidelines, then a resource dependency edge is established from the thoracic surgery outpatient resource unit of the main campus to the blood routine examination resource unit of the main campus.

[0090] Step 1.4: Construct a resource dependency graph using all resource units as nodes and the resource dependency edges as directed edges. Label each dependency edge with its dependency type and normalized dependency strength. Dependency strength is determined by the frequency of collaboration. This value is directly used as a normalized value; for example, 0.65 mentioned above represents the dependency strength of this dependency edge, indicating the outpatient resource unit. There is a 65% chance that a complete blood count (CBC) test will be required after completion. .

[0091] Step 2: Obtain patient treatment data and parse it to generate a structured resource requirement list. Map each treatment requirement in the list to an appointment node. Combine the treatment time sequence constraints with the resource dependency graph to establish directed edges between appointment nodes with dependencies. After loop verification, construct a directed acyclic appointment chain.

[0092] like Figure 3 As shown, specifically, step 2 includes the following sub-steps:

[0093] Step 2.1: Obtain patient medical data, which includes medical orders, diagnostic codes, and historical medical records. For example, patient Li's diagnostic code is K80.1 (gallbladder stones), and the medical orders are: "Laparoscopic cholecystectomy, preoperative examinations include complete blood count, coagulation function, abdominal CT, and electrocardiogram. Admission is scheduled for the day before surgery, and surgery will be performed on Friday morning."

[0094] Step 2.2: Perform natural language processing on the medical order content. First, segment and tag the medical order text to identify medical entities. A rule-based and dictionary-based entity extraction method is used: A dictionary of treatment items is established (containing entries such as "laparoscopic cholecystectomy," "complete blood count," "coagulation function," "abdominal CT," and "electrocardiogram," along with corresponding resource type mappings; for example, "laparoscopic cholecystectomy" is mapped to the surgical resource type, and "complete blood count" to the medical technology resource type); a dictionary of time constraints is established (containing entries such as "one day before surgery," "Friday morning," "fasting," and "the night before," along with corresponding time parsing rules); and a dictionary of priorities is established (containing entries such as "urgent," "routine," and "outpatient," along with corresponding priority values). A forward maximum matching scan is performed on the medical order text, and matched entities are labeled as treatment items, time constraints, or priority types. Time constraints are parsed into specific time windows (e.g., "Friday morning" is parsed as Friday 8:00-12:00, and "one day before surgery" is parsed as requiring bed occupancy the day before the surgery appointment). All extracted results are aggregated to generate a structured resource requirement list consisting of treatment needs. For pending treatment items with preconditions such as "depending on examination results" or "to be executed after patient consent" in the medical order, the system marks them as condition-triggered appointment nodes when generating appointment nodes, records the identifiers of the precondition nodes they depend on and the triggering conditions, and stores them in the status attribute field of the appointment node. Each treatment requirement includes the required resource type, expected time window, and priority label.

[0095] Step 2.3: Convert each medical request into an appointment node. Each appointment node includes the type of resource required, the expected time window, and the patient's attribute requirements. Patient attribute requirements are determined by the diagnosis code and basic patient information, including the disease code, nursing level requirements, and equipment requirements.

[0096] Step 2.4: Extract the temporal constraints between treatment items from medical orders and historical medical records. For example, extract the following temporal constraints from medical orders: After the outpatient follow-up visit, perform a complete blood count, coagulation function test, abdominal CT scan, and electrocardiogram in sequence; after all four tests are completed, have an outpatient follow-up visit to check the results; confirm surgery after the outpatient follow-up visit; check into a bed the day before surgery. For each pair of appointment nodes with a sequential relationship, assuming that the treatment need corresponding to node A needs to occur before the treatment need corresponding to node B, first map node A to its corresponding resource unit type, and also map node B to its corresponding resource unit type. Then, in the adjacency list of the resource dependency graph, using the resource unit corresponding to node A as the key, query whether its direct successor node list contains the resource unit corresponding to node B. If it does, it means that there is a dependency edge from A to B in the graph and the direction is consistent with the temporal constraint, so directly establish a directed edge from appointment node A to appointment node B. If the dependency edge does not exist in the graph, but the timing constraint in the medical order explicitly requires A to precede B, then a directed edge from appointment node A to appointment node B is added and marked as "artificial rule dependency".

[0097] Step 2.5: Use depth-first search to traverse the reservation node network and detect the existence of directed cycles. Starting from any reservation node, the depth-first search recursively visits successor nodes along directed edges. If a node that has been visited but not yet searched is encountered during the search, a directed cycle is determined to exist. When a directed cycle is detected, based on the priority labels extracted in Step 2.2, identify the lowest priority dependency edge in the cycle and delete it. After deletion, perform the depth-first search again, repeating the above verification until all directed cycles are eliminated, resulting in a directed acyclic reservation chain.

[0098] Step 3: For appointment nodes involving bed resources in the appointment chain, predict the discharge time and confidence level based on patient characteristics and clinical pathway constraints, and include the bed resources into the appointmentable resource pool according to the confidence level; and directly include the resource units with sufficient available capacity in outpatient, medical technology and surgical resources into the appointmentable resource pool.

[0099] Specifically, step 3 includes:

[0100] Step 3.1: For appointment nodes involving bed resources, extract the patient's disease code, age, number of comorbidities, number of surgical procedures, and number of medication regimens as static features. For example, patient Li's disease code is K80.1 (gallstones), age is 55, number of comorbidities is 1 (hypertension), number of surgical procedures is 1 (laparoscopic cholecystectomy), and number of medication regimens is 3. These static features are all scalars and can be directly arranged into a 5-dimensional vector. Simultaneously, extract the vital signs sequence of historical patients with the same disease during their hospitalization as temporal features, including the daily average body temperature sequence and the daily average white blood cell count sequence since admission. These temporal features are variable-length sequences, which are converted into fixed-length feature vectors through statistical aggregation: taking the mean, maximum, and last day's body temperature value (3 values ​​in total) of the body temperature sequence, and taking the mean, minimum, and last day's white blood cell count value (3 values ​​in total) of the white blood cell count sequence, resulting in a 6-dimensional temporal feature vector. By concatenating the 5-dimensional static feature vector and the 6-dimensional temporal feature vector in sequence, an 11-dimensional discharge prediction feature vector is obtained.

[0101] Step 3.2: Input the discharge prediction feature vector into a temporal neural network model embedded in the clinical pathway constraint layer. This model, based on a conventional gated recurrent unit (GRU) network, inserts a clinical pathway constraint layer before the output layer. The constraint layer queries the standard length of hospital stay range in the preset clinical pathway knowledge base based on the disease code. The preset clinical pathway knowledge base is obtained by statistically analyzing the length of hospital stay of historically discharged patients with the same disease: the actual length of hospital stay of historically discharged patients with the same disease is sorted from shortest to longest, and the 50th percentile is taken as the lower limit. Take the 95th percentile as the upper limit. For example, the 50th percentile of historical hospital stay data for "cholelithiasis-laparoscopic cholecystectomy" is 5 days, and the 95th percentile is 9 days; therefore, the standard range for hospital stay is 5 to 9 days. The constraint layer affects the original predicted hospital stay output of the GRU network. Apply boundary constraints, the constraint formula is:

[0102]

[0103] The meaning of this formula is: when the original predicted value Within the standard length of stay Within the time frame, the constrained predicted value It equals the original predicted value; when the original predicted value is lower than the lower limit of 5, it is taken as 5, and when the original predicted value is higher than the upper limit of 9, it is taken as 9.

[0104] Step 3.3: Calculate the predicted discharge time based on the admission date and the constrained prediction value. For example, if patient Li's scheduled admission date is Thursday afternoon and the constrained prediction value is 7 days, then the predicted discharge time is next Thursday afternoon. Obtain the standard deviation of the prediction error for this disease on the test set from the model training phase. and the average length of hospital stay for this disease. The base confidence level is calculated as follows: For example, if the average length of hospital stay for this disease is 7 days, and the standard deviation of the model's prediction error on the test set is 1.4 days, then the baseline confidence level is... Next, the degree to which the patient's age deviates from the average age for this disease (the average age for this disease is 50 years old, and Mr. Li is 55 years old, the degree of deviation is...) Multiply by a preset coefficient of 0.1, multiply the number of comorbidities (1) by a preset coefficient of 0.05, and subtract both from the base confidence level of 0.8 to obtain the final confidence level. .

[0105] Step 3.4: Compare the final confidence level with the tiered thresholds: Beds with a confidence level above the high threshold are included in the available reservation pool in advance; beds with a confidence level between the high and low thresholds are included near the admission date and marked as conditionally available; beds with a confidence level below the low threshold are not included. The high threshold is set at 0.8, and the low threshold is set at 0.5. Mr. Li's final confidence level is 0.74, which falls between 0.5 and 0.8. Therefore, this bed is included in the available reservation pool one day in advance and marked as "conditionally available," starting from the predicted discharge time (next Thursday afternoon).

[0106] Step 3.5: Process the appointment nodes for outpatient, medical technology, and surgical resources one by one according to the chronological order of the appointment chain. For each appointment node, extract its required resource type and expected time window. Iterate through all resource units across all campuses under this resource requirement, performing a two-step check for each unit: First, check the available appointment time window for that resource unit. Does it fully cover the expected time window of the reservation node, i.e., does it meet the requirements? and If the conditions are met, proceed to the second step of the check; if not, exclude the resource unit. The second step checks whether the current occupancy status of the resource unit is "available." If it is "available," calculate the remaining capacity (carrying capacity minus the currently reserved amount). If the remaining capacity is greater than 0, include the resource unit in the reserveable resource pool; if the remaining capacity is equal to 0 or the current occupancy status is not "available," exclude the resource unit.

[0107] Step 4: Following the time sequence of the reservation chain, select candidate resource units from the available resource pool for each reservation node to generate cross-campus candidate allocation schemes, and calculate a comprehensive score including attribute matching degree, load balancing degree and cross-campus movement cost; dynamically adjust the score weights according to the actual time sequence deviation of the executed nodes, and select the highest score scheme as the scheduling result of the node.

[0108] Specifically, step 4 includes the following sub-steps:

[0109] Step 4.1: Following the chronological order of the appointment chain, for each appointment node to be scheduled, candidate resource units that meet the basic conditions are selected from the available resource pool according to the required resource type, expected time window, and patient attribute requirements. Multiple candidate resource allocation schemes are then generated. Each candidate scheme includes a hospital area identifier and a resource unit identifier, indicating that the appointment node is assigned to that resource unit within that hospital area. For pending treatment items with preconditions such as 'depending on examination results' or 'to be executed after patient consent' in the medical order, the system marks them as condition-triggered appointment nodes when generating the appointment node, records the identifiers of the precondition nodes they depend on and the triggering conditions, and stores them in the status attribute field of the appointment node.

[0110] Step 4.2: Calculate the comprehensive score for each group of candidate solutions. The calculation formula is as follows:

[0111]

[0112] in The resource attribute matching score is calculated by weighting the degree of specialty matching, nursing matching, and equipment matching between the diagnosis and treatment type and the candidate resources. The resource load balance score for each campus is obtained by normalizing the variance of the load rate of similar resources in each campus. For cross-hospital mobility cost scoring, the highest value is taken when the candidate option is in the same hospital as the previous node, and the calculation is based on deductions for cross-hospital mobility factors such as travel time, travel cost and patient mobility ability when the option is in a different hospital. , , The initial default values ​​for the corresponding rating items are set to 0.5, 0.2, and 0.3 respectively, and satisfy the following conditions: .

[0113] Resource attribute matching score The calculation formula is:

[0114]

[0115] in The degree of specialization between the diagnosed disease and the department to which the candidate resource belongs is set as 1 if they match, and 0 otherwise. The formula for calculating the match between patient care level and ward nursing capacity is as follows: , The numerical value representing the level of care required by the patient. This represents the highest level of nursing care that can be provided in the ward. This is the maximum value for the nursing care level (e.g., if the nursing care level is divided into 1-3 levels, then...). ); The matching degree between the demand for diagnostic and treatment equipment and the configuration of candidate resource equipment is obtained by dividing the number of intersections between the demand equipment list and the configuration equipment list by the number of equipment in the demand list. , , The preset sub-item weights are set to 0.5, 0.3, and 0.2 respectively.

[0116] Resource load balancing score The calculation formula is:

[0117]

[0118] in The variance of the load rate of the corresponding resource type in each campus within the reservation time window is given by the load rate as "reserved quantity / carrying capacity". This represents the maximum variance of the load rate of this type of resource during the same historical period.

[0119] Inter-campus mobility cost rating The calculation rule is as follows: if the resource campus of the candidate solution is the same as the campus of the previous reservation node, then... Take the highest value of 1; if the hospital areas are different, the calculation formula is as follows:

[0120]

[0121] in This is the normalized value of the estimated travel time. For actual estimated travel time, The maximum acceptable travel time (e.g., 2 hours); This is the normalized value of transportation costs. To estimate transportation costs, The maximum acceptable transportation cost (e.g., 100 yuan); The mobility factor is determined by the patient's age and the number of comorbidities, and the calculation formula is as follows: ,in This represents the portion of the age that is over 50; if the age is under 50, then it is 0. , , The preset coefficients are set to 0.4, 0.3, and 0.3 respectively.

[0122] Step 4.3: Obtain the deviation between the actual completion time and the expected time of the executed appointment node. The preset deviation threshold is set to 15 minutes. If the deviation exceeds the preset threshold, the attribute matching weight is reduced. Increase the weight of inter-campus mobility costs Weight adjustment step size Set to 0.1. After adjustment , , Keeping it constant at 0.2, the three weights are automatically normalized to satisfy... If the deviation does not exceed the threshold, the initial weights remain unchanged.

[0123] Step 4.4: Recalculate the comprehensive score based on the adjusted weights, and select the highest-scoring scheme as the scheduling result for the reservation node.

[0124] Step 5: Perform full-link conflict detection on the scheduling results of all nodes. When a conflict is detected, backtrack along the directed edges of the reservation chain to determine the conflict propagation mode and propagate backward to determine the range of affected successor nodes. Generate candidate resolution schemes for nodes within the affected range. When multiple schemes compete for scarce resources, solve the game equilibrium based on the urgency of diagnosis and treatment, and select the optimal resolution scheme to correct the reservation chain.

[0125] Specifically, step 5 includes the following sub-steps:

[0126] Step 5.1: Obtain the scheduling results of all appointment nodes and perform conflict detection, including: detecting whether the total number of appointment requests for the same resource unit in the same time window exceeds the carrying capacity; detecting whether the time interval between adjacent appointment nodes that have sequential dependencies and belong to different hospital areas meets the sum of the inter-hospital traffic time and the waiting buffer time of the target hospital area; and detecting whether the completion time of the predecessor node is later than the appointment start time of the successor node.

[0127] The formula for calculating the travel time between campuses is as follows: in This represents the distance (in kilometers) between campuses recorded in the campus connectivity map. Select the average speed (km / h) for the patient's chosen mode of transportation, such as 40km / h for driving and 25km / h for public transportation. The traffic congestion coefficient for the current period is 1.5 for the morning peak (7:00-9:00), 1.0 for the off-peak period, and 1.3 for the evening peak (17:00-19:00).

[0128] The formula for calculating the waiting buffer time at the target hospital area is: in This represents the total number of completed appointments for this type of resource at this hospital within the past 30 days. For the first The actual start time of the scheduled medical visit. For the first The start time of the reservation; if the calculated value is less than the preset minimum buffer time (e.g., 5 minutes), then the minimum buffer time is used.

[0129] Step 5.2: Upon detecting a conflict, starting from the conflicting node, trace back along the directed edges of the reservation chain to find the root cause of the conflict, addressing the predecessor dependent nodes. Determine the conflict propagation mode based on the dependency types of each node on the backtracking path: if the backtracking path consists entirely of sequential dependency edges, it is "serial propagation," where the conflict will affect subsequent nodes sequentially along the sequential dependencies; if the backtracking path contains parallel dependency edges, it is "parallel branch propagation," where the conflict's impact is distributed across multiple branches at the parallel branch. Propagate backward along the directed edges, including all affected successor nodes within the scope of the conflict's influence, until the reservation chain terminates; predecessor nodes that have already completed execution are not included in the scope of the conflict's influence.

[0130] Step 5.3: For all reservation nodes within the scope of the conflict's impact, generate candidate resolution schemes one by one. For each reservation node, generate three types of candidate resolution schemes:

[0131] (1) Adjusting the reservation time: Based on the current reservation time window, search for other available time windows for the same resource unit in the available resource pool. The search range is a 7-day interval before and after the current time window, with a step size of the available time granularity of the resource unit (e.g., 30 minutes). If an available time window is found, a candidate plan for adjusting the reservation time is generated, which involves adjusting the reservation time to that available time window.

[0132] (2) Replacement of alternative resource scheme: Search for other resource units under the same campus and resource type in the available resource pool. The available time window of the resource unit must cover the expected time window of the reservation node and the remaining capacity must be greater than 0. If a suitable alternative resource unit is found, a candidate scheme for replacing the alternative resource is generated. The scheme is to replace the resource unit with the alternative resource unit.

[0133] (3) Switching to another hospital: Search for similar resource units in other hospitals within the available appointment resource pool, using the same search criteria as for changing alternative resource units. If available resource units in other hospitals are found, a candidate solution for switching to another hospital is generated. The three types of candidate solutions for each appointment node are combined using a Cartesian product to generate multiple sets of candidate resolution solutions covering all appointment nodes within the affected area. Each set of candidate resolution solutions includes specific adjustment operations for each appointment node within the affected area.

[0134] Step 5.4: Extract the urgency level of each appointment node. The formula for calculating the urgency level is:

[0135]

[0136] in The normalized values ​​for the severity level of the illness indicated on the medical order are: 1.0 for "urgent", 0.5 for "routine", and 0.2 for "casual visit". This is the timeliness requirement coefficient for diagnosis and treatment items (e.g., chemotherapy items have high timeliness requirements and are rated at 1.0, while routine follow-up examinations have low timeliness requirements and are rated at 0.3). The normalized value of the patient's tolerance for waiting for this treatment is calculated using the following formula: , This refers to the number of times a patient cancels an appointment due to excessively long waiting times. This refers to the total number of appointments made by the patient in their history. but Take 0.5; , , The preset weighting coefficients are set to 0.5, 0.3, and 0.2 respectively.

[0137] When multiple reservation chains compete for the same scarce cross-campus resource, a game-theoretic payoff matrix is ​​constructed. The rows of the matrix correspond to the competing reservation chains, and the columns correspond to the candidate solutions. The matrix's [number of columns]... Line 1 The value of a column element is the first... The candidate resolution schemes for the first... The degree to which each reservation node satisfies time matching and resource matching needs is multiplied by the [number]th [node]. The urgency of treatment at each appointment node. Here, time matching degree is the overlap ratio between the time window provided by the plan and the expected time window (1 for complete overlap, 0 for no overlap). The calculation method for resource matching degree is the same as... The same applies. The average of the time matching degree and the resource matching degree is used as the satisfaction level, which is then multiplied by the urgency of treatment to obtain the payoff matrix element values. Using the urgency of treatment for each reservation node as the priority weight, an iterative elimination method to remove strictly inferior strategies is used to solve the Nash equilibrium: For each game participant, the payoff values ​​of each candidate solution are compared. If a solution is inferior to another solution under all other participants' strategy combinations, it is eliminated; this elimination is repeated until the remaining strategy combinations no longer change, and the remaining strategy combinations are the Nash equilibrium solutions. The solution corresponding to the equilibrium is selected to correct the scheduling results of the affected nodes in each reservation chain. As a safety net, if multiple equivalent Nash equilibrium solutions still exist after the above iterative elimination or if convergence to a unique pure strategy Nash equilibrium fails, the order of arrival time of reservation requests from each competing reservation node is used as the final selection rule, and the solution corresponding to the reservation node with the earliest arrival time is given priority.

[0138] Step 6: Decompose the revised reservation chain into resource reservation instructions for each hospital area and distribute them for execution; monitor resource status changes in real time, calculate the degree of impact of resource anomalies on each reservation node, and trigger rescheduling only for nodes exceeding the threshold.

[0139] Specifically, step 6 includes the following sub-steps:

[0140] Step 6.1: The revised directed acyclic reservation chain is broken down into independent resource reservation instructions according to the respective hospital area, and distributed to the resource management systems of the corresponding hospital areas for reservation execution via data interface. Each resource reservation instruction includes a patient identifier, a resource unit identifier, a reservation time window, and reservation status information. After receiving the instruction, the resource management systems of each hospital area perform resource locking and reservation confirmation operations in their respective systems, and feed the execution results back to the scheduling system.

[0141] Step 6.2: Periodically monitor the status of resource units in each hospital area in real time, with a monitoring cycle set at 30 seconds. Monitored events include equipment failure leading to resource unavailability, staff absences resulting in appointment slot release, emergency admissions causing changes in bed occupancy, and public health emergencies leading to resource requisition. When any of these events are detected, the subsequent anomaly response process is triggered.

[0142] For appointment nodes marked as conditionally triggered, the system provides a status change interface to receive confirmation or cancellation operations initiated by doctors or patients. When a cancellation instruction is received, the system uses that node as the starting point and queries all direct or indirect successor nodes layer by layer along the directed edges of the appointment chain. All nodes that have not yet been executed are automatically marked as cancelled, while executed nodes remain unchanged. When a confirmation instruction is received, the system activates that node and re-executes step 4 (resource allocation) and step 5 (conflict resolution) from that node onwards, ensuring that the scheduling results of subsequent chains are consistent with the newly activated node.

[0143] Step 6.3: When an abnormal resource status event is detected, locate the directly affected reservation nodes and map them to resource unit nodes in the resource dependency graph. Based on the resource dependency graph, calculate the degree of transmission influence between the abnormal resource unit node and the corresponding resource unit nodes of each reservation node in the reservation chain. The formula for calculating the degree of transmission influence is:

[0144] in To transmit the degree of influence, This represents the dependency strength between two resource unit nodes in the resource dependency graph. This represents the shortest dependency path length between two resource unit nodes in the resource dependency graph. The shortest dependency path length is calculated using a breadth-first search algorithm: starting from the anomalous resource unit node, expanding outwards layer by layer along the dependency edges, recording the number of edges traversed when first reaching each target resource unit node; this number of edges is the shortest dependency path length. Dependency strength. Take the product of the dependency strengths of each dependency edge on the shortest dependency path. For example, if the path length from the abnormal bed node to the surgical node is 1 and the dependency strength is 0.8, then... ; Path length to outpatient node is 3, dependency strength product ,but .

[0145] Step 6.4: Compare the transmission impact level of each node with a preset threshold of 0.3. Only the appointment nodes whose transmission impact level exceeds the threshold are marked as pending rescheduling. For appointment nodes marked as pending rescheduling, re-execute the resource allocation in Step 4 and the conflict resolution in Step 5. For example, the transmission impact level of the surgical appointment node is 0.8, which exceeds the 0.3 threshold, and it is marked as pending rescheduling; while the transmission impact level of the outpatient appointment node is 0.064, which does not exceed the 0.3 threshold, and does not trigger rescheduling. This precise rescheduling mechanism based on the transmission impact level only makes local adjustments to appointment nodes that are significantly affected by abnormal events, avoiding the problem of excessive disturbance range caused by rescheduling all nodes in the appointment chain in traditional solutions.

[0146] Example 2

[0147] This embodiment proposes a multi-hospital-area medical resource intelligent scheduling system, the system comprising:

[0148] The resource modeling module collects medical resource data and historical scheduling records from multiple hospital areas, abstracts each medical resource into a resource unit, mines the diagnostic and treatment collaboration dependencies between resource units from the historical scheduling records, and constructs a resource dependency graph with dependency strength.

[0149] The appointment chain construction module acquires patient diagnosis and treatment data and parses it to generate a structured resource demand list. It maps each diagnosis and treatment demand in the list to an appointment node. Combining the diagnosis and treatment time sequence constraints with the resource dependency relationship graph, it establishes directed edges between appointment nodes with dependencies. After loop verification, it constructs a directed acyclic appointment chain.

[0150] The resource pool management module predicts discharge time and confidence level for appointment nodes involving bed resources in the appointment chain based on patient characteristics and clinical pathway constraints, and includes bed resources into the appointmentable resource pool according to confidence level; it also directly includes resource units with sufficient available capacity from outpatient, medical technology, and surgical resources into the appointmentable resource pool.

[0151] The scheduling decision module, following the chronological order of the reservation chain, selects candidate resource units from the available resource pool for each reservation node to generate cross-campus candidate allocation schemes, and calculates a comprehensive score including attribute matching degree, load balancing degree, and cross-campus movement cost; dynamically adjusts the score weights according to the actual time sequence deviation of the executed nodes, and selects the highest-scoring scheme as the scheduling result for that node;

[0152] The conflict resolution module performs full-link conflict detection on the scheduling results of all nodes. When a conflict is detected, it backtracks along the directed edges of the reservation chain to determine the conflict propagation mode and propagates backward to determine the range of affected successor nodes. For nodes within the affected range, it generates candidate resolution schemes. When multiple schemes compete for scarce resources, it performs game theory optimization based on the urgency of diagnosis and treatment. If there are multiple equivalent optimization results, it uses the arrival time of the reservation request as the bottom rule and selects the optimal resolution scheme to correct the reservation chain.

[0153] The dynamic scheduling module breaks down the corrected reservation chain into resource reservation instructions for each hospital area and distributes them for execution; it monitors changes in resource status in real time, calculates the impact of resource anomalies on each reservation node, and only triggers rescheduling for nodes that exceed the threshold.

[0154] The circuits, electronic components, and modules involved are all existing technologies, which can be fully implemented by those skilled in the art, and need not be elaborated upon. The scope of protection of this invention does not involve any improvement to the software and methods.

[0155] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0156] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent scheduling of medical resources across multiple hospital campuses, characterized in that, Includes the following steps: Step 1: Collect medical resource data and historical scheduling records from multiple hospital campuses, abstract each medical resource into a resource unit, mine the diagnostic and treatment collaboration dependencies between resource units from the historical scheduling records, and construct a resource dependency relationship map with dependency strength; Step 2: Obtain patient treatment data and parse it to generate a structured resource requirement list. Map each treatment requirement in the list to an appointment node. Combine the treatment time sequence constraints with the resource dependency graph to establish directed edges between appointment nodes with dependencies. After loop verification, construct a directed acyclic appointment chain. Step 3: For appointment nodes involving bed resources in the appointment chain, predict the discharge time and confidence level based on patient characteristics and clinical pathway constraints, and include the bed resources into the appointmentable resource pool according to the confidence level; and directly include the resource units with sufficient available capacity in outpatient, medical technology, and surgical resources into the appointmentable resource pool. Step 4: Following the chronological order of the reservation chain, select candidate resource units from the available resource pool for each reservation node to generate cross-campus candidate allocation schemes, and calculate a comprehensive score including attribute matching degree, load balancing degree, and cross-campus movement cost; dynamically adjust the score weights according to the actual time sequence deviation of the executed nodes, and select the highest-scoring scheme as the scheduling result for that node. Step 5: Perform full-link conflict detection on the scheduling results of all nodes. When a conflict is detected, backtrack along the directed edges of the reservation chain to determine the conflict propagation mode and propagate backward to determine the range of affected successor nodes. Generate candidate resolution schemes for nodes within the affected range. When multiple schemes compete for scarce resources, perform game optimization based on the urgency of diagnosis and treatment. If there are multiple equivalent optimization results, use the arrival time of the reservation request as the bottom rule and select the optimal resolution scheme to correct the reservation chain. Step 6: Break down the revised appointment chain into resource appointment instructions for each hospital area and distribute them for execution; Real-time monitoring of resource status changes, calculation of the impact of resource anomalies on each reserved node, and triggering rescheduling only for nodes exceeding the threshold.

2. The intelligent scheduling method for medical resources across multiple hospital campuses according to claim 1, characterized in that, Step 1 specifically includes: By using the data interface of the hospital information system, we can obtain real-time status data of medical resources in each hospital area and the historical scheduling records of the entire disease course that have been completed. Each available time slot for each type of resource is abstracted into a resource unit. Each resource unit includes the resource type, the hospital area, the available time window, the carrying capacity, and the current occupancy status. Completed patient appointment sequences throughout their entire treatment course are extracted from historical scheduling records. The frequency of collaboration between different resource units as adjacent treatment steps within the same appointment sequence is calculated using the following formula: in Let A be the frequency of coordination between resource unit A and resource unit B. This represents the total number of times 'a' appears immediately after 'b' in the historical sequence. This represents the total number of all resource units that appear immediately following 'a'. when When the threshold is exceeded and the treatment guidelines are followed, a resource dependency edge is established from a to b. A resource dependency graph is constructed using all resource units as nodes and the resource dependency edges as directed edges. Each dependency edge is labeled with its dependency type and normalized dependency strength.

3. The intelligent scheduling method for medical resources across multiple hospital campuses according to claim 1, characterized in that, Step 2 specifically includes: Acquire patient medical data, which includes medical orders, diagnostic codes, and historical medical records; Natural language processing is performed on the medical orders to extract treatment items, resource types, timing requirements and priorities, and a structured resource requirement list consisting of treatment needs is generated. Each medical need is converted into an appointment node, which includes the type of resource needed, the expected time window, and the patient's attribute needs. Extract the temporal constraints between medical treatment items from medical orders and historical medical records. For each pair of appointment nodes with a sequential relationship, query the resource dependency graph: if there are corresponding dependency edges in the graph and the directions are consistent, directly establish directed edges between appointment nodes; if there are no corresponding dependency edges, supplement them and mark them as manual rule dependencies. A depth-first search is used to traverse the reservation node network to detect whether there are directed cycles. If they exist, the lowest priority dependency edge in the cycle is deleted according to priority. The detection is repeated until all directed cycles are eliminated, resulting in a directed acyclic reservation chain. Furthermore, in the appointment node, pending medical treatment needs described in the medical order as depending on the results of prior medical treatment or the patient's wishes to decide whether to execute are marked as condition-triggered appointment nodes, and a condition-triggered identifier and the identifier of the prior node it depends on are attached to the node.

4. The intelligent scheduling method for medical resources across multiple hospital campuses according to claim 1, characterized in that, Step 3 specifically includes: For appointment nodes involving bed resources, the patient's disease code, age, number of comorbidities, number of surgical procedures and number of medication regimens are extracted as static features, and the vital signs sequence of historical patients with the same disease during hospitalization is extracted as temporal features. The static features and temporal features are concatenated to form a discharge prediction feature vector. The discharge prediction feature vector is input into a temporal neural network model embedded in the clinical pathway constraint layer. The constraint layer queries the standard hospitalization day range in the preset clinical pathway knowledge base according to the disease code, applies boundary constraints to the predicted hospitalization days output by the model, and outputs the constraint prediction value. The predicted discharge time is calculated based on the admission date and the constraint prediction value; the baseline confidence level is calculated by the ratio of the prediction error of the model on the test set to the average length of hospital stay, and then the final confidence level is obtained by correcting the patient's age deviation and the number of comorbidities. The final confidence level is compared with the classification threshold: bed resources above the high threshold are included in the bookable resource pool starting from the predicted discharge time; bed resources between the high and low thresholds are included and marked as conditionally bookable starting from the predicted discharge time and the date of admission; those below the low threshold are not included. For appointment nodes of outpatient, medical technology, and surgical resources, check the current occupancy status of each resource unit within the demand time window, and directly include resource units with sufficient available capacity into the appointment-available resource pool.

5. The intelligent scheduling method for medical resources across multiple hospital campuses according to claim 1, characterized in that, Step 4 specifically includes: Following the chronological order of the appointment chain, each appointment node to be scheduled is selected one by one. Based on the type of resource required, the expected time window, and the patient's attribute requirements, candidate resource units that meet the basic conditions are screened from the available resource pool. The patient's physiological calendar constraint information is also obtained. The physiological calendar constraint information includes at least the fasting time and contraindication period required by the treatment project, the drug washout period and the contraindication period of the female menstrual cycle. The appointment time window of the candidate resource unit is compared with the above-mentioned contraindication periods. If there is an overlap, the candidate resource unit is removed. Multiple sets of candidate resource allocation schemes are generated. Calculate the overall score for each group of candidate solutions using the following formula: in The resource attribute matching score is calculated by weighting the degree of specialty matching, nursing matching, and equipment matching between the diagnosed disease and the candidate resources. The resource load balance score for each campus is obtained by normalizing the variance of the load rate of similar resources in each campus. For cross-hospital mobility cost scoring, the highest value is taken when the candidate option is in the same hospital as the previous node, and the calculation is based on deductions for cross-hospital mobility factors such as travel time, travel cost and patient mobility ability when the option is in a different hospital. , , Let be the initial weights for the corresponding rating items, and satisfy . ; Obtain the deviation between the actual completion time and the expected time of the executed appointment node: if the deviation exceeds a preset threshold, reduce the attribute matching weight. Increase the weight of inter-campus mobility costs If the deviation does not exceed the threshold, the initial weights remain unchanged. The overall score is recalculated based on the adjusted weights, and the highest-scoring scheme is selected as the scheduling result for the reservation node.

6. The intelligent scheduling method for medical resources across multiple hospital campuses according to claim 1, characterized in that, Step 5 specifically includes: Obtain the scheduling results of all appointment nodes and perform conflict detection, including: detecting whether the total number of appointment requests for the same resource unit in the same time window exceeds the carrying capacity; detecting whether the time interval between adjacent appointment nodes that have sequential dependencies and belong to different hospital areas meets the sum of the inter-hospital traffic time and the waiting buffer time of the target hospital area; and detecting whether the completion time of the predecessor node is later than the appointment start time of the successor node. Upon detecting a conflict, starting from the conflict node, the system traces back along the directed edges of the reservation chain to identify the predecessor dependent nodes to trace the root cause of the conflict. The conflict propagation mode is determined based on the dependency type of each node on the backtracking path. The system then propagates backward along the directed edges, including all subsequent nodes affected by the conflict within the scope of the conflict's influence, until the reservation chain terminates. Predecessor nodes that have already been executed are not included in the scope of the conflict's influence. For all appointment nodes within the scope of the conflict, three candidate resolution solutions are generated: adjusting appointment time, changing alternative resources, and switching the hospital area. The urgency of treatment for each appointment node is extracted. The urgency is calculated by weighting the severity of the illness indicated by the medical order, the timeliness requirements of the treatment items, and the patient's tolerance for waiting. When multiple appointment chains compete for the same scarce cross-hospital shared resource, a game payoff matrix is ​​constructed. The value of the element in the i-th row and j-th column of the matrix is ​​the degree to which the j-th candidate solution satisfies the i-th appointment node in terms of time matching and resource matching, multiplied by the urgency of treatment for the i-th appointment node. Using the urgency of treatment for each node as a priority weight, an iterative elimination of strictly inferior strategies is used to solve the Nash equilibrium. If there are multiple equivalent equilibrium solutions or if it cannot converge to a pure strategy Nash equilibrium, the arrival time of the appointment requests of each competing appointment node is used as a safety net rule, and the solution corresponding to the first arrival is selected to correct each appointment chain.

7. The intelligent scheduling method for medical resources across multiple hospital campuses according to claim 6, characterized in that, The revised reservation chain further includes: responding to confirmation or cancellation instructions initiated by the external system to the condition-triggered reservation node; when the state change instruction is cancellation, starting from the condition-triggered node, propagating the cancellation state backward along the directed edges of the reservation chain, and automatically marking all unexecuted successor nodes that have a direct or indirect dependency relationship with the node as cancellation; when the state change instruction is confirmation, activating the condition-triggered node, and re-executing the scheduling process of steps 4 to 5 for the node and its unexecuted successor nodes.

8. The intelligent scheduling method for medical resources across multiple hospital campuses according to claim 1, characterized in that, Step 6 specifically includes: The revised directed acyclic reservation chain is broken down into independent resource reservation instructions according to the respective hospital area, and distributed to the resource management systems of the corresponding hospital area for reservation execution through the data interface. The status of resource units in each hospital area is monitored periodically in real time. The monitored events include equipment failures that cause resources to be unavailable, medical staff stopping their services and releasing appointment slots, emergency admissions that cause changes in bed occupancy, and public emergencies that lead to resource requisition. When an abnormal resource status event is detected, the directly affected reservation node is located and mapped to the resource unit node in the resource dependency graph. Based on the resource dependency graph, the degree of transmission influence between the abnormal resource unit node and the corresponding resource unit nodes of each reservation node in the reservation chain is calculated. The degree of transmission impact of each node is compared with the preset threshold. Only the reservation nodes whose degree of transmission impact exceeds the threshold are marked as pending rescheduling. Steps 4 and 5 are re-executed for the reservation nodes marked as pending rescheduling.

9. The intelligent scheduling method for medical resources in multiple hospital areas according to claim 8, characterized in that, The formula for calculating the degree of conduction influence is as follows: in To transmit the degree of influence, This represents the dependency strength between two resource unit nodes in the resource dependency graph. This represents the shortest dependency path length between two resource unit nodes in the resource dependency graph.

10. A multi-campus medical resource intelligent scheduling system, characterized in that, The system for performing the method according to any one of claims 1 to 9, the system comprising: The resource modeling module collects medical resource data and historical scheduling records from multiple hospital areas, abstracts each medical resource into a resource unit, mines the diagnostic and treatment collaboration dependencies between resource units from the historical scheduling records, and constructs a resource dependency graph with dependency strength. The appointment chain construction module acquires patient diagnosis and treatment data and parses it to generate a structured resource demand list. It maps each diagnosis and treatment demand in the list to an appointment node. Combining the diagnosis and treatment time sequence constraints with the resource dependency relationship graph, it establishes directed edges between appointment nodes with dependencies. After loop verification, it constructs a directed acyclic appointment chain. The resource pool management module predicts discharge time and confidence level for appointment nodes involving bed resources in the appointment chain based on patient characteristics and clinical pathway constraints, and includes bed resources into the appointmentable resource pool according to confidence level; it also directly includes resource units with sufficient available capacity from outpatient, medical technology, and surgical resources into the appointmentable resource pool. The scheduling decision module, following the chronological order of the reservation chain, selects candidate resource units from the available resource pool for each reservation node to generate cross-campus candidate allocation schemes, and calculates a comprehensive score including attribute matching degree, load balancing degree, and cross-campus movement cost; dynamically adjusts the score weights according to the actual time sequence deviation of the executed nodes, and selects the highest-scoring scheme as the scheduling result for that node; The conflict resolution module performs full-link conflict detection on the scheduling results of all nodes. When a conflict is detected, it backtracks along the directed edges of the reservation chain to determine the conflict propagation mode and propagates backward to determine the range of affected successor nodes. For nodes within the affected range, it generates candidate resolution schemes. When multiple schemes compete for scarce resources, it performs game theory optimization based on the urgency of diagnosis and treatment. If there are multiple equivalent optimization results, it uses the arrival time of the reservation request as the bottom rule and selects the optimal resolution scheme to correct the reservation chain. The dynamic scheduling module breaks down the corrected reservation chain into resource reservation instructions for each hospital area and distributes them for execution; it monitors changes in resource status in real time, calculates the impact of resource anomalies on each reservation node, and only triggers rescheduling for nodes that exceed the threshold.

Images