Multi-institutional human dynamic deployment decision support system based on hierarchical response mechanism

Abstract

The present application relates to a multi-hospital area human resource dynamic allocation decision support system based on a hierarchical response mechanism, which integrates hospital information systems, electronic medical records and office automation data in real time through a multi-source data fusion module, and builds a distributed data warehouse after standardization cleaning. In the system operation, the data acquisition layer captures abnormal events in the hospital area in real time, and after cleaning and verification, drives the prediction engine to output early warning. When the response threshold is reached, the decision unit automatically matches the optimal scheduling scheme, and synchronizes the task instructions through the mobile terminal. The whole chain tracking is implemented in the execution process, and the resource state is dynamically updated. After the event, multi-index scoring is carried out based on human efficiency, business load and other dimensions, driving the strategy library iteration and model weight adjustment, forming a "monitoring-prediction-decision-execution-evaluation" closed loop. The present application realizes the intelligent cooperation and flexible scheduling of multi-hospital area human resources, and significantly improves the response efficiency in emergency scenarios and the utilization rate of medical resources.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of medical information technology, in particular to a multi-hospital computing architecture based on artificial intelligence and real-time data fusion, which is used to solve the technical problems of response delay and resource mismatch in dynamic collaboration of cross-hospital human resources. BACKGROUND

[0002] At present, large hospitals with multiple campuses generally face the problem of dynamic allocation of multi-campus human resources. The independent operation of each hospital information system (such as HIS, EMR, OA) forms a data island, and key information such as patient flow fluctuation, on-duty status of medical staff, and equipment resources cannot be real-time interchanged. The independent operation of each hospital information system (such as HIS, EMR, OA) forms a data island, and the underlying data format and communication protocol are heterogeneous, which leads to the inability of real-time interchanging of key information, which is a typical technical obstacle. Traditional manual scheduling relies on experience-based decision-making, which cannot achieve millisecond-level data synchronization and computing response, and it is difficult to quickly respond to sudden scenes such as emergency flow surge, resulting in technical consequences of resource mismatch and response delay.

[0003] The prior art lacks precise human demand prediction capability, and its prediction model mostly uses static statistical model, which has low calculation efficiency and cannot effectively capture nonlinear and high-dimensional time sequence features; there are also serious shortcomings in the decision-making link: the emergency plan library is fixed, and no decision-making algorithm is established in conjunction with real-time traffic data and equipment status data; most systems only complete resource scheduling instruction issuing, and due to the system architecture not supporting full-process data tracking, there is a lack of closed-loop feedback algorithm for automatically generating quantitative scores and driving strategy iteration, resulting in historical execution data not being systematized for decision parameter self-correction. SUMMARY

[0004] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0005] According to the first aspect of the present application, a multi-hospital human dynamic allocation decision support system based on a hierarchical response mechanism is requested to be protected, comprising:

[0006] A multi-source data fusion module acquires patient diagnosis and treatment data of a hospital information system (HIS) and an electronic medical record system (EMR), and on-duty status data of medical staff of an office automation system (OA) in real time through a standardized interface, and performs data cleaning and format unification through ETL technology to construct a distributed data warehouse;

[0007] An intelligent prediction engine predicts the patient flow of each hospital department in a first preset time period in the future using an LSTM neural network, calculates the absence probability combining a Bayesian model, and outputs a quantitative value of dynamic human gap;

[0008] A four-level response decision unit performs resource scheduling under a preset emergency scenario;

[0009] A dynamic strategy library stores the deployment path, scheduling adjustment rules and priority strategies corresponding to different response levels, and dynamically optimizes resource scheduling parameters according to historical execution effects;

[0010] A closed-loop evaluation module constructs a four-dimensional index library of human efficiency, business load, service quality and human structure, generates a hospital area comprehensive evaluation report based on a percentage scoring system, and automatically corrects the prediction model weight according to the scoring result.

[0011] Further, the multi-source data fusion module comprises:

[0012] An API gateway cluster deployed in the hospital intranet synchronizes full data with a second preset time length as a cycle, and captures incremental data at intervals of a third preset time length;

[0013] A data cleaning unit uses regular expression to check the timestamp format, and removes outliers based on the box plot algorithm;

[0014] When the field missing rate is > 5%, trigger real-time recording alarm to the office automation OA system;

[0015] During daily full synchronization, automatically check the patient ID consistency of the HIS system and the EMR system, and use EMR as the standard when there is a conflict;

[0016] The box plot algorithm identifies abnormal values of the number of registrations, triggering an artificial review process;

[0017] Using the recording mechanism, if the field missing alarm lasts for 30 minutes without processing, automatically fill in the average of the last three days.

[0018] Further, the running logic of the intelligent prediction engine comprises:

[0019] Using LSTM neural network to set 128 neuron hidden layer, input the hourly outpatient flow time series data in the first cycle;

[0020] Using Bayesian absence model, the model analyzes the correlation between historical absence data, meteorological data and statutory holiday dates to automatically generate dynamic seasonal correction factors;

[0021] Output the human resource demand prediction report marked with department-level gap distribution heat map;

[0022] Automatically retrain the LSTM model at a preset time period every week, and add new data to the sliding time window;

[0023] The system preset month parameter triggers the automatic loading of the seasonal correction factor, and the respiratory and critical care medicine department sick leave probability benchmark value is increased by 15%;

[0024] Generate a heat map report, red area indicates the gap > 5 people or the same post gap is greater than or equal to 15% of the department, yellow indicates the gap 2-4 people, or the same post gap is greater than or equal to 10% less than 15% of the department.

[0025] Further, the fourth level response decision unit comprises:

[0026] The first level response module: in the major public health event, the cross-hospital area integrated team dispatching and reserve manpower calling are started;

[0027] The second level response module: the inter-hospital area specialist cooperation and multi-department linkage are triggered for single-hospital area mass injury event;

[0028] The third level response module: the elastic scheduling and cross-department support are executed for the continuously overloaded department;

[0029] The fourth level response module: the department level independent fine-tuning scheduling and the rotation personnel calling are supported;

[0030] The trigger condition of the fourth level response decision unit is:

[0031] The first level response: the whole hospital manpower gap > 30% or the public health event red alert is issued, after receiving the red instruction of the early warning platform, the reserve manpower library is started within 10 seconds;

[0032] The second level response: the single-hospital area emergency department 30-minute reception volume is over the threshold value 150%, in the mass injury event, the trauma specialist team within < 10 kilometers from the target hospital area is automatically matched;

[0033] The third level response: the department continuously overloaded > 48 hours and the average hospitalization day growth < 80%, after the overloaded department marker is continuously marked for 48 hours, the support request is sent to the related department;

[0034] The fourth level response: the department scheduling conflict or the temporary leave vacancy is less than or equal to 2 people, when the early shift vacancy is detected in the scheduling conflict detection, the rotation personnel list of the department is automatically searched.

[0035] Further, the optimization mechanism of the dynamic strategy library comprises:

[0036] The on-site delay rate and the skill matching deviation degree of each deployment are recorded;

[0037] When the on-site delay rate > 20%, the priority weight of the deployment path is automatically promoted;

[0038] The effectiveness matrix of the strategy is generated every month to drive the iteration of the rules;

[0039] The actual on-site time and the skill usage rate are stored after each dispatch;

[0040] When the path delay rate is greater than 20%, the strategy is adjusted, the priority is automatically degraded, and the alternative path is tested;

[0041] Monthly iterations are performed to generate matrix reports to compare the efficiency of different hospital combinations and obtain the preferred hospital combination.

[0042] Further, the index library of the closed-loop evaluation module comprises:

[0043] Human efficiency dimension: number of diagnoses per capita, absence impact coefficient;

[0044] Business load dimension: emergency rescue success rate, gap rate of rural training;

[0045] Service quality dimension: medical error rate, waiting time compliance rate;

[0046] Human structure dimension: absence replacement rate, personnel compliance;

[0047] Each index is calculated by dynamic weight to obtain a percentage comprehensive score;

[0048] Absence impact coefficient = (actual number of on-site personnel / standard number of personnel) x 100%;

[0049] Waiting time compliance rate = (<30 minutes patient number / total patient number) x 100%;

[0050] The scoring rule is that when the single index achievement is less than 60%, the highest score of this dimension is 79 points to pass.

[0051] Further, the application logic of the scoring result is that if the score of two consecutive quarters is less than 70 points, a special optimization plan is triggered, and personnel adjustment suggestions are automatically pushed to the human resource system. In the teaching hospital scenario, the "training coverage rate" weight is increased to 20%;

[0052] Continuous low-score hospital triggers personnel optimization, and automatically compares the number of diagnoses per capita with similar hospitals;

[0053] In the teaching hospital mode, the weight is adjusted, and the training coverage rate weight is increased from 10% to 20%;

[0054] After the personnel suggestions are pushed to the HR system, the personnel staffing quantity suggestion value is automatically associated and the personnel gap quantity is promoted, and the system is linked with the recruitment system.

[0055] Further, the system further comprises:

[0056] An integrated visual interaction layer is provided, which renders a hospital human saturation heat map through an ECharts engine, uses red, yellow and green three colors for early warning, supports drag adjustment scheme and real-time pop-up display of index changes, and pushes a task progress tracking view when the mobile terminal APP responds to three levels or more.

[0057] Perform a click red overload department operation on the heat map, pop up a list of schedulable personnel and an estimated arrival time;

[0058] Drag the physician to the target department to adjust the plan, and display the real-time waiting time prediction value change;

[0059] Mobile monitoring is adopted, and the APP pushes a task progress bar when the secondary response is triggered.

[0060] Further, the system further comprises:

[0061] A distributed computing architecture is adopted, the intelligent prediction engine is deployed on a GPU server cluster, the response decision unit is run on a memory database, and the data warehouse adopts columnar storage to optimize the concurrent query of thousands of people; each hospital area OA system is connected through a micro-service API gateway;

[0062] The GPU cluster performs parallel computation on the flow prediction of each hospital area and performs prediction acceleration;

[0063] The memory database preloads the strategy library for real-time decision-making, and the response trigger is generated to the scheme generation within <10 seconds;

[0064] The OA system receives scheduling instructions through RESTful API for micro-service integration, and returns the personnel confirmation state.

[0065] Further, the system further comprises:

[0066] A sandbox simulation module is set, the administrator modifies the commuting time threshold and the response level parameter, the system automatically simulates the load balancing change curve within 72 hours, and outputs the index comparison report of the optimized scheme and the baseline scheme;

[0067] The administrator modifies the commuting time parameter, and the system simulates the cross-hospital area scheduling simulation running data within the first preset time length;

[0068] The report output comparison displays the waiting time reduction data and the manpower cost data of the new scheme.

[0069] The application relates to a multi-hospital area human resource dynamic allocation decision support system based on a hierarchical response mechanism, which integrates a hospital information system, an electronic medical record and office automation data in real time through a multi-source data fusion module, and constructs a distributed data warehouse after standardization cleaning. In the system operation, the data acquisition layer captures abnormal events in the hospital area in real time, and drives the prediction engine to output early warning after cleaning and verification. When the response threshold is reached, the decision unit automatically matches the optimal scheduling scheme, and synchronizes the task instructions through the mobile terminal. The execution process implements full-chain tracking, and dynamically updates the resource state. After the event, multi-index scoring is carried out based on human efficiency, business load and other dimensions, driving the strategy library iteration and model weight adjustment, forming a "monitoring-prediction-decision-execution-evaluation" closed loop. The application realizes intelligent cooperation and flexible scheduling of multi-hospital area human resources, and significantly improves the response efficiency in emergency scenarios and the utilization rate of medical resources. BRIEF DESCRIPTION OF DRAWINGS

[0070] Figure 1 A structure module diagram of a multi-hospital area human resource dynamic allocation decision support system based on a hierarchical response mechanism is requested to be protected by the embodiments of the application. DETAILED DESCRIPTION

[0071] The technical solutions in the embodiments of the application will be clearly and completely described below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only part of the embodiments of the application, rather than all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the application.

[0072] The terms "first", "second", "third" in the application are only for description purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features limited by "first", "second", "third" can explicitly or implicitly include at least one of the features. In the description of the application, the meaning of "multiple" is at least two, for example, two, three, etc., unless otherwise explicitly and specifically limited. All directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the application are only used to explain the relative position relationship, movement condition, etc. between the components in a certain posture (as shown in the drawings), and if the certain posture changes, the directional indications also change accordingly. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device.

[0073] Reference to“an embodiment” herein means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the application. The appearances of the phrase“in an embodiment” in various places in the specification are not necessarily all referring to the same embodiment, nor are they necessarily all directed to the same embodiment, nor are they necessarily all mutually exclusive or alternative embodiments. It is expressly understood that any of the embodiments described herein can be incorporated into other embodiments.

[0074] According to the first embodiment of the application, the application claims to protect a multi-hospital district human dynamic deployment decision support system based on a hierarchical response mechanism, referring to Figure 1 , comprising:

[0075] A multi-source data fusion module collects patient diagnosis and treatment data of a hospital information system (HIS) and an electronic medical record system (EMR), and on-duty state data of medical staff of an office automation (OA) system in real time through a standardized interface, and performs data cleaning and format unification through an ETL technology to construct a distributed data warehouse;

[0076] An intelligent prediction engine predicts patient flow of each hospital department in a first preset time period in the future using an LSTM neural network, calculates a sick leave probability in combination with a Bayesian model, and outputs a quantitative value of a dynamic human gap;

[0077] A four-level response decision unit performs resource scheduling in a preset emergency scenario;

[0078] A dynamic strategy library stores deployment paths, scheduling adjustment rules, and priority strategies corresponding to different response levels, and dynamically optimizes resource scheduling parameters according to historical execution effects;

[0079] A closed-loop evaluation module constructs a four-dimensional index library of human efficiency, business load, service quality, and human structure, generates a hospital district comprehensive evaluation report based on a percentage grading system, and automatically corrects the weight of the prediction model according to the grading results.

[0080] In this embodiment, the multi-source data fusion module solves the technical problems of“data silos”and“heterogeneous data”through a standardized interface, an ETL technology, and the construction of a distributed data warehouse;

[0081] The use of specific technical means such as an API gateway cluster, regular expression verification, and box plot algorithm brings technical effects such as“improved data quality”and“real-time exception alarm”.

[0082] Data collection 7:30 The API gateway captures a sudden surge in the instantaneous flow of the emergency department of hospital A from the HIS system, with 45 registered patients in 10 minutes, which is 160% higher than the baseline value;

[0083] The OA system reported that three physicians in the Department of Respiratory and Critical Care Medicine at Campus B were on sick leave, and the sick leave rate rose to 35% during the winter flu season.

[0084] The EMR system shows a backlog of elective orthopedic surgeries in Campus C, and the average length of stay in that ward is expected to increase by 0.6 days this week.

[0085] The cleaning process automatically corrects timestamp errors in the HIS system, including 5 records where the "2025-13-15" record is converted to a null value;

[0086] The box plot algorithm identified and removed outlier registration values, including one record of "1200 people registered in a single hour";

[0087] The Department of Respiratory and Critical Care Medicine had a 6.2% missing field rate for manpower shortages, which triggered an alarm in the OA system. The on-duty nurse had to enter the missing data within 5 minutes.

[0088] Specifically, the time-related processing flow is as follows:

[0089] 7:28 AM: The infrared sensing array in the emergency department of Campus A detected a sudden increase in the density of people in the waiting area, and generated heat map coordinate data every minute.

[0090] 7:30 AM: The HIS system registration records show that 22 new trauma patients were added within 10 minutes, including 5 critically injured patients. The system automatically marked the data timestamp as "2024-12-15 07:30:15".

[0091] Anomaly cleaning detected three registration records with timestamps of "2025-13-15" where the month value was overflowing. The cleaning unit replaced these records with the timestamp of the previous valid record, "2024-12-14 19:22:03", and triggered a manual review.

[0092] Implement sick leave data linkage, including:

[0093] 7:32 AM: The OA system pushes sick leave applications from Dr. Wang and Dr. Li of the Department of Respiratory and Critical Care Medicine in Campus B. During the winter flu season, the sick leave probability correction factor is activated, and the system automatically associates their vacant work hours from 08:00 to 17:00.

[0094] When the field was being supplemented, the missing field rate of the Respiratory and Critical Care Medicine Department exceeded the threshold. The on-duty nurse supplemented Dr. Zhang's training extension information through a mobile terminal. The data warehouse was updated in real time to show that there were 4 available personnel.

[0095] Surgical backlog analysis:

[0096] 7:35 AM: The EMR system reads the status of three operating rooms in the orthopedics department of Campus C:

[0097] Room 1: Total hip replacement surgery exceeded the time limit by 120 minutes;

[0098] Room 2: Knee arthroscopy surgery queue delay of 45 minutes;

[0099] The system automatically calculates that the average length of stay in this ward is expected to increase by 0.6 days this week, which exceeds 10%, triggering a yellow alert.

[0100] The multi-source data fusion module includes:

[0101] The API gateway cluster deployed on the hospital's intranet synchronizes full data at a second preset interval and captures incremental data at a third preset interval.

[0102] The data cleaning unit uses regular expressions to validate timestamp formats and removes outliers based on box plot algorithms.

[0103] When the field missing rate exceeds 5%, a real-time data entry alert is triggered to the office automation (OA) system.

[0104] During daily full synchronization, the consistency of patient IDs between the HIS system and the EMR system is automatically checked, and in case of conflict, the EMR system takes precedence.

[0105] Box plot algorithm identifies outliers in the number of registered patients, triggering a manual review process;

[0106] A data entry supplementation mechanism is adopted. If a field missing alarm remains unresolved for 30 minutes, the average value of the most recent three days will be automatically filled in.

[0107] In this embodiment, 127 cases of HIS / EMR patient ID conflicts were found during full synchronization, and the EMR version was used for automatic overwriting.

[0108] After outlier removal, the volatility of outpatient traffic data decreased by 76%;

[0109] The traffic prediction model was activated by inputting the time-series data of emergency room traffic from 7:00 to 9:00 AM for the past 90 days from Campus A. The LSTM model identified the Monday morning peak pattern:

[0110] The baseline number of people between 7:00 and 8:00 was 25; the predicted number between 8:00 and 9:00 today is 50.

[0111] The actual instantaneous flow at 8:00 was 45 people, with an error rate of 4%.

[0112] The winter correction Bayesian model, with an ambient temperature of -5°C, increased the baseline value for sick leave probability in the Department of Respiratory and Critical Care Medicine from 17% to the historical winter average of 32%.

[0113] Execute the second-level response trigger mechanism:

[0114] 8:00 AM: System statistics on the number of emergency patients received in Campus A within 30 minutes:

[0115] The number of trauma patients was 45 (including 5 critically ill patients), which is 150% of the number exceeding the threshold of 30.

[0116] Combined with mobile CT fault alarm (2 units pending repair) → meets the dual conditions for Level 2 response.

[0117] The dynamic strategy database was used to search for the "mass injury incident - orthopedics priority" rule for contingency plan matching, and the team led by Director Li of the joint group of the orthopedics department in Campus C was identified, with a historical response delay rate of 5.3%.

[0118] The operating logic of the intelligent prediction engine includes:

[0119] An LSTM neural network with 128 hidden neurons was used as input, along with hourly outpatient flow time-series data from the first cycle.

[0120] A Bayesian absence model was adopted, which automatically generates a dynamic seasonal correction factor by analyzing the correlation between historical absence data and meteorological data and statutory holiday dates.

[0121] Output a human resource demand forecast report with a heat map showing the distribution of departmental-level shortages;

[0122] The LSTM model is automatically retrained every week during a preset time period, and new data is added to the sliding time window.

[0123] The system's preset monthly parameters trigger the automatic loading of seasonal correction factors, increasing the baseline value for sick leave probability in the Department of Respiratory and Critical Care Medicine by 15%.

[0124] Generate a heat map report. Red areas indicate departments with a shortage of more than 5 people or a shortage of 15% or more for the same position. Yellow areas indicate departments with a shortage of 2-4 people or a shortage of 10% or more but less than 15% for the same position.

[0125] In this embodiment, "LSTM neural network" (a specific algorithm structure) and "Bayesian model" are technical algorithms used to handle uncertainty problems, while "automatic retraining" and "sliding time window" are technical mechanisms for the model to continuously learn and adapt to data changes.

[0126] After the winter mode was activated, the accuracy rate of sick leave prediction in the Department of Respiratory and Critical Care Medicine reached 91%.

[0127] The heat map shows a red alert for the trauma group in Area A of the hospital, indicating a shortage of more than 5 people.

[0128] The entire chain of Level 4 response decision-making and execution constitutes a closed loop for resource scheduling.

[0129] 8:05 AM: The OA system pushes a task order to Dr. Li's team in the Orthopedic Joint Group at Campus C:

[0130] Title: "Level II Response - Trauma Support for Campus A";

[0131] Requirements: 3 orthopedic surgeons + 1 anesthesiologist, latest arrival time 8:45;

[0132] Commuting route: Plan to avoid the morning rush hour congestion points on the Third Ring Road.

[0133] Execution process tracing includes:

[0134] 8:12 AM: The orthopedic joint group, led by Director Li, confirmed the task, and the system updated its status to "on the way" in real time.

[0135] 8:30 AM: The emergency department's queuing system automatically postpones the queue for non-critical patients;

[0136] 8:35 AM: The medical team clocks in with their fingerprints and handles trauma support surgeries.

[0137] In response to cross-campus collaboration, temporary vacancies in the Department of Respiratory and Critical Care Medicine at Campus B will be filled by physicians from Campus A providing support.

[0138] The system searched for on-duty physicians with qualifications in respiratory and critical care medicine and identified two physicians located 2.1 kilometers away from Campus B.

[0139] Automatically retain access to the electronic medical records of patients under its original jurisdiction.

[0140] The four-level response decision unit includes:

[0141] Level 1 Response Module: Initiating cross-hospital-wide team deployment and reserve personnel mobilization during major public health emergencies;

[0142] Level 2 Response Module: Targets inter-hospital specialist collaboration and multi-departmental coordination triggered by mass casualty incidents within a single hospital campus;

[0143] Level 3 Response Module: Implementing flexible scheduling and cross-departmental support for departments that are continuously overloaded;

[0144] Level 4 Response Module: Supports departmental-level autonomous fine-tuning of shift schedules and staff rotation;

[0145] The triggering condition for the fourth-level response decision unit is:

[0146] Level 1 Response: If the hospital-wide staffing shortage exceeds 30% or a red alert for public health emergencies is issued, the reserve staffing pool will be activated within 10 seconds of receiving the red alert from the platform.

[0147] Level 2 Response: If the number of emergency patients in a single hospital exceeds the threshold by 150% within 30 minutes, in the event of a mass casualty incident, a trauma specialist team within 10 kilometers of the target hospital will be automatically matched.

[0148] Level 3 Response: If a department is continuously overloaded for more than 48 hours and the average length of stay is expected to increase by more than 10%, after the overloaded department has been marked for 48 hours, a support request will be sent to the related departments.

[0149] Level 4 Response: When there are 2 or fewer vacancies due to scheduling conflicts or temporary leave in the department, the department will automatically retrieve the list of staff on leave when a vacancy for the early shift is detected.

[0150] In this embodiment, the first-level response is as follows: after issuing a red alert for influenza, 50 doctors from the reserve pool are activated within 120 seconds.

[0151] Level 3 Response: Orthopedics backlog at Campus C exceeds 48 hours → Automatic coordination with Campus A to support 2 surgical teams.

[0152] Specific indicators were collected in real time, including human resource efficiency, namely, the average number of cases handled per person in the orthopedics department of Hospital A during the event reached 8.2 cases / 4 hours, compared to a daily average of 4.5 cases; and service quality, including the reduction of the first consultation time for critically ill patients to 9 minutes, with a standard requirement of ≤15 minutes.

[0153] The scoring logic is based on the business load dimension score calculation:

[0154] Emergency resuscitation success rate 94% → Achievement rate 100% → Maximum score of 25 points for this item;

[0155] Bed turnover rate recovered to 89% → Achievement rate 92% → Score 23 points;

[0156] The overall score of 87 points was attributed to the fact that the outpatient waiting time was still 29 minutes, while the target value was 25 minutes.

[0157] The system detects winter sick leave prediction bias during model dynamic optimization:

[0158] Actual sick leave: 3 vs. predicted: 5 → Deviation rate: 40%. The winter correction factor weight is automatically reduced by 3%, and a "Winter Model Calibration Recommendation" is generated and pushed to the administrator.

[0159] Furthermore, the optimization mechanism of the dynamic strategy library includes:

[0160] Record the arrival delay rate and skill matching deviation for each deployment;

[0161] When the arrival delay rate is greater than 20%, the priority weight of the allocation path is automatically increased;

[0162] The effectiveness matrix of the generated strategy is used monthly to drive rule iteration;

[0163] After each scheduling, store the actual arrival time and skill utilization rate;

[0164] When the latency of a certain path exceeds 20%, the strategy is adjusted, its priority is automatically downgraded, and alternative paths are tested.

[0165] Perform monthly iterations, generate matrix reports to compare the efficiency of different hospital combinations, and obtain the optimal hospital combination.

[0166] In this embodiment, the record shows that the commuting delay rate from C to A campus is 18% (<20% threshold), and the priority of this route is retained.

[0167] The monthly matrix report indicates that the support efficiency of the B→C campus is 37% lower than that of the A→C campus, indicating a potential downgrade path.

[0168] Latency calculation includes:

[0169] The orthopedic joint group, led by Director Li, was scheduled to arrive at 8:45 AM but actually arrived at 8:35 AM, 10 minutes earlier than planned.

[0170] The historical latency rate for this path has been updated from 5.3% to 4.1%.

[0171] The system prioritizes the "Campus C → Campus A" pathway in traumatic events.

[0172] The indicator library of the closed-loop evaluation module includes:

[0173] Human resource efficiency dimensions: average number of medical visits per person, absenteeism impact coefficient;

[0174] Business load dimensions: emergency rescue success rate, shortage rate of rural training programs;

[0175] Service quality dimensions: medical error rate, waiting time compliance rate;

[0176] Human resource structure dimensions: absenteeism replacement rate, staff suitability;

[0177] Each indicator is weighted dynamically to calculate a comprehensive score out of 100.

[0178] Absence impact coefficient = (Actual number of employees on duty / Standard number of employees) × 100%;

[0179] Waiting time compliance rate = (Number of patients with waiting time <30 minutes / Total number of patients) × 100%;

[0180] The scoring rule is that when the achievement rate of a single indicator is less than 60%, the highest score for that dimension is 79 points, which is considered passing.

[0181] In this embodiment, the indicator library of the closed-loop evaluation module includes four distinct and quantifiable evaluation dimensions:

[0182] Human resource efficiency dimension: includes the indicators of "average number of outpatient visits" and "absence impact coefficient". Among them, the absence impact coefficient = (actual number of employees on duty / standard number of employees) × 100%, which is used to quantify the degree of matching between the on-duty human resources and the standard requirements.

[0183] Workload dimension: Includes indicators such as "emergency resuscitation success rate" and "bed occupancy rate". The data comes directly from the hospital information system (HIS) emergency resuscitation records and inpatient ward statistical reports, and is used to objectively reflect the clinical workload and quality.

[0184] Service quality dimension: includes the indicators of "medical error rate" and "waiting time compliance rate". Among them, the waiting time compliance rate = (number of patients with a waiting time of less than 30 minutes / total number of patients) × 100%, and this data is automatically collected by the triage system.

[0185] Human Resources Structure Dimension: Includes indicators of "Skill Matching" and "Cross-Campus Deployment Success Rate". Skill Matching is evaluated by the department head nurse in the OA system using a 100-point scale based on the professional skills of support personnel; Cross-Campus Deployment Success Rate = (Number of Successfully Arrived Personnel / Total Number of Deployment Initiations) × 100%.

[0186] All indicators across dimensions are combined and calculated into a 100-point overall score using preset dynamic weights. The system is set so that if the achievement rate of any single indicator in any dimension is below 60%, the maximum score for that dimension is limited to 79 points (the passing score).

[0187] In this embodiment, the system performs a monthly evaluation of the "Emergency Department" of "XX City People's Hospital Central Campus".

[0188] Data Collection and Calculation: The system automatically collected data from the HIS showing a total of 9,000 emergency department visits for the month, with 30 medical staff actually on duty and a standard staffing level of 35. Based on this, the following calculations were performed:

[0189] Average number of outpatient visits per person = 9000 / 30 = 300 visits / person.

[0190] Absence impact factor = (30 / 35) * 100% ≈ 85.7%.

[0191] Meanwhile, the head nurse of the department gave a score of 90 for the "skill matching degree" of the physician who provided cross-hospital support this month, obtained from the OA system.

[0192] Scoring Logic Application: The system compares the calculated values ​​with preset standard values. Assuming the calculated "absence impact coefficient" achievement rate is 85.7%, which is higher than the 60% threshold, this indicator can be included in the score based on its actual value. If the achievement rate of another indicator, "bed occupancy rate," is only 55% (lower than 60%), then the overall score for the "workload dimension" will be limited to 79 points.

[0193] Results and Feedback: The system ultimately generated a comprehensive score of 82 for the department. Since the score was greater than 80 but less than 90, the system determined it to be at a "good" level and provided improvement suggestions in the comprehensive evaluation report, such as "It is recommended to pay attention to bed occupancy rate and optimize patient admission and discharge procedures."

[0194] Model self-optimization: The scoring result is simultaneously fed back to the intelligent prediction engine as part of the historical data, which is used to fine-tune the weights of the future prediction model, making it more accurate in predicting the department's manpower needs, thereby achieving closed-loop optimization.

[0195] Furthermore, the application logic of the scoring results is as follows: if the score is less than 70 points for two consecutive quarters, a special optimization plan will be triggered, and personnel adjustment suggestions will be automatically pushed to the human resources system. In the teaching hospital scenario, the weight of "training coverage rate" will be increased to 20%.

[0196] A hospital with consistently low ratings will trigger staff optimization, automatically comparing the average number of outpatient visits per patient with similar hospitals.

[0197] Under the teaching hospital model, the weighting of continuing education coverage rate will be adjusted, increasing it from 10% to 20%.

[0198] After personnel suggestions are pushed to the HR system, the system automatically reserves recruitment slots and executes the linkage.

[0199] In this embodiment, Campus B received a score of 68 for two consecutive quarters, triggering a special program.

[0200] Automatically compare the average number of outpatient visits per patient in Campus A (24.2 cases) versus Campus B (18.1 cases);

[0201] The staffing recommendations were pushed to the HR system: "Increase staffing by 2 people" and "Add 2 physicians in the Department of Respiratory and Critical Care Medicine".

[0202] In addition, the system also includes:

[0203] An integrated visualization and interaction layer is used to render a heat map of the hospital's manpower saturation through the ECharts engine. It adopts a three-color warning system (red, yellow, and green), supports drag-and-drop adjustment of the scheme, and displays the changes of indicators in real-time pop-up windows. The mobile APP pushes a task progress tracking view when the response level is three or higher.

[0204] When clicking on a red overloaded department on the heat map, a list of available personnel and their estimated arrival time will pop up.

[0205] Drag and drop doctors to the target department to adjust treatment plans, and display real-time changes in predicted waiting times;

[0206] Mobile monitoring is used, and the app pushes a task progress bar when a level 2 response is initiated.

[0207] In this embodiment, the administrator drags an anesthesiologist from Campus C to Campus A → a real-time pop-up message appears: "The backlog of surgeries is expected to decrease by 3, but the load of the delivery center in Campus B has risen to a yellow alert."

[0208] The app pushed a level-two response progress notification: "The orthopedic joint group's Director Li's team has set off," and actually arrived at 8:35.

[0209] Furthermore, the system also includes:

[0210] The system employs a distributed computing architecture, with the intelligent prediction engine deployed on a GPU server cluster, the response decision unit running in an in-memory database, and the data warehouse using columnar storage to optimize concurrent queries for thousands of users; it connects to the OA systems of each campus through a microservice API gateway.

[0211] GPU clusters perform parallel computation of traffic predictions for each campus and accelerate the prediction process.

[0212] The in-memory database preloads the strategy library for real-time decision-making, with the response triggering and solution generation taking less than 10 seconds.

[0213] The OA system verifies the status of personnel.

[0214] Furthermore, the system includes a sandbox simulation module, where administrators can modify commuting time thresholds and response level parameters. The system automatically simulates the load balancing change curve over 72 hours and outputs a comparison report of the optimized and baseline solutions.

[0215] Modify the administrator's commute time parameter, and the system simulates cross-campus scheduling simulation data within the first preset time period;

[0216] The report output compares data on reduced waiting times and lower labor costs associated with the new plan.

[0217] In this embodiment, the "sandbox simulation module" is a simulation computing environment that allows administrators to modify parameters (such as commuting time), and the system can automatically perform large-scale computational simulations and output quantitative comparison reports. This demonstrates the simulation computing capabilities of complex systems.

[0218] In summary, this invention constructs a complete technical architecture system, including:

[0219] 1) Based on distributed computing and microservices data fusion technology, the problem of real-time integration and cleaning of multi-source heterogeneous medical data has been solved;

[0220] 2) The prediction algorithm combining LSTM and Bayesian models improves the accuracy and timeliness of human resource demand forecasting;

[0221] 3) A four-level response decision-making mechanism based on a rule engine and real-time computing was established, achieving a second-level response from early warning to solution generation;

[0222] 4) An automated closed-loop control system that includes data acquisition, prediction, decision-making, execution, and feedback was designed and implemented, which automatically drives the iterative optimization of models and strategies through quantitative evaluation results.

[0223] Therefore, the present invention is first and foremost an improved computer system that, through the comprehensive application of the aforementioned technical means, achieves efficient collaboration and flexible scheduling of human resources across multiple hospital campuses at the technical level, significantly improving the system response efficiency and medical resource utilization rate.

[0224] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between apparatuses or units, and may be electrical, mechanical, or other forms.

[0225] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in hardware or as software functional units. The above are merely embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made based on the description and drawings of this application, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

[0226] The specific embodiments of the invention have been described in detail above, but they are only examples, and this application is not limited to the specific embodiments described above. For those skilled in the art, any equivalent modifications or substitutions to the invention are also within the scope of this application. Therefore, all equivalent changes, modifications, and improvements made without departing from the spirit and principles of this application should be covered within the scope of this application.

Claims

1. A multi-campus human dynamic deployment decision support system based on a hierarchical response mechanism, characterized in that, Comprise: A multi-source data fusion module that collects patient diagnosis and treatment data from hospital information systems (HIS) and electronic medical record systems (EMR) and medical staff scheduling and on-duty status data from office automation systems (OA) in real time through standardized interfaces, cleans and formats the data through ETL technology, and constructs a distributed data warehouse; An intelligent prediction engine that uses an LSTM neural network to predict patient flow in each hospital department for a first predetermined period of time, calculates the absence probability using a Bayesian model, and outputs a dynamic manpower gap quantification value; A four-level response decision unit that performs resource scheduling in a predetermined emergency scenario; A dynamic strategy library that stores deployment paths, scheduling adjustment rules, and priority strategies corresponding to different response levels, and dynamically optimizes resource scheduling parameters based on historical execution results; A closed-loop evaluation module that constructs a four-dimensional index library of manpower efficiency, business load, service quality, and manpower structure, generates a hospital comprehensive evaluation report based on a percentage grading system, and automatically corrects the prediction model weights based on the grading results; The operation logic of the intelligent prediction engine includes: Using an LSTM neural network to set up a 128 neuron hidden layer and inputting hourly outpatient flow time series data in the first period; Using a Bayesian absence model that automatically generates dynamic seasonal correction factors by analyzing the correlation between historical absence data, meteorological data, and statutory holiday dates; Outputting a manpower demand prediction report labeled with a department-level gap distribution heat map; Automatically retraining the LSTM model every week in a predetermined time period and adding new data to a sliding time window; The system's preset monthly parameter triggers the automatic loading of the seasonal correction factor, and the benchmark value of the respiratory and critical care medicine department's absence probability is increased by 15%; Generating a heat map report, with red areas indicating departments with a gap of >5 people or a same-post gap of ≥15%, and yellow areas indicating departments with a gap of 2-4 people or a same-post gap of ≥10% but <15%; The four-level response decision unit includes: A first-level response module that initiates cross-hospital district team scheduling and reserves manpower in major public health events; A second-level response module that triggers inter-hospital district specialist collaboration and multi-department coordination for single-hospital district mass injury events; A third-level response module that performs flexible scheduling and cross-department support for continuously overloaded departments; A fourth-level response module that supports department-level self-tuning scheduling and the calling of rest personnel; The triggering conditions for the four-level response decision unit are: First-level response: hospital-wide manpower gap >30% or public health event red alert issued, after receiving the red alert from the warning platform, reserve manpower library is activated within 10 seconds; Second-level response: single-hospital district emergency department 30-minute reception volume exceeds threshold of 150%, and in mass injury events, automatically match trauma specialist teams within <10 km of the target hospital district; Third-level response: department continuously overloaded for >48 hours and average hospital stay growth <80%, after the overloaded department is marked for 48 hours, send a request for assistance to associated departments; Fourth-level response: department scheduling conflict or temporary leave vacancy ≤2 people, when a morning shift vacancy is detected, automatically retrieve the department's rest personnel list.

2. The multi-campus human dynamic deployment decision support system based on hierarchical response mechanism according to claim 1, characterized in that, The multi-source data fusion module includes: The API gateway cluster deployed in the hospital intranet synchronizes full data with a second preset time length as the cycle and captures incremental data at intervals of a third preset time length; The data cleaning unit checks the timestamp format using regular expressions and removes outliers based on the box plot algorithm; When the field missing rate is >5%, trigger real-time recording alarm to the office automation OA system; During daily full synchronization, automatically verify the patient ID consistency of HIS system and EMR system, and use EMR as the standard when there is a conflict; The box plot algorithm identifies abnormal values in the number of registrations and triggers an artificial review process; Use the recording mechanism, if the field missing alarm lasts for 30 minutes without processing, automatically fill in the average of the last three days. 3.The multi-campus human dynamic deployment decision support system based on hierarchical response mechanism according to claim 1, wherein, The optimization mechanism of the dynamic strategy library includes: Record the on-site delay rate and skill matching deviation of each deployment; When the on-site delay rate >20%, automatically increase the priority weight of the deployment path; Generate an effectiveness matrix of the strategy every month to drive iteration; Store the actual arrival time and skill usage rate after each dispatch; When the delay rate of a certain path >20%, adjust the strategy, automatically degrade its priority, and test alternative paths; Perform monthly iteration to generate a matrix report to compare the efficiency of different hospital combinations and obtain the preferred hospital combination.

4. The multi-campus human dynamic deployment decision support system based on hierarchical response mechanism according to claim 1, characterized in that, The index library of the closed-loop evaluation module includes: Human efficiency dimension: average number of diagnoses and treatments per person, absence impact coefficient; Business load dimension: emergency rescue success rate, gap rate of rural training; Service quality dimension: medical error rate, waiting time compliance rate; Human structure dimension: absence replacement rate, personnel compliance rate; Each index is calculated as a percentage of the comprehensive score according to the dynamic weight; Absence impact coefficient = (actual number of people on duty / standard number of people) x 100%; Waiting time compliance rate = (<30 minutes patient number / total patient number) x 100%; The scoring rule is that when the single index achievement is <60%, the highest score of this dimension is 79 points to pass.

5. The multi-campus human dynamic deployment decision support system based on hierarchical response mechanism according to claim 4, characterized in that, The application logic of the scoring result is that if the score of the last two quarters is <70 points, trigger a special optimization plan, automatically push personnel adjustment suggestions to the human resources system, and in the teaching hospital scenario, increase the "training coverage rate" weight to 20%; Trigger personnel optimization for consecutive low-scoring hospitals, automatically compare the average number of diagnoses and treatments per person with similar hospitals; Adjust the weight in the teaching hospital mode, increase the training coverage rate weight from 10% to 20%; After pushing the personnel suggestions to the HR system, automatically associate the personnel establishment quantity suggestion value and promote the personnel gap quantity, and link with the recruitment system.

6. The multi-campus human dynamic deployment decision support system based on hierarchical response mechanism according to claim 1, characterized in that, It also includes: Integrate the visual interactive layer, render the hospital human saturation heat map through the ECharts engine, use red, yellow and green three colors for early warning, support drag adjustment scheme and real-time pop-up display of index changes, and push task progress tracking view on mobile APP when the response is above three levels; Click on the red overloaded department in the heat map to pop up a list of schedulable personnel and an estimated arrival time; Drag the doctor to the target department to adjust the scheme, and display the real-time waiting time prediction value changes; Use mobile monitoring, and APP pushes the task progress bar when the response is at the secondary level.

7. The multi-campus human dynamic deployment decision support system based on hierarchical response mechanism according to claim 1, characterized in that, It also includes: Adopting distributed computing architecture, the intelligent prediction engine is deployed on GPU server cluster, the response decision unit runs on in-memory database, and the data warehouse adopts columnar storage to optimize concurrent query of thousands of people; Through the micro-service API gateway, connect the OA systems of each hospital area; GPU cluster parallel computing each hospital area traffic prediction and prediction acceleration; In-memory database preloading strategy library for real-time decision-making, response trigger to scheme generation <10 seconds; OA system receives scheduling instructions through RESTful API for micro-service integration, and returns personnel confirmation status.

8. The multi-campus human dynamic deployment decision support system based on hierarchical response mechanism according to claim 1, characterized in that, Also includes: Set up a sandbox simulation module, the administrator modifies the commuting time threshold and response level parameters, the system automatically simulates the load balancing change curve within 72 hours, and outputs the index comparison report of the optimization scheme and the baseline scheme; Modify the administrator's commuting time parameters, and the system simulates the cross-hospital area scheduling simulation running data within the first preset time period; Report output comparison shows the waiting time reduction data and human cost data of the new scheme.

Images