Intelligent emergency pre-examination triage method and system based on big data

Abstract

The invention discloses an intelligent emergency pre-examination triage method and system based on big data, and relates to the technical field of pre-examination triage. According to the intelligent emergency pre-examination triage method based on the big data, the pre-examination data of the patient including the real-time physiological monitoring data and the main information is obtained, the pre-triage department and the disease deterioration probability are determined, then the treatment capacity coefficient of the pre-triage department is evaluated, and therefore the final triage strategy is determined, accurate matching of the departments can be achieved based on the big data, and the accuracy of the triage strategy is improved. The condition deterioration probability is dynamically evaluated, and the department treatment ability is comprehensively considered, so that a scientific triage strategy is formulated, the emergency triage efficiency and accuracy are improved, and a more timely and effective treatment opportunity is won for patients; the problems that in traditional emergency pre-examination triage, department matching is not accurate, disease deterioration risk assessment is not timely, and hospital disposal capacity is not sufficiently considered can be solved.

Description

Technical Field

[0001] This invention relates to the field of pre-screening and triage technology, specifically to an intelligent emergency pre-screening and triage method and system based on big data. Background Technology

[0002] In emergency medical services, rapid and accurate triage of patients is a key step in ensuring timely treatment and optimizing the allocation of medical resources.

[0003] Chinese patent application CN120032835A discloses an emergency triage method and system, relating to the field of emergency medical technology. The method includes: after confirming the patient's identity information, obtaining the patient's vital signs information and chief complaint information; triaging the patient based on a pre-stored triage knowledge base, vital signs information, and chief complaint information, generating the patient's triage information; the triage knowledge base stores vital signs information and chief complaint information corresponding to different triage levels; and displaying the triage information.

[0004] However, the traditional emergency triage process suffers from problems such as inaccurate department matching, untimely assessment of the risk of worsening condition, and insufficient consideration of the hospital's handling capacity. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent emergency triage method and system based on big data, which solves the problems of inaccurate department matching, untimely assessment of the risk of worsening condition, and insufficient consideration of hospital treatment capacity in traditional emergency triage.

[0006] To achieve the above objectives, the present invention provides the following technical solution: an intelligent emergency triage method based on big data, comprising the following steps: acquiring patient triage data; determining the patient's pre-triage department and the probability of disease deterioration based on the patient triage data; the patient triage data includes real-time physiological monitoring data and patient description information; the real-time physiological monitoring data f1 includes circulatory system parameters, respiratory system parameters, nervous system parameters, and characteristic monitoring parameters; the patient description information c1 includes pain location, pain intensity, pain duration, and past medical history; assessing the treatment capacity of the pre-triage department and determining the emergency treatment capacity coefficient; determining the final triage strategy based on the probability of disease deterioration and the emergency treatment capacity coefficient; the final triage strategy includes treatment in the pre-triage department of the hospital and transfer to another hospital for treatment: if the probability of disease deterioration is greater than the disease deterioration threshold stored in the database, and the emergency treatment capacity coefficient is less than the emergency treatment capacity threshold stored in the database, then the final triage strategy is determined to be transfer to another hospital for treatment, and the selected transfer hospital is determined based on multi-objective optimization; otherwise, the final triage strategy is determined to be treatment in the pre-triage department of the hospital.

[0007] Furthermore, determining the patient's pre-triage department and the probability of disease deterioration based on patient pre-screening data includes the following steps: obtaining the disease feature datasets of each emergency department stored in the database, including disease feature data, which includes physiological parameter feature data f2 and parameter description information c2; performing similarity matching between the patient pre-screening data and the disease feature data in the disease feature datasets of each emergency department to obtain the disease similarity matching value JB. xc JB xc =σ(f1, f2) + σ(c1, c2), where σ(·) is the cosine similarity function; determine the main disease feature data corresponding to the largest disease similarity matching value, and record it as the most similar main disease feature data. Determine the corresponding emergency department as the patient's pre-triage department through the main disease feature dataset to which the most similar main disease feature data belongs;

[0008] Retrieve the diagnostic data-deterioration probability mapping set corresponding to the most similar disease characteristics data from the database. This set includes several diagnostic data-deterioration probability branches, and each diagnostic data includes several historical physiological monitoring data fb. i According to the set time interval, physiological monitoring sample data cb corresponding to several time points are obtained from the real-time physiological monitoring data. i The similarity analysis of physiological monitoring sample data and historical physiological monitoring data was performed to obtain the monitoring data similarity comparison value JC. xs JC xs =∑σ(cb′) i ,fb′ i ); Determine the diagnostic data corresponding to the largest monitoring data similarity comparison value, and determine the predicted probability of disease deterioration;

[0009] Dynamic physiological time-series subsequences are obtained based on real-time physiological monitoring data, and static risk factor subsequences are also obtained. The dynamic physiological time-series subsequences and static risk factor subsequences are concatenated to obtain a fused feature sequence. The auxiliary probability of disease deterioration is determined using a Transformer-Bayes model. The predicted probability of disease deterioration and the auxiliary probability of disease deterioration are weighted and summed to obtain the probability of disease deterioration.

[0010] Furthermore, the Transformer-Bayes model includes a Transformer module and a Bayesian inference module. The Transformer-Bayes model is used to determine the auxiliary probability of disease deterioration, which includes the following steps: the fused feature sequence is input into the Transformer module, which includes a multi-layer encoder and an output layer. Each encoder layer includes a multi-head self-attention mechanism and a feedforward neural network. After the encoder layers in the Transformer module process the fused feature sequence, the processing results are input into the output layer. The output layer uses the Sigmoid activation function to generate the auxiliary probability of disease deterioration at multiple future time points.

[0011] The Bayesian inference module samples the data using the Monte Carlo dropout method to obtain an auxiliary probability distribution sample of disease deterioration at each future time point. It calculates the mean of this auxiliary probability distribution sample as a reference value for the deterioration probability at that time point, and uses the reference value to calculate the sample standard deviation as an uncertainty index. This determines the probability distribution interval for each time point, which is the reference value for deterioration ± 2 times the sample standard deviation. Based on the probability distribution interval-confidence factor mapping set stored in the database, it determines the confidence factor corresponding to the probability distribution interval for each time point, which serves as the weighting factor for the auxiliary probability of disease deterioration. The sum of the weighting factor for the auxiliary probability of disease deterioration and the weighting factor for the predicted probability of disease deterioration is set to 1, and the probability of disease deterioration is then calculated.

[0012] Further, the emergency treatment capacity coefficient is determined by the following steps: obtaining emergency department treatment capacity assessment data, including emergency physician response data, departmental resource data, and departmental environment data; determining the emergency physician response factor TVI based on the emergency physician response data; determining the departmental resource coupling factor KsO based on the departmental resource data; determining the environmental impact factor HjY based on the departmental environment data; and determining the emergency treatment capacity coefficient CZ by combining the emergency physician response factor TVI, the departmental resource coupling factor KsO, the environmental impact factor HjY, and the probability of patient deterioration Pe. nl :

[0013] Where α1, α2, β1 and β2 are all weighting factors, and e is a natural constant.

[0014] Furthermore, the emergency physician response data includes professional matching degree, physician experience index, cumulative fatigue coefficient, and workload index. The acquisition process is as follows: Based on the most similar primary disease feature data, the corresponding disease label is determined; the primary disease label set of each emergency physician in the patient pre-triage department is obtained; and the proportion of all primary disease label sets containing the corresponding disease label is calculated and denoted as professional matching degree B. zzThe doctor experience index YjY is obtained by multiplying the seniority of each emergency physician by the success rate of each case, summing and averaging the results; the average working hours of emergency physicians are obtained and compared with the working hours-fatigue coefficient mapping set stored in the database to obtain the cumulative fatigue coefficient Pl; the warning threshold for the probability of disease deterioration stored in the database is obtained, the proportion of patients currently being treated in the department whose probability of disease deterioration is greater than the warning threshold is calculated, the ratio of the number of patients waiting in the department to the number of patients at the critical waiting point is obtained, and the proportion and the ratio are weighted and summed to obtain the load index Fh.

[0015] The first response factor (TVI1) for emergency physicians was determined based on emergency physician response data.

[0016]

[0017] Wherein, λ1, λ2 and λ3 are all weighting factors, and yjY is the benchmark value of the doctor's experience index;

[0018] Obtain static basic data and dynamic work data of emergency physicians. Determine the second response factor TVI2 of emergency physicians based on the static basic data and dynamic work data of emergency physicians. Perform a weighted summation of the first response factor TVI1 and the second response factor TVI2 of emergency physicians to obtain the emergency physician response factor TVI.

[0019] Departmental resource data includes the percentage of available equipment (Ks). b and bed turnover coefficient Cw z Determine the departmental resource coupling factor based on departmental resource data:

[0020]

[0021] Where, N cl To determine the number of beds ready for disinfection, N zg H represents the total number of beds. pb The nursing care matching factor, which is the ratio of the number of nurses to the number of patients, Gr jq This represents the infection rate over the past month.

[0022] Departmental environmental data includes personnel density per unit area (ρ). sj Electromagnetic interference intensity Dc sj and aerosol concentration qr sj Based on departmental environmental data, the environmental impact factor HjY was determined.

[0023]

[0024] ρ jz Dc is the threshold value for personnel density per unit area. jz qr is the threshold value for electromagnetic interference intensity. jzThis is the threshold value for aerosol concentration.

[0025] Furthermore, the static basic data of emergency physicians includes seniority, success rate of cases, and disease labels of patients treated. The dynamic work data of emergency physicians includes the number of patients seen per unit time, time spent on a single case, team collaboration response speed, continuous working hours, medical record completeness score, and recent misdiagnosis / missed diagnosis records. The second response factor (TVI2) for emergency physicians is determined by the following steps:

[0026] After standardizing and integrating the static basic data and dynamic work data of emergency physicians, an original feature vector was formed. Deep feature extraction was performed on the original feature vector using a Deep Belief Network (DBN) to obtain a deep feature vector, including clinical experience depth, professional domain matching depth, collaboration efficiency coefficient, and work stability index. Physiological indicator data of emergency physicians was acquired, and an initial observation sequence arranged in chronological order was obtained based on the deep feature vector and the physiological indicator data. The initial observation sequence was weighted to obtain a dynamic observation sequence with attention weights. A Hidden Markov Model (HMM) was trained based on the dynamic observation sequence with attention weights to obtain the probability distribution of hidden states, including highly focused states, mild fatigue states, cognitive overload states, and collaboration-dependent states. Weighting factors for the deep feature vector were determined based on reinforcement learning and the probability distribution of the hidden states. The deep feature vectors were then weighted and summed to obtain the emergency physician's second response factor, TVI2.

[0027] Furthermore, the selection of hospitals to be transferred based on multi-objective optimization includes the following steps: constructing a multi-objective function and setting constraints; using triplets composed of [candidate hospital identifier, transfer route number, departure timestamp] as basic gene units to form an initial chromosome population; performing a screening of the initial chromosome population based on the constraints to obtain a screened initial chromosome population; performing Pareto front analysis and approximation ideal solution sorting on the screened initial chromosome population to obtain initial candidate basic gene units; performing genetic operations on the initial chromosome population, specifically: performing a two-point crossover operation on the candidate hospitals, a single-point mutation operation on the transfer route, and a Gaussian distribution-based mutation operation on the departure time to form new basic gene units, constituting a new chromosome population;

[0028] The new chromosome population is screened again based on constraints, followed by Pareto front analysis and ranking of near-ideal solutions to obtain candidate basic gene units after a second iteration. After several iterations, all candidate basic gene units are summarized. A fitness function is constructed based on a multi-objective function to determine the candidate basic gene unit corresponding to the maximum fitness function value, thus obtaining the selected hospitals for transfer. The fitness function F is formulated as follows:

[0029] Further, Pareto front analysis and approximation ideal solution ranking are performed, including the following steps: For any two basic gene units A and B, if the emergency treatment capability coefficient and cosine similarity of treatment plan of A are not less than the emergency treatment capability coefficient and cosine similarity of treatment plan of B, and at least one of the emergency treatment capability coefficient and cosine similarity of treatment plan of A is greater than the emergency treatment capability coefficient and cosine similarity of treatment plan of B, then A is said to dominate B; Basic gene units not dominated by other basic gene units are formed into a first-level non-dominated set, which serves as the first layer of the Pareto front and as the optimal solution set obtained from the Pareto front analysis; Approximation ideal solution ranking is performed on the optimal solution set: the positive ideal solution of the multi-objective function stored in the database is obtained; Euclidean distance is calculated between the actual solution of the multi-objective function of each basic gene unit in the optimal solution set and the positive ideal solution, and the basic gene unit corresponding to the smallest Euclidean distance value is recorded as the candidate basic gene unit.

[0030] Furthermore, the multi-objective function includes: the emergency treatment capacity coefficient CZ of the corresponding department of the optional transfer hospital. nl ′;The cosine similarity σ(ZL1, ZL2) between the current treatment plan ZL1 and the alternative transfer hospital plan ZL1;

[0031] The constraints include: CZ nl CZ nl ′ min ;σ(ZL1, ZL2)>σ(ZL1, ZL2) min CZ nl ′ min To minimize the disposal capacity coefficient, σ(ZL1, ZL2) min This represents the minimum similarity value.

[0032] A big data-based intelligent emergency triage system includes a triage module for acquiring patient triage data and determining the pre-triage department and probability of disease deterioration based on the patient triage data. The patient triage data includes real-time physiological monitoring data and patient description information. Real-time physiological monitoring data f1 includes circulatory system parameters, respiratory system parameters, nervous system parameters, and characteristic monitoring parameters. Patient description information c1 includes pain location, pain intensity, pain duration, and past medical history. A treatment capacity assessment module is used to evaluate the treatment capacity of the pre-triage department and determine the emergency treatment capacity coefficient. A triage module is used to determine the final triage strategy based on the probability of disease deterioration and the emergency treatment capacity coefficient. The final triage strategy includes treatment in the pre-triage department of the hospital and transfer to another hospital: if the probability of disease deterioration is greater than the disease deterioration threshold stored in the database, and the emergency treatment capacity coefficient is less than the emergency treatment capacity threshold stored in the database, then the final triage strategy is determined to be transfer to another hospital, and the selected transfer hospital is determined based on multi-objective optimization; otherwise, the final triage strategy is determined to be treatment in the pre-triage department of the hospital.

[0033] The present invention has the following beneficial effects:

[0034] This big data-based intelligent emergency triage method acquires pre-screening data from patients, including real-time physiological monitoring data and patient descriptions, to determine the pre-triage department and the probability of disease deterioration. It then assesses the treatment capacity coefficient of the pre-triage department to determine the final triage strategy. Based on big data, it achieves precise department matching, dynamically assesses the probability of disease deterioration, comprehensively considers departmental treatment capabilities, and formulates scientific triage strategies. This improves the efficiency and accuracy of emergency triage, securing more timely and effective treatment opportunities for patients. It addresses the problems of inaccurate department matching, untimely assessment of disease deterioration risk, and insufficient consideration of hospital treatment capabilities in traditional emergency triage.

[0035] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0036] Figure 1 This is a flowchart of the intelligent emergency triage method based on big data according to the present invention.

[0037] Figure 2 This is a flowchart of the intelligent emergency triage device based on big data according to the present invention. Detailed Implementation

[0038] Please see Figure 1 The present invention provides a technical solution: an intelligent emergency triage method based on big data, comprising the following steps: acquiring patient triage data, determining the patient's pre-triage department and the probability of disease deterioration based on the patient triage data, wherein the patient triage data includes real-time physiological monitoring data and patient description information;

[0039] Real-time physiological monitoring data f1 includes circulatory system parameters (heart rate, blood pressure, blood oxygen saturation, and electrocardiogram data), respiratory system parameters (respiratory rate, airway pressure, and end-tidal carbon dioxide), neurological system parameters (Glasgow Coma Scale, binary pupillary light reflex markers (marked 1 for presence, 2 for absence), and EEG entropy), and characteristic monitoring parameters (invasive blood pressure, central venous pressure, and cardiac output); the patient's subjective information c1 includes pain location, pain intensity, pain duration, and past medical history. By integrating real-time physiological monitoring data with the patient's subjective information, a comprehensive patient profile is constructed. Real-time physiological parameters reflect the current functional state of the body, while the patient's subjective information supplements subjective symptoms and medical history, forming a two-dimensional data system of "objective indicators + subjective description."

[0040] Determining the pre-triage department and the probability of disease deterioration based on patient pre-screening data includes the following steps: Obtaining the disease feature datasets of each emergency department stored in the database, including feature data for each disease, which includes physiological parameter feature data f2 and parameter description information c2; performing similarity matching between the patient pre-screening data and the disease feature data in the disease feature datasets of each emergency department to obtain a disease similarity matching value JB. xc The main disease feature data corresponding to the largest disease similarity matching value is identified and denoted as the most similar main disease feature data. The corresponding emergency department is determined as the patient's pre-triage department by the main disease feature dataset to which the most similar main disease feature data belongs.

[0041] Disease similarity matching value JB xc :

[0042] JB xc =σ(f1, f2) + σ(c1, c2), where σ(·) is the cosine similarity function;

[0043] The system calculates cosine similarity between patient data and the emergency department's disease feature database, enabling precise departmental triage through quantified matching. The disease feature data includes physiological parameters (such as typical ST-segment elevation and abnormal blood pressure in myocardial infarction patients) and patient complaints (such as a description of substernal squeezing pain in angina patients), forming standardized disease templates. The disease feature database, trained on large datasets, covers common diseases across multiple departments. The cosine similarity formula is used to calculate the matching degree between patient data and each disease template, avoiding subjective judgment errors. For example, if a patient's ECG shows ST-segment elevation and they complain of substernal pain, the system can quickly match the acute myocardial infarction feature template from the cardiology department, directly triaging them to cardiology and reducing triage errors.

[0044] Automated matching replaces traditional manual inquiry and judgment, shortening triage time. It is especially suitable for large numbers of emergency patients (such as mass trauma events) and ensures that the triage principle of "prioritizing the serious and the urgent" is implemented.

[0045] Retrieve the diagnostic data-deterioration probability mapping set corresponding to the most similar disease characteristics data from the database. This set includes several diagnostic data-deterioration probability branches, and each diagnostic data includes several historical physiological monitoring data fb. i According to the set time interval, physiological monitoring sample data cb corresponding to several time points are obtained from the real-time physiological monitoring data. i The similarity analysis of physiological monitoring sample data and historical physiological monitoring data was performed to obtain the monitoring data similarity comparison value JC. xs Determine the diagnostic data corresponding to the largest similarity comparison value of the monitoring data, and determine the predicted probability of disease deterioration.

[0046] Similarity comparison value of monitoring data:

[0047] JC xs =∑σ(cb) i ′,fb i ′).

[0048] A "diagnostic data-deterioration probability mapping set" is used to establish a correlation between historical data and current condition. This mapping set stores historical physiological monitoring data for different disease types (such as blood pressure and heart rate change sequences for patients with septic shock) and their corresponding deterioration probabilities. Current physiological data is sampled at time intervals (such as heart rate and blood oxygen samples extracted every 5 minutes), and similarity analysis is performed with historical data. Finally, the deterioration probability is determined by the maximum similarity match.

[0049] Quantitative analysis based on historical case data can identify potential critical illnesses in advance. For example, if a patient's initial blood pressure is normal but their heart rate continues to rise, the system can provide an early warning of the probability of the condition worsening by comparing the data with historical septic shock data. This alerts medical staff to prepare resuscitation measures in advance. Sampling at time intervals (e.g., every 10 minutes) can capture sudden changes in the condition, avoiding the lag of static assessments.

[0050] Based on real-time physiological monitoring data, dynamic physiological time-series subsequences are obtained (at 5-minute intervals, parameters such as heart rate, blood pressure, blood oxygen saturation, and respiratory rate are extracted from real-time physiological monitoring data to form dynamic physiological time-series subsequences (e.g., containing 12 consecutive time points, i.e., 60 minutes of data)). At the same time, static risk factor subsequences are obtained (e.g., history of diabetes marked as 1 / 0, history of hypertension marked as 1 / 0, age risk coefficient, and basic organ function score). The dynamic physiological time-series subsequences and static risk factor subsequences are concatenated, and the static risk factor subsequences are expanded to have the same length as the dynamic physiological time-series subsequences (each time point carries the same static features) to obtain a fused feature sequence.

[0051] The auxiliary probability of disease deterioration was determined using the Transformer-Bayesian model; the predicted probability of disease deterioration and the auxiliary probability of disease deterioration were weighted and summed to obtain the probability of disease deterioration.

[0052] The Transformer-Bayesian model includes a Transformer module and a Bayesian inference module. Determining the auxiliary probability of disease deterioration using the Transformer-Bayesian model involves the following steps: The fused feature sequence is input into a Transformer module comprising multiple encoders and an output layer. In this embodiment, it is set to four layers. Each encoder layer includes a multi-head self-attention mechanism and a feedforward neural network (hidden layer dimension is 256). In this embodiment, an eight-head self-attention mechanism is used. After processing the fused feature sequence, each encoder layer in the Transformer module inputs the processing result into the output layer. The output layer uses the Sigmoid activation function to generate the auxiliary probability of disease deterioration at multiple future time points.

[0053] The Bayesian inference module uses the Monte Carlo dropout method for sampling (dropout rate set to 0.15, 100 samplings) to obtain the auxiliary probability distribution sample of disease deterioration at each future time point. The mean of the auxiliary probability distribution sample is calculated as the reference value of the deterioration probability at that time point. The standard deviation of the sample is calculated using the deterioration probability reference value as an uncertainty index, thereby determining the probability distribution interval for each time point, i.e., the deterioration probability reference value ± 2 times the sample standard deviation. Based on the probability distribution interval-confidence factor mapping set stored in the database, the confidence factor corresponding to the probability distribution interval of each time point is determined as the weight factor of the auxiliary probability of disease deterioration. The sum of the weight factor of the auxiliary probability of disease deterioration and the weight factor of the predicted probability of disease deterioration is set to 1, and the probability of disease deterioration is calculated.

[0054] Assess the handling capacity of pre-triage departments and determine the emergency handling capacity coefficient;

[0055] Acquire emergency department capacity assessment data, including emergency physician response data, departmental resource data, and departmental environment data; determine the emergency physician response factor TVI based on the emergency physician response data; determine the departmental resource coupling factor KsO based on the departmental resource data; determine the environmental impact factor HjY based on the departmental environment data; and determine the emergency department capacity coefficient CZ by combining the emergency physician response factor TVI, departmental resource coupling factor KsO, environmental impact factor HjY, and the probability of patient deterioration Pe. nl :

[0056] Where α1, α2, β1 and β2 are all weighting factors, and e is a natural constant.

[0057] Emergency physician response data (such as professional matching degree and fatigue coefficient) avoids the dilemma of "having equipment but no doctors"; departmental resource data (percentage of available equipment and bed turnover coefficient) prevents the bottleneck of "having doctors but no resources"; environmental data (personnel density and aerosol concentration) avoids the risk of "environmental interference with treatment" (such as high electromagnetic interference affecting the accuracy of monitoring equipment). All three types of data are collected in real time (such as the dynamic change of physician workload index with the number of waiting patients and real-time updates of bed status), ensuring that the assessment results accurately reflect the actual operating status of the emergency department.

[0058] An evaluation system is constructed from three levels: medical service providers (doctors), hardware resources (equipment / beds), and environmental conditions, forming a three-dimensional evaluation framework of "people-materials-environment". Doctor response data reflects the availability and professionalism of medical personnel, departmental resource data quantifies hardware support capabilities, and environmental data assesses the potential impact of the medical environment on treatment.

[0059] Emergency physician response data includes professional matching degree, physician experience index, cumulative fatigue coefficient, and workload index. The acquisition process is as follows:

[0060] Based on the most similar disease characteristics data, the corresponding disease label is determined. The disease label set of each emergency physician in the patient's pre-triage department is obtained. The proportion of all disease label sets containing the corresponding disease label is calculated and denoted as the professional matching degree B. zz Ensure that patients are assigned to specialized medical teams. For example, patients with myocardial infarction are matched with departments with a high proportion of cardiologists to improve diagnostic accuracy and avoid treatment delays caused by non-specialist doctors handling complex cases. For instance, neurologists are significantly more efficient at treating stroke than other departments.

[0061] The product of seniority and success rate of each emergency physician is obtained, and the sum and average are processed to obtain the physician experience index YjY. Seniority reflects basic clinical accumulation, while success rate reflects actual diagnostic and treatment ability. The product of the two avoids the one-sided assessment of "high seniority and low efficiency" or "low seniority and high risk". The average calculation is used to quickly assess the overall experience level of the department and provide data support for prioritizing the allocation of departments with high experience index for critically ill patients.

[0062] The average working hours of emergency room doctors are obtained. A comparison is made with the working hours-fatigue coefficient mapping set stored in the database to obtain the cumulative fatigue coefficient Pl (the working hours-fatigue coefficient mapping set includes multiple branches, each corresponding to a working hours and fatigue coefficient; the Euclidean distance between the average working hours of emergency room doctors and the working hours of each branch is calculated, and the working hours with the smallest difference from the average working hours of emergency room doctors are determined, along with their corresponding fatigue coefficient, denoted as the cumulative fatigue coefficient, which has no unit). Based on the working hours-fatigue coefficient mapping set, the nearest working hours branch is matched using Euclidean distance to quantify the doctor's fatigue level.

[0063] The system retrieves the disease deterioration probability warning threshold stored in the database, calculates the percentage of patients in the department whose disease deterioration probability exceeds the warning threshold, obtains the ratio of the number of patients waiting in the department to the critical number of patients waiting, and then performs a weighted summation of the percentage and the ratio to obtain the load index Fh. The percentage of patients with a high probability of disease deterioration reflects the urgency of treatment, while the ratio of the number of patients waiting reflects the degree of resource congestion. Combining the two avoids the crude assessment of "only looking at the number of patients and not the disease".

[0064] The first response factor (TVI1) for emergency physicians was determined based on emergency physician response data.

[0065]

[0066] Among them, λ1, λ2 and λ3 are all weighting factors, and yjY is the benchmark value of the doctor's experience index. The positive factors (professional matching degree, experience index) and negative factors (fatigue coefficient, load index) are offset to calculate and intuitively reflect the actual available capabilities of the doctor team.

[0067] Obtain static basic data and dynamic work data of emergency physicians. Determine the second response factor TVI2 of emergency physicians based on the static basic data and dynamic work data of emergency physicians. Perform a weighted summation of the first response factor TVI1 and the second response factor TVI2 of emergency physicians to obtain the emergency physician response factor TVI.

[0068] Static baseline data for emergency physicians includes seniority, success rate, and primary disease labels. Dynamic work data for emergency physicians includes patient volume per unit time, time spent treating a single case, team collaboration response speed, continuous working hours, medical record completeness score, and recent misdiagnosis / missed diagnosis records. The second response factor (TVI2) for emergency physicians is determined through the following steps:

[0069] After standardizing and integrating the static basic data and dynamic work data of emergency physicians, the original feature vector was obtained. Based on the deep belief network (DBN), deep feature extraction was performed on the original feature vector to obtain the deep feature vector, which includes clinical experience depth value, professional domain matching depth value, collaboration efficiency coefficient and work stability index.

[0070] The Deep Belief Network (DBN) consists of three Restricted Boltzmann Machines (RBMs) and one output layer. The first RBM layer has the same number of visible neurons as the original feature vector (including nine dimensions: seniority, success rate, set of treated disease labels, number of patients treated per unit time, time spent treating a single case, team collaboration response speed, continuous working hours, medical record completeness score, and recent misdiagnosis / missed diagnosis records). The hidden layer has 16 neurons. Through unsupervised pre-training of the first RBM layer, a 16-dimensional initial feature is output, which is an abstract expression of the original features after the first nonlinear transformation.

[0071] The second RBM has 16 visible layer neurons, matching the 16-dimensional initial feature output of the first RBM, and 8 hidden layer neurons. Through unsupervised pre-training of the second RBM, it outputs 8-dimensional intermediate features, which are a further abstract expression of the 16-dimensional initial features after the second nonlinear transformation.

[0072] The visible layer of the third RBM has 8 neurons, which matches the 8-dimensional intermediate feature output of the second RBM. The hidden layer has 4 neurons. Through unsupervised pre-training of the third RBM, a 4-dimensional deep feature is output. This feature is a high-order abstract expression of the 8-dimensional intermediate feature after nonlinear transformation of the third layer.

[0073] After pre-training, fine-tuning is performed using the backpropagation algorithm: the 4-dimensional deep features output by the third RBM layer are input to the output layer, which has 4 neurons. After fine-tuning, a 4-dimensional final deep feature vector is output, which includes clinical experience depth value, professional domain matching depth value, collaboration efficiency coefficient, and work stability index.

[0074] Acquire physiological index data of emergency physicians, obtain an initial observation sequence arranged in time based on deep feature vectors and physiological index data of emergency physicians, and obtain a dynamic observation sequence with attention weights by weighting the initial observation sequence.

[0075] The 4-dimensional deep feature vector output by the DBN (including clinical experience depth value, professional domain matching depth value, collaboration efficiency coefficient, and work stability index) is fused with real-time collected physician physiological index data, which includes heart rate per minute, average reaction speed (unit: seconds), respiratory rate (breaths / minute), and skin conductance (microSiemens). Data collection and fusion are performed at set time intervals (e.g., every 15 minutes) to form an initial observation sequence arranged chronologically. The observation value at each time point is a 7-dimensional vector (4-dimensional deep features + 3-dimensional real-time physiological index).

[0076] Based on the timestamp information of the initial observation sequence, a time decay factor is set (e.g., with an 8-hour baseline working time, the time decay factor increases by 0.15 for every additional hour beyond the baseline). The time decay factor is multiplied by the base weights to obtain the attention weights at each time point. The 7-dimensional observation value at each time point in the initial observation sequence is then multiplied dimension by dimension by the corresponding attention weight to obtain the weighted 7-dimensional observation value. The weighted observation values ​​at all time points are arranged in their original temporal order to form a dynamic observation sequence with attention weights.

[0077] The Hidden Markov Model is trained based on dynamic observation sequences with attention weights to obtain the probability distribution of hidden states, which include highly focused states, mild fatigue states, cognitive overload states, and collaborative dependency states.

[0078] The Hidden States Model (HMM) is set to four states: "Highly Attentive State," "Mild Fatigue State," "Cognitive Overload State," and "Collaboration-Dependent State." Weighted observation sequences are used as training data and input into the HMM for parameter training, resulting in a state transition probability matrix (4×4 matrix, elements representing the probability of transitioning from one state to another) and an observation probability matrix (4×7 matrix, elements representing the probability of observing a specific 7-dimensional value in a given state). Newly collected real-time doctor data is used to construct an initial observation sequence using the method described in the first step. After calculating attention weights in the second step and generating a weighted observation sequence in the third step, the sequence is input into the trained HMM. The forward-backward algorithm is used to calculate the probability that the doctor is currently in one of the four states: "Highly Attentive State," "Mild Fatigue State," "Cognitive Overload State," or "Collaboration-Dependent State," and the probability distribution of each state is output.

[0079] The weighting factors of deep feature vectors are determined based on reinforcement learning and the probability distribution of hidden states. The deep feature vectors are then weighted and summed to obtain the emergency physician's second response factor, TVI2.

[0080] Obtain the 4-dimensional deep feature values ​​of the DBN output, including the clinical experience depth value (denoted as F1), the professional domain matching depth value (denoted as F2), the collaboration efficacy coefficient (denoted as F3), and the job stability index (denoted as F4); obtain the probability distribution of the 4 hidden states of the HMM output, including the probability of highly focused state (denoted as P1), the probability of mild fatigue state (denoted as P2), the probability of cognitive overload state (denoted as P3), and the probability of collaboration dependence state (denoted as P4).

[0081] Set the reinforcement learning (Q-learning) parameters. Define the Q-learning state as the probability distribution of the HMM output ((P1, P2, P3, P4)); define the action space as the weight adjustment amount corresponding to the four deep feature values, with an adjustment step size of 0.05, and the value range of each weight is 0.1-0.5; define the reward function R as the negative value of the absolute value of the deviation between the doctor's response ability coefficient and the actual treatment efficiency (the smaller the deviation, the larger the reward).

[0082] The weights are dynamically adjusted using Q-learning. The current state is input into the Q-learning model, and a weight adjustment action is selected based on an ε-greedy policy (ε = 0.1) to obtain the adjusted dynamic weights. The Q-value is updated through feedback via a reward function, and the final dynamic weights are determined after 50 iterations. The deep feature values ​​of the DBN are multiplied by the final dynamic weights obtained in step three, and then summed to obtain the emergency physician's second response factor, TVI2.

[0083] Departmental resource data includes the percentage of available equipment (Ks). b and bed turnover coefficient Cw z Determine the departmental resource coupling factor based on departmental resource data:

[0084]

[0085] Where, N cl To determine the number of beds ready for disinfection, N zg H represents the total number of beds. pb The nursing care matching factor, which is the ratio of the number of nurses to the number of patients, Gr jq The infection rate is the rate over the past month; the proportion of available equipment and the bed turnover coefficient reflect the resource utilization rate, the infection rate is adjusted inversely to ensure medical safety, and the nursing staffing ratio quantifies the degree of manpower shortage to avoid the bottleneck of "sufficient equipment but insufficient nursing".

[0086] Departmental environmental data includes personnel density per unit area (ρ). sj Electromagnetic interference intensity Dc sj and aerosol concentration qr sj Based on departmental environmental data, the environmental impact factor HjY was determined.

[0087]

[0088] ρ jz Dc is the threshold value for personnel density per unit area. jz qr is the threshold value for electromagnetic interference intensity. jz This represents the threshold for aerosol concentration. When the density of personnel, electromagnetic interference, or aerosol concentration exceeds the standard, the exponential function rapidly lowers environmental factors, indicating a risk of cross-infection or equipment malfunction.

[0089] The final triage strategy is determined based on the probability of disease deterioration and the emergency treatment capacity coefficient. The final triage strategy includes treatment in the pre-triage department of the hospital and treatment by transfer to another hospital: if the probability of disease deterioration is greater than the disease deterioration threshold stored in the database and the emergency treatment capacity coefficient is less than the emergency treatment capacity threshold stored in the database, then the final triage strategy is determined to be treatment by transfer to another hospital, and the selected hospital to be transferred is determined based on multi-objective optimization; otherwise, the final triage strategy is determined to be treatment in the pre-triage department of the hospital.

[0090] A multi-objective function is constructed and constraints are set. Triples consisting of [candidate hospital identifier, transport route number, and departure timestamp] are used as basic gene units to form an initial chromosome population. Key transport decision-making elements (hospital selection, route planning, and departure time) are encoded as triples, serving as the basic operational units of the genetic algorithm to achieve the digital expression of decision variables. The gene unit format is adapted to the crossover and mutation operations of the genetic algorithm, facilitating the search for the optimal solution through evolutionary computation.

[0091] The initial chromosome population is screened based on constraints to obtain a new initial chromosome population. This avoids the inclusion of infeasible solutions (such as unqualified hospitals or completely mismatched treatment plans), laying the foundation for subsequent optimization. Pareto front analysis and approximation of ideal solutions are then performed on the screened initial chromosome population to obtain initial candidate basic gene units. Genetic operations are then performed on the initial chromosome population, specifically: a two-point crossover operation is performed on the candidate hospitals, a single-point mutation operation is performed on the transport path, and a Gaussian distribution-based mutation operation is performed on the departure time, forming new basic gene units and constituting a new chromosome population.

[0092] The new chromosome population is screened again based on constraints, followed by Pareto front analysis and ranking of near-ideal solutions to obtain candidate basic gene units after a second iteration. After several iterations, all candidate basic gene units are summarized. A fitness function is constructed based on a multi-objective function to determine the candidate basic gene unit corresponding to the maximum fitness function value, thus obtaining the selected hospitals for transfer. The fitness function F is formulated as follows:

[0093]

[0094] The multi-objective function includes: the emergency treatment capacity coefficient CZ of the corresponding department of the optional transfer hospital. nl The cosine similarity σ(ZL1, ZL2) between the current treatment plan ZL1 and the planned treatment plan ZL1 at the alternative transfer hospital is used. The treatment capacity coefficient integrates data from doctors, resources, and the environment (as described above) to ensure transfer to a hospital with stronger comprehensive capabilities, avoiding the risk of insufficient post-transfer treatment capacity. The cosine similarity of the treatment plan quantifies the compatibility between the current plan and the transfer hospital's plan (such as matching medication regimens and surgical plans), reducing the risk of treatment interruption.

[0095] The constraints include: CZ nl CZ nl ′ min ;σ(ZL1, ZL2)>σ(ZL1, ZL2) min CZ nl ′ min To minimize the disposal capacity coefficient, σ(ZL1, ZL2) min The goal is to minimize similarity. Hospitals with poor treatment capabilities (e.g., lack of corresponding specialists or equipment) or significantly different treatment plans (e.g., conflict between the current thrombolysis protocol and the surgical plan of the transport hospital) should be avoided to reduce the error rate in transport decisions.

[0096] The Pareto front analysis and approximation ideal solution ranking process includes the following steps: For any two basic gene units A and B, if the emergency treatment capability coefficient and cosine similarity of treatment plan of A are not less than the emergency treatment capability coefficient and cosine similarity of treatment plan of B, and at least one of the emergency treatment capability coefficient and cosine similarity of treatment plan of A is greater than the emergency treatment capability coefficient and cosine similarity of treatment plan of B, then A is said to dominate B; basic gene units not dominated by other basic gene units are grouped into a first-level non-dominated set, which serves as the first layer of the Pareto front and as the optimal solution set obtained from the Pareto front analysis; the optimal solution set is then subjected to approximation ideal solution ranking process.

[0097] It should be noted that the emergency treatment capability coefficient and cosine similarity of treatment plan A are both no less than the emergency treatment capability coefficient and cosine similarity of treatment plan B, indicating that the emergency treatment capability coefficient of A is no less than the emergency treatment capability coefficient of B, and the cosine similarity of treatment plan A is no less than the cosine similarity of treatment plan B; and at least one of the emergency treatment capability coefficient and cosine similarity of treatment plan A is greater than the corresponding emergency treatment capability coefficient and cosine similarity of treatment plan B.

[0098] The dominance relationship is defined by "non-inferiority in both objectives and superiority in at least one objective," meaning that A dominates B when A's treatment capacity coefficient is ≥ B and its treatment plan similarity is ≥ B, and at least one of these is strictly superior to B. This avoids the one-sidedness of a single-objective optimal solution (such as selecting only the hospital with the highest treatment capacity but completely mismatched treatment plans), ensuring that the selected solution has advantages in both key objectives. By retaining solutions with advantages in different dimensions through non-dominance relationships, multiple choices are provided for subsequent decision-making.

[0099] The optimized solution set is sorted to approximate the ideal solution, including the following steps: Obtain the positive ideal solution of the multi-objective function stored in the database; calculate the Euclidean distance between the actual solution of the multi-objective function and the positive ideal solution for each basic gene unit in the optimized solution set, and record the basic gene unit corresponding to the smallest Euclidean distance value as the candidate basic gene unit. Select the gene unit with the smallest Euclidean distance as the candidate solution, i.e., the transport scheme closest to the positive ideal solution. The positive ideal solution can be automatically adjusted with database updates to ensure that the optimization logic always adapts to the latest distribution of medical resources.

[0100] A smart emergency triage system based on big data, such as Figure 2 As shown, the system includes a pre-screening module for acquiring patient pre-screening data and determining the pre-triage department and the probability of disease deterioration based on the patient pre-screening data. The patient pre-screening data includes real-time physiological monitoring data and patient description information. A treatment capacity assessment module is used to evaluate the treatment capacity of the pre-triage department and determine the emergency treatment capacity coefficient. A triage module is used to determine the final triage strategy based on the probability of disease deterioration and the emergency treatment capacity coefficient. The final triage strategy includes treatment in the pre-triage department of the hospital and transfer to another hospital: if the probability of disease deterioration is greater than the disease deterioration threshold stored in the database, and the emergency treatment capacity coefficient is less than the emergency treatment capacity threshold stored in the database, then the final triage strategy is determined to be transfer to another hospital, and the selected transfer hospital is determined based on multi-objective optimization; otherwise, the final triage strategy is determined to be treatment in the pre-triage department of the hospital.

[0101] An electronic device includes: a processor; and a memory storing computer program instructions, which, when executed by the processor, cause the processor to perform the big data-based intelligent emergency triage method as described above.

[0102] A computer-readable storage medium for storing a program that, when executed by a processor, implements the big data-based intelligent emergency triage method described above.

[0103] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0104] This invention is described with reference to flowchart illustrations and / or block diagrams of systems, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0105] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0106] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0107] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0108] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A smart emergency triage method based on big data, characterized in that, Includes the following steps: Acquire patient pre-examination data, and determine the pre-triage department and the probability of disease deterioration based on the patient pre-examination data. The patient pre-examination data includes real-time physiological monitoring data and patient description information. Real-time physiological monitoring data f1 includes circulatory system parameters, respiratory system parameters, nervous system parameters and characteristic monitoring parameters; patient description information c1 includes pain location, pain intensity, pain duration and past medical history. Assess the handling capacity of pre-triage departments and determine the emergency handling capacity coefficient; The final triage strategy is determined based on the probability of disease deterioration and the emergency department's capacity to handle the situation. The final triage strategy includes treatment within the pre-triage department of this hospital and treatment upon transfer to another hospital. If the probability of the condition worsening is greater than the threshold for the condition worsening stored in the database, and the emergency treatment capacity coefficient is less than the threshold for the emergency treatment capacity stored in the database, then the final triage strategy is determined to be transfer to another hospital for treatment, and the selected hospital to be transferred is determined based on multi-objective optimization. Otherwise, the final triage strategy will be to have the patient treated by the pre-triage department of this hospital.

2. The intelligent emergency triage method based on big data according to claim 1, characterized in that, Determining the pre-triage department and the probability of disease progression based on patient pre-screening data includes the following steps: Obtain the dataset of disease characteristics of each emergency department stored in the database, including the disease characteristic data, which includes physiological parameter characteristic data f2 and parameter description information c2. The patient pre-screening data is matched with the characteristic data of each major disease in the characteristic dataset of each emergency department to obtain the disease similarity matching value JB. xc : JB xc =σ(f1, f2) + σ(c1, c2), where σ(·) is the cosine similarity function; The main disease feature data corresponding to the largest disease similarity matching value is identified and denoted as the most similar main disease feature data. The corresponding emergency department is determined as the patient's pre-triage department through the main disease feature dataset to which the most similar main disease feature data belongs. Retrieve the diagnostic data-deterioration probability mapping set corresponding to the most similar disease characteristics data from the database. This set includes several diagnostic data-deterioration probability branches, and each diagnostic data includes several historical physiological monitoring data fb. i ′; According to the set time intervals, physiological monitoring sample data cb corresponding to several time points are obtained from the real-time physiological monitoring data. i The similarity analysis of physiological monitoring sample data and historical physiological monitoring data was performed to obtain the monitoring data similarity comparison value JC. xs JC xs =∑σ(cb′) i ,fb′ i ); Determine the diagnostic data corresponding to the largest similarity comparison value of the monitoring data, and determine the predicted probability of disease deterioration; Dynamic physiological time-series subsequences are obtained based on real-time physiological monitoring data, and static risk factor subsequences are also obtained. The dynamic physiological time-series subsequences and static risk factor subsequences are concatenated to obtain a fused feature sequence. Using a Transformer-Bayesian model to determine the auxiliary probability of disease progression; The probability of disease deterioration is obtained by weighted summation of the predicted probability and the auxiliary probability of disease deterioration.

3. The intelligent emergency triage method based on big data according to claim 2, characterized in that, The Transformer-Bayesian model includes a Transformer module and a Bayesian inference module. The Transformer-Bayesian model is used to determine the auxiliary probability of disease progression, including the following steps: The fused feature sequence is input into a Transformer module that includes multiple encoders and an output layer. Each encoder layer includes a multi-head self-attention mechanism and a feedforward neural network. After processing the fused feature sequence, the encoders in the Transformer module input the processing results into the output layer. The output layer uses the Sigmoid activation function to generate auxiliary probabilities of disease deterioration at multiple future time points. The Bayesian inference module samples using the Monte Carlo dropout method to obtain an auxiliary probability distribution sample of disease deterioration at each future time point. It calculates the mean of the auxiliary probability distribution sample as a reference value for the deterioration probability at that time point, and uses the deterioration probability reference value to calculate the sample standard deviation as an uncertainty index. Then, it determines the probability distribution interval for each time point, which is the deterioration probability reference value ± 2 times the sample standard deviation. The confidence factor corresponding to the probability distribution interval at each time point is determined based on the probability distribution interval-confidence factor mapping set stored in the database, and is used as a weighting factor for the auxiliary probability of disease deterioration. The weighting factor for the auxiliary probability of disease deterioration and the weighting factor for the predicted probability of disease deterioration are set to 1, and the probability of disease deterioration is calculated.

4. The intelligent emergency triage method based on big data according to claim 2, characterized in that, Determining the emergency treatment capacity coefficient includes the following steps: Obtain emergency department response capacity assessment data, including emergency physician response data, department resource data, and department environment data; The Emergency Physician Response Factor (TVI) was determined based on emergency physician response data. The departmental resource coupling factor KsO is determined based on departmental resource data; Environmental impact factor HjY was determined based on departmental environmental data. The emergency treatment capacity coefficient CZ is determined by combining the emergency physician response factor TVI, the departmental resource coupling factor KsO, the environmental impact factor HjY, and the probability of disease deterioration Pe. nl : Where α1, α2, β1 and β2 are all weighting factors, and e is a natural constant.

5. The intelligent emergency triage method based on big data according to claim 4, characterized in that, Emergency physician response data includes professional matching degree, physician experience index, cumulative fatigue coefficient, and workload index. The acquisition process is as follows: Based on the most similar disease characteristics data, the corresponding disease label is determined. The disease label set of each emergency physician in the patient's pre-triage department is obtained. The proportion of all disease label sets containing the corresponding disease label is calculated and denoted as the professional matching degree B. zz ; The product of each emergency physician's years of experience and success rate is summed and averaged to obtain the physician experience index YjY; The average working hours of emergency room doctors are obtained, and the cumulative fatigue coefficient Pl is obtained by comparing it with the working hours-fatigue coefficient mapping set stored in the database. Obtain the warning threshold for the probability of disease deterioration stored in the database, count the proportion of patients in the department whose probability of disease deterioration is greater than the warning threshold, obtain the ratio of the number of patients waiting in the department to the number of patients at the critical waiting point, and sum the proportion and the ratio by weight to obtain the load index Fh. The first response factor (TVI1) for emergency physicians was determined based on emergency physician response data. Wherein, λ1, λ2 and λ3 are all weighting factors, and yjY is the benchmark value of the doctor's experience index; Obtain static basic data and dynamic work data of emergency physicians. Determine the second response factor TVI2 of emergency physicians based on the static basic data and dynamic work data of emergency physicians. Perform a weighted summation of the first response factor TVI1 and the second response factor TVI2 of emergency physicians to obtain the emergency physician response factor TVI. Departmental resource data includes the percentage of available equipment (Ks). b and bed turnover coefficient Cw z Determine the departmental resource coupling factor based on departmental resource data: Where, N cl To determine the number of beds ready for disinfection, N zg H represents the total number of beds. pb The nursing care matching factor, which is the ratio of the number of nurses to the number of patients, Gr jq This represents the infection rate over the past month. Departmental environmental data includes personnel density per unit area (ρ). sj Electromagnetic interference intensity Dc sj and aerosol concentration qr sj Based on departmental environmental data, the environmental impact factor HjY was determined. Where, ρ jz Dc is the threshold value for personnel density per unit area. jz qr is the threshold value for electromagnetic interference intensity. jz This is the threshold value for aerosol concentration.

6. The intelligent emergency triage method based on big data according to claim 5, characterized in that, Static baseline data for emergency physicians includes seniority, success rate, and primary disease labels. Dynamic work data for emergency physicians includes patient volume per unit time, time spent treating a single case, team collaboration response speed, continuous working hours, medical record completeness score, and recent misdiagnosis / missed diagnosis records. The second response factor (TVI2) for emergency physicians is determined through the following steps: The static basic data and dynamic work data of emergency physicians are standardized and integrated into the original feature vector; Deep feature extraction is performed on the original feature vector based on the deep belief network (DBN) to obtain deep feature vectors, including clinical experience depth value, professional domain matching depth value, collaboration efficiency coefficient and job stability index. Acquire physiological index data of emergency physicians, obtain an initial observation sequence arranged in time based on deep feature vectors and physiological index data of emergency physicians, and obtain a dynamic observation sequence with attention weights by weighting the initial observation sequence. The Hidden Markov Model is trained based on dynamic observation sequences with attention weights to obtain the probability distribution of hidden states, which include highly focused states, mild fatigue states, cognitive overload states, and collaborative dependency states. The weighting factors of deep feature vectors are determined based on reinforcement learning and the probability distribution of hidden states. The deep feature vectors are then weighted and summed to obtain the emergency physician's second response factor, TVI2.

7. The intelligent emergency triage method based on big data according to claim 4, characterized in that, The selection of hospitals for transfer is determined based on multi-objective optimization, including the following steps: Construct a multi-objective function and set constraints; The initial chromosome population is formed by using a triple consisting of [candidate hospital identifier, transfer route number, departure timestamp] as the basic gene unit; The initial chromosome population is screened based on constraints to obtain the initial chromosome population after screening. Pareto front analysis and approximation of ideal solution sorting are then performed on the initial chromosome population after screening to obtain the initial candidate basic gene units. Genetic manipulation was performed on the initial chromosome population, specifically: a two-point crossover operation was performed on the selected hospitals, a single-point mutation operation was performed on the transport route, and a Gaussian distribution-based mutation operation was performed on the departure time to form new basic gene units and constitute a new chromosome population. The new chromosome population was screened again based on the constraints, followed by Pareto front analysis and sorting to approximate the ideal solution, to obtain the candidate basic gene units after the second iteration. After several iterations, all candidate basic gene units are summarized. A fitness function is constructed based on a multi-objective function. The candidate basic gene units corresponding to the maximum fitness function value are determined, and the selected hospitals for transfer are obtained. The formula for the fitness function F is:

8. The intelligent emergency triage method based on big data according to claim 7, characterized in that, Pareto front analysis and sorting of solutions approximating the ideal solution are performed, including the following steps: For any two basic gene units A and B, if the emergency treatment capability coefficient and the cosine similarity of the treatment plan of A are both not less than the emergency treatment capability coefficient and the cosine similarity of the treatment plan of B, and at least one of the emergency treatment capability coefficient and the cosine similarity of the treatment plan of A is greater than the emergency treatment capability coefficient and the cosine similarity of the treatment plan of B, then A is said to dominate B. The basic gene units that are not dominated by other basic gene units are grouped into the first-level non-dominated set, which serves as the first layer of the Pareto front and as the set of optimal solutions obtained from Pareto front analysis. The optimized solution set is sorted to approximate the ideal solution: Obtain the positive ideal solution of the multi-objective function stored in the database; The Euclidean distance between the actual solution and the positive ideal solution of the multi-objective function of each basic gene unit in the optimized solution set is calculated, and the basic gene unit corresponding to the smallest Euclidean distance value is recorded as the candidate basic gene unit.

9. The intelligent emergency triage method based on big data according to claim 8, characterized in that, Multi-objective functions include: Emergency treatment capacity coefficient CZ of the corresponding department of the optional transfer hospital nl ′; The cosine similarity σ(ZL1, ZL2) between the current treatment regimen ZL1 and the alternative transfer hospital's planned treatment regimen ZL1; The constraints include: CZ nl ′>CZ nl ′ min ;σ(ZL1,ZL2)>σ(ZL1,ZL2) min ; Among them, CZ nl ′ min To minimize the disposal capacity coefficient, σ(ZL1, ZL2) min This represents the minimum similarity value.

10. A big data-based intelligent emergency triage system, used in accordance with any one of claims 1-9, characterized in that, include: The pre-screening module is used to acquire patient pre-screening data and determine the pre-triage department and the probability of disease deterioration based on the patient pre-screening data. The patient pre-screening data includes real-time physiological monitoring data and patient description information. Real-time physiological monitoring data f1 includes circulatory system parameters, respiratory system parameters, nervous system parameters and characteristic monitoring parameters; patient description information c1 includes pain location, pain intensity, pain duration and past medical history. The emergency response capacity assessment is used to evaluate the response capacity of the pre-triage department and determine the emergency response capacity coefficient. The triage module is used to determine the final triage strategy based on the probability of disease deterioration and the emergency treatment capacity coefficient. The final triage strategy includes treatment in the pre-triage department of this hospital and treatment by transfer to another hospital. If the probability of the condition worsening is greater than the threshold for the condition worsening stored in the database, and the emergency treatment capacity coefficient is less than the threshold for the emergency treatment capacity stored in the database, then the final triage strategy is determined to be transfer to another hospital for treatment, and the selected hospital to be transferred is determined based on multi-objective optimization. Otherwise, the final triage strategy will be to have the patient treated by the pre-triage department of this hospital.

Images