Intensive care cross-institution collaborative data processing method

Abstract

The invention relates to an intensive care cross-institution collaborative data processing method, which comprises the following steps of: defining a hierarchical data model based on the intensive care field, and modeling various types of information including vital signs, clinical diagnosis and laboratory detection results of a patient according to a time sequence; a self-adaptive standardized interface is designed, and data formats from different hospitals or devices are identified and converted; real-time dynamic tags are introduced for patient data through a tagging technology; various health data, including vital signs, medical history and treatment feedback, of a critical patient are converted into a multi-dimensional health track, and sequential analysis is used for conducting continuous modeling on the condition of the patient; a deep learning and reinforcement learning method is combined to establish a disease evolution deduction engine; the disease development is predicted according to historical data, and the influence of different intervention measures on the disease is simulated to intelligently recommend a treatment strategy; and according to the dynamic deduction result, generating a plurality of treatment schemes, and dynamically adjusting the intervention strategy according to the real-time response of the patient.

Description

Technical Field

[0001] The present invention relates to a collaborative data processing method, specifically to a cross-institutional collaborative data processing method for critical care. Background Art

[0002] Currently, although the cross-institutional collaborative data processing method for critical care has been gradually widely applied and has improved the nursing efficiency and resource utilization rate to a certain extent, there are still many deficiencies and drawbacks. These problems not only limit the actual effect of cross-institutional collaboration, but also affect the treatment quality of patients and the overall operation efficiency of the medical system to a certain extent.

[0003] First of all, most of the existing cross-institutional collaborative data processing methods focus on information exchange and sharing, lacking in-depth analysis and intelligent processing of data. In traditional medical information sharing platforms, the collaboration between hospitals is mostly limited to information-level sharing and transmission, often without sufficient data integration and analysis. The data formats of different hospitals, departments, and devices are inconsistent, and there are no standardized and normalized interfaces, making it difficult to integrate cross-institutional data. Even if different hospitals can exchange data, due to the lack of a unified data standard, information often becomes distorted or misunderstood during transmission, affecting the accuracy and timeliness of the data. In addition, many existing systems do not take into account the dynamics and complexity of data. The health status of patients is constantly changing, and traditional static data processing methods cannot effectively cope with the complex fluctuations of patients' conditions and the changing treatment needs. Secondly, the allocation and sharing of resources in cross-institutional collaboration are also a significant problem. Currently, cross-institutional collaboration mostly relies on manual scheduling and decision-making, lacking an efficient and real-time resource scheduling system. In the traditional medical environment, the resource allocation between hospitals often follows their respective management systems and rules, with low resource scheduling efficiency and the inability to respond in real time to the changes and urgent needs of patients' conditions, resulting in overuse of resources in some hospitals while some other hospitals may face resource shortages. For example, a specific treatment device needed by a patient may be idle in a certain hospital, but due to insufficient information sharing between hospitals, the patient cannot receive timely treatment, affecting the treatment effect. This inefficiency in resource scheduling directly affects the treatment quality, especially in critical care, where the consequences of delayed treatment are often catastrophic.

[0004] In addition, most existing cross-institutional collaboration methods ignore the workload and availability issues of doctors and nursing staff. Although some systems can provide doctor scheduling and resource allocation, they lack personalized and dynamic scheduling for doctors. For example, some systems fail to accurately evaluate the professional expertise and current load of each doctor, resulting in patients being assigned to non-specialist doctors during the treatment process, or being unable to find an available doctor in a timely manner when treatment is most needed. This not only affects the treatment efficiency, but also increases the waiting time of patients, and may even cause errors during the diagnosis and treatment process. Moreover, existing decision-making systems for cross-institutional collaboration often rely on traditional rule engines or expert systems, lacking the application of deep learning and artificial intelligence technologies. Traditional decision-making methods are mainly based on historical experience or expert judgment. Although these methods can provide certain guidance, they often cannot provide sufficient accuracy and flexibility when faced with complex and changing critical conditions. Especially when multiple medical institutions and multiple departments participate in the treatment together, it is difficult for expert judgment to integrate information from different institutions in real time, and it is difficult for the system to automatically adjust strategies, lacking the ability to respond to emergencies. For example, when a patient's condition suddenly changes during the transfer process, the existing system may not be able to adjust the treatment plan in a timely manner, and may even cause treatment delays due to information lag or inability to identify key data. In addition, existing cross-institutional collaboration methods also face the problem of patient privacy protection. Summary of the Invention

[0005] The purpose of the present invention is to provide a cross-institutional collaborative data processing method for critical care, so as to solve some of the drawbacks and deficiencies pointed out in the background technology.

[0006] The technical solutions adopted by the present invention to solve its above technical problems include the following steps:

[0007] S1. Cross-institutional data standardization and dynamic adaptation based on critical care characteristics:

[0008] S1.1. Define a hierarchical data model based on the critical care field, and model various types of patient information including vital signs, clinical diagnoses, and laboratory test results in time series.

[0009] S1.2. Design an adaptive standardization interface to identify and convert data formats from different hospitals or devices.

[0010] S1.3. Introduce real-time dynamic tags for patient data through tagging technology.

[0011] S2. Dynamic condition deduction and intervention strategies based on the patient's health trajectory:

[0012] S2.1. Transform various health data of critically ill patients, including vital signs, medical history, and treatment feedback, into multi-dimensional health trajectories, and use time series analysis to continuously model the patient's condition;

[0013] S2.2. Combine deep learning and reinforcement learning methods to establish a deduction engine for the evolution of the condition; predict the development of the condition based on historical data, and intelligently recommend treatment strategies by simulating the impact of different intervention measures on the condition;

[0014] S2.3. Generate multiple treatment plans according to the dynamic deduction results, and dynamically adjust the intervention strategy based on the patient's real-time response;

[0015] S3. Real-time adjustment of cross-institutional treatment paths based on collaborative optimization:

[0016] S3.1. Design an algorithm model based on multi-objective optimization to calculate the health status, treatment plan, doctor availability, hospital resources of critically ill patients, and the collaboration ability between institutions, and generate the optimal treatment path;

[0017] S3.2. Use real-time data including changes in the patient's condition and resource occupancy to dynamically adjust the treatment path;

[0018] S4. Intelligent decision support system and cross-institutional collaborative decision-making framework:

[0019] S4.1. Establish a decision support model based on an expert knowledge base and machine learning, integrate the patient's condition data, historical treatment records, and the treatment capabilities of medical institutions to generate personalized treatment recommendations;

[0020] S4.2. In a cross-institutional collaborative environment, by recommending the best treatment plan, and each medical institution makes real-time cross-institutional decision collaboration based on the patient data input, adjustment and feedback of the treatment plan;

[0021] S4.3. Build a real-time collaborative decision-making platform participated by multiple parties, allowing doctors and nursing staff from different institutions to share patient information, treatment progress and risk assessment in real time to make collaborative decisions.

[0022] Furthermore, the dynamic condition deduction and intervention strategy method based on the patient's health trajectory includes:

[0023] Through multi-dimensional data fusion, integrate multiple data dimensions from vital signs, patient medical history, and treatment feedback to form a complete health trajectory; use time series analysis technology for modeling, and use long short-term memory networks to capture the time series information in the data; the calculation formula is:

[0024]

[0025] Where Xt is the health state vector at time t, representing health data composed of vital signs Hτ, medical history B(τ), and treatment feedback (F(τ)); the function ψ(·) is a multi-dimensional data fusion function, aiming to combine heterogeneous data including continuous physiological signals and discrete medical history information; the symbol τ represents the time variable, and dτ represents the integral operation over time.

[0026] Furthermore, the method for dynamic disease progression deduction and intervention strategy based on the patient's health trajectory includes:

[0027] Each treatment decision in reinforcement learning obtains an immediate reward including the degree of improvement of the patient's state, and optimizes the treatment strategy through long-term feedback; an expected reward formula is introduced for strategy optimization to maximize the treatment effect;

[0028] J(θ) = E π(θ) [∫ 0 T γ t r t (θ, X t ) dt]

[0029] where J(θ) is the expected reward of the treatment strategy, representing the overall reward value calculated by time integration under the current strategy; π(θ) is the treatment strategy, the optimal strategy trained through reinforcement learning; r t (θ, X t ) is the immediate reward function, measuring the reward value corresponding to the patient's health state X t when the strategy θ is taken at time t; γ is the discount factor, used to balance the influence of current and future rewards; T is the time endpoint, representing the time span of the deduction process.

[0030] Furthermore, the method for dynamic disease progression deduction and intervention strategy based on the patient's health trajectory includes:

[0031] By combining reinforcement learning and simulation intervention, according to real-time patient data and historical records, simulate the effects of multiple treatment measures, and evaluate the long-term impact of each treatment plan; based on the simulation results, dynamically adjust the treatment strategy to provide a personalized treatment path;

[0032]

[0033] where, is the optimization degree of the treatment plan, representing the optimization evaluation of all treatment strategies by the system at time t; ρ k is the weight of the intervention measure T k , representing the contribution degree of each treatment measure to the optimization goal; η k (Xt , T k ) is the effect evaluation function, representing the treatment effect brought about by taking the intervention measure T t under the current state X k ; e -λt is the attenuation factor, controlling the attenuation degree of the treatment effect over time; λ is the attenuation constant, used to adjust the attenuation rate in terms of time.

[0034] Furthermore, the real-time adjustment method for the cross-institutional treatment path based on collaborative optimization:

[0035] Introduce a multi-objective optimization algorithm model, integrate the patient's health status, doctor availability, hospital resource occupancy, and cross-hospital collaboration ability into a unified optimization framework, and calculate the optimal treatment path; the objective function comprehensively considers multiple objectives and constraints through weighted summation, and determines the global optimal treatment path according to the weights of each objective; the optimization function is as follows:

[0036] where x ∈ R n

[0037] where, F(x) represents the overall objective function, which is the comprehensive treatment path index to be optimized, reflecting the weighted comprehensive effect of multiple objectives; f i (x) is the i-th objective function, representing the optimization objective related to the i-th objective; h j (x) is the j-th constraint function, representing the function related to resource allocation, hospital capacity, and doctor availability resource constraints; α i and γ j are the weight coefficients of the objective function and the constraint function respectively; x is the vector of optimization variables, representing specific treatment path selection, resource allocation, doctor, and hospital scheduling decision factors.

[0038] Furthermore, the real-time adjustment method for the cross-institutional treatment path based on collaborative optimization:

[0039] Introduce a real-time data feedback mechanism, and combine it with a dynamic control model to adjust the treatment path in real time according to the real-time data of the condition, resource occupancy, and doctor availability; the algorithm performs dynamic adjustment through the following differential equation model:

[0040]

[0041] where:

[0042] It represents the rate of change of the patient's health status, reflecting the change of the patient's condition at time t; A(t) is a dynamic matrix describing the impact of the treatment plan on the patient's health status, representing the effect of different treatment measures over time; x(t) is the health status vector of the patient, representing the specific manifestation at time t; B(t) is a matrix describing the impact of external factors such as resource utilization and doctor availability on the patient's health status; u(t) is a control variable, representing decision variables such as treatment plans, resource scheduling, and doctor arrangements; C(t) is a matrix representing the impact of collaboration effectiveness, determining the real-time impact of information sharing and resource collaboration between hospitals; d(t) is a variable of external disturbance or hospital collaboration, reflecting the real-time data of collaboration between different hospitals and the impact on the patient's treatment path.

[0043] Furthermore, the real-time adjustment method of the cross-institutional treatment path based on collaborative optimization:

[0044] Incorporate factors such as resource sharing, information flow, and doctor availability between hospitals into the optimization model for cross-institutional resource scheduling optimization; the collaboration effectiveness between hospitals is modeled through the following time-dependent collaboration effectiveness function:

[0045]

[0046] Where:

[0047] C collab (R) represents the collaboration effectiveness between hospitals, reflecting the overall effect of cross-hospital resource sharing and cooperation; β k is the weight coefficient of each collaboration dimension, representing the impact of different collaboration factors on the overall collaboration effectiveness; g k (R(t)) is the effectiveness function of the kth collaboration dimension, representing the effectiveness of a specific collaboration at time t; R(t) is the configuration vector of cross-institutional resources, reflecting the resource utilization of different hospitals at time t; τ 1 and τ 2 are the time intervals of collaboration, representing the duration of collaboration. Brief Description of the Drawings

[0048] Figure 1 It is a flowchart of the cross-institutional collaborative data processing method for intensive care of the present invention.

[0049] Figure 2 It is a flowchart of the dynamic disease deduction and intervention strategy method based on the patient's health trajectory of the present invention.

[0050] Figure 3 It is a flowchart of the real-time adjustment method of the cross-institutional treatment path based on collaborative optimization of the present invention. Detailed Embodiments

[0051] The following will make a detailed description of the specific embodiments of the present invention in conjunction with the drawings.

[0052] The cross-institutional collaborative data processing method for critical care first proposed a key concept, namely, cross-institutional data standardization and dynamic adaptation based on the characteristics of critical care. Specifically, in the field of critical care, patients' health data does not only come from a single hospital or device, but is scattered in different medical institutions and equipment, and these data formats may be incompatible due to the lack of unified standards between institutions. Therefore, cross-institutional collaboration requires a unified processing framework. In order to ensure that these scattered and formatted data can be effectively shared and analyzed in linkage, data standardization and dynamic adaptation must be performed first. First, in view of the particularity of critical care, a hierarchical data model is designed, which can be modeled based on multi-dimensional information such as patient health status, clinical diagnosis, vital signs, and laboratory test results. This modeling method can not only reflect the longitudinal association of data (i.e., the changes in the patient's condition on the timeline), but also show the horizontal multi-dimensional characteristics of data. Each data dimension (such as vital signs, diagnosis, test results, etc.) has a clear classification and weight, which facilitates subsequent dynamic analysis and multi-objective optimization. Next, in order to solve the problem of inconsistent data formats between different hospitals and equipment, an adaptive standardized interface is designed. This interface can identify data formats from different sources and automatically convert them into a unified standard format through rules or algorithms, ensuring that data can be seamlessly transmitted and integrated across institutions. It is worth noting that this standardized interface is not static, but highly adaptable. It can dynamically adapt with the technological upgrade of medical institution equipment and the introduction of new data types, ensuring the long-term availability and scalability of the system.

[0053] In order to further enhance the dynamics and availability of data, labeling technology is introduced in data processing. Through labeling technology, real-time dynamic labels can be given to each patient's health data. These labels not only represent the patient's health status at a certain moment, but can also be updated in real time according to the process of the patient's condition changing. For example, if the patient's vital signs such as heart rate and blood pressure change, the system will automatically generate new labels, such as "acute heart failure is progressing" or "good treatment response". These labels will become the core basis for subsequent data analysis and decision support.

[0054] By converting various health data of critically ill patients into multi-dimensional health trajectories, long-term and detailed tracking of the patients' conditions can be carried out. The health data includes vital signs (such as heart rate, blood pressure, blood oxygen saturation, etc.), medical history (such as past medical history, family history, etc.), and treatment feedback (such as drug use, surgical treatment response, etc.). These data are often presented in the form of time series, reflecting the dynamic changes in the patients' health conditions. Therefore, by modeling them through time series analysis, the potential laws and trends in the development of the patients' conditions can be captured. Through time series analysis methods, especially models such as long short-term memory network (LSTM) or gated recurrent unit (GRU) in deep learning, the long-term dependencies in the changes of the patients' conditions can be effectively identified. This enables the system to not only rely on current data for prediction, but also deduce the possible paths of future condition development based on the patients' historical health data.

[0055] Next, in order to further improve the accuracy and flexibility of condition deduction, a multi-level condition deduction engine is constructed by combining deep learning and reinforcement learning methods. The deep learning part uses complex neural network structures to deeply explore the non-linear relationships and potential patterns in the patients' data, while the reinforcement learning part can self-learn and optimize treatment plans through interaction with the environment. Especially during the process of the patients' condition changes, the treatment effect can be improved by continuously adjusting treatment strategies. The introduction of reinforcement learning makes the deduction not only limited to simple condition prediction, but also able to simulate the effects of different intervention measures and recommend the most suitable treatment plan for doctors. During the deduction process, the system will use historical data to simulate the condition, predict the possible future development of the patients' conditions, and generate multiple treatment plans based on these predictions. These plans not only consider the patients' current conditions, but also the potential impacts of different intervention measures on the condition development, such as changing drug treatment, adjusting treatment intensity, or choosing different surgical methods, etc. Most importantly, according to the patients' real-time responses during the treatment process, the system will dynamically adjust the intervention strategy. Specifically, when the patients' health conditions change, the system can monitor these changes in real time and, through comparative analysis with historical data, automatically adjust the treatment plan to ensure the timeliness and personalization of the treatment. For example, if the patients' responses after receiving a certain treatment are not as expected, the system will adjust the treatment plan according to the new health trajectory prediction and provide optimized intervention measures. In this way, the system can continuously track the patients' conditions, conduct dynamic deduction, and make the optimal treatment decision in a timely manner according to the condition changes.

[0056] Real-time adjustment of cross-institutional treatment paths based on collaborative optimization proposes a multi-objective optimization algorithm model for calculating and generating optimal treatment paths. The core idea of this model is to make a comprehensive trade-off among multiple variables, considering the patient's health status, treatment options, doctor availability, hospital resources, and the collaborative capabilities among institutions to achieve a globally optimal treatment path. Specifically, the algorithm first collects and analyzes the patient's current health status, including vital signs, medical history, treatment feedback, etc., while evaluating the currently available treatment options and the professional capabilities and availability of doctors. At the same time, the status of hospital resources, such as the availability of beds, equipment, drugs, and other medical support resources, is also incorporated into the model to ensure that the treatment plan can be implemented within the scope of available resources. In addition, the algorithm also takes into account the collaborative capabilities among different medical institutions, which not only includes physical resources (such as shared beds, equipment) and human resources (such as the allocation of doctors and nurses), but also includes the information flow and data sharing capabilities among hospitals, which is particularly important because cross-institutional collaboration can provide more treatment options and support, especially in cases where the treatment needs of critically ill patients are high and resources are scarce.

[0057] By integrating these factors, the algorithm can make a trade-off among multiple objectives and find an optimal treatment path. The optimal path not only meets the patient's medical needs but also maximizes the utilization efficiency of resources in actual operation and ensures smooth cross-institutional collaboration. In addition, to cope with the rapidly changing environment in critical care, the model also introduces a dynamic adjustment mechanism. By continuously monitoring changes in the patient's condition, the occupancy of medical resources, and doctor availability, the algorithm can obtain new data in real time and adjust the treatment path. For example, when the patient's condition suddenly deteriorates, the original treatment plan may require higher-priority resources, or some resources (such as ICU beds, specific equipment) become unavailable, the algorithm will immediately recalculate the optimal path and dynamically adjust the treatment plan. In addition, when the resources of the hospital where the patient is located reach a bottleneck, the model will consider whether the patient needs to be transferred to other medical institutions and optimize the allocation of resources and doctors among different institutions.

[0058] By constructing a decision support model based on an expert knowledge base and machine learning, this model combines the experience and expertise of a large number of medical experts and incorporates machine learning algorithms, enabling the system to automatically generate personalized treatment recommendations that are most suitable for each patient according to the patient's condition, historical treatment records, and the treatment capabilities of medical institutions. The expert knowledge base contains profound accumulations in various fields such as various diseases, treatment plans, drug use, and treatment efficacy. The machine learning model can learn from a large number of historical cases, extract potential patterns, and make more accurate decisions based on the characteristics of different patients and in combination with real-time data. For example, when data such as the patient's vital signs, laboratory test results, and treatment feedback are input into the system, the machine learning algorithm will analyze this data and, in combination with the knowledge in the expert database, recommend the best treatment path for the patient or adjust the existing treatment plan according to the patient's special circumstances. The introduction of this intelligent decision support model makes treatment recommendations not only rely on the personal experience of doctors but are based on global big data analysis and intelligent learning, being more objective and scientific. Secondly, in a cross-institutional collaborative environment, by recommending the best treatment plan, the system can provide real-time and accurate treatment guidance for each medical institution. Each medical institution inputs the patient's health data and treatment feedback into the system, and the system analyzes the data in real time and automatically adjusts the treatment plan according to the changes in the patient's condition.

[0059] In addition, whenever the treatment plan is adjusted, the system will send update information to all relevant medical institutions to ensure that all participating doctors, nurses, and other medical staff can timely obtain the latest treatment plan and make the next treatment decision based on it. The most crucial thing is that all participating institutions and medical staff can conduct real-time interaction and information sharing through a centralized collaborative decision-making platform. The platform not only allows doctors and nursing staff from different institutions to share patient information but also enables the sharing of treatment progress, risk assessment, and feedback information on treatment plans. Through this real-time circulation of information, all medical institutions can jointly discuss and reach a consensus to ensure the accuracy and consistency of treatment decisions. The platform will also dynamically adjust the treatment plan according to the resource situation of different institutions and the needs of patients to ensure that cross-institutional collaboration can maximize its effectiveness.

[0060] Example 1:

[0061] A critically ill patient named Mr. Li was hospitalized for severe heart failure. During the treatment process, the changes in Mr. Li's vital signs, medical history, and treatment feedback were monitored and recorded in real time. According to one of the core technologies of the cross-institutional collaborative data processing method for critical care, the method of dynamic disease evolution and intervention strategy based on the patient's health trajectory is adopted, and Mr. Li's health trajectory is formed through multi-dimensional data fusion. Mr. Li's vital signs include physiological data such as heart rate, blood pressure, and blood oxygen saturation. The medical history includes past medical histories of hypertension and diabetes, as well as treatment feedback data (such as drug reactions, body position change reactions, etc.). These data are collected from multiple different sources (such as hospital monitoring equipment, historical medical records, drug usage records, etc.) to form a multi-dimensional health data input system.

[0062] First, in order to achieve multi-dimensional data fusion, a health state vector X is defined t , which represents the patient's health state at time t. This health state vector is composed of three data dimensions: vital signs H(τ), medical history B(τ), and treatment feedback F(τ). Each dimension changes over time, and these data types are different: vital signs are continuous (e.g., heart rate changes at any time), medical history is discrete (e.g., the patient's medical histories of hypertension and diabetes), and treatment feedback is discrete data based on the patient's reaction to drugs or treatments. Therefore, a multi-dimensional data fusion function ψ(·) is used, and the role of this function is to integrate these heterogeneous data together.

[0063] For example, during Mr. Li's treatment, assume that the health state data at time t - 1 is H(t - 1) = 80 beats per minute (heart rate), B(t - 1) = 1 (hypertension history flag, with a value of 1 indicating a medical history), and the treatment feedback F(t - 1) = 1.2 (indicating the reaction of blood pressure to the drug, and 1.2 means the blood pressure has decreased). Based on this data, time series modeling can be performed through the following formula:

[0064]

[0065] Among them, ψ(H(τ), B(τ), F(τ)) is a multi-dimensional data fusion function, and the specific implementation may process these data through weighted average or more complex non-linear functions. Assume that the weights assigned by this function to different dimension data are: the weight of heart rate data is 0.5, the weight of medical history data is 0.3, and the weight of treatment feedback data is 0.2. Then the result of this fusion function can be expressed by the following formula:

[0066] ψ(H(τ), B(τ), F(τ)) = 0.5·H(τ) + 0.3·B(τ) + 0.2·F(τ)

[0067] In Mr. Li's actual data, his health data at time t is set as follows: heart rate H(t) = 82 beats per minute, medical history flag B(t) = 1, treatment feedback F(t) = 1.1. Then the value of the fusion function is:

[0068] ψ(82, 1, 1.1) = 0.5·82 + 0.3·1 + 0.2·1.1 = 41 + 0.3 + 0.22 = 41.52

[0069] Next, perform the time series integration operation. By integrating the health status data from time t - 1 to t, assuming Mr. Li's health data is relatively stable during this period, that is, it can be considered that the data changes little during this period, and the integration result is:

[0070]

[0071] At this time, Mr. Li's health status vector X t is 41.52, and this value can represent the comprehensive health status of the patient at time t. After obtaining the health status vector, next, use the long short-term memory network (LSTM) in deep learning to capture the time series changes of the patient's condition and establish a disease progression model. LSTM can predict the patient's future health status based on historical data. Considering the long-term dependence, it can better reflect the long-term trend of the patient's condition. By training this LSTM model, the system can simulate the changes in Mr. Li's condition and predict the possible disease development trajectory of Mr. Li in combination with historical data.

[0072] Assume that through the prediction of the LSTM model, Mr. Li's health status will deteriorate within the next few hours, the heart rate will increase to 90 beats per minute, and the treatment feedback data shows that the effect of the drug is gradually weakening. At this time, the system will, based on these deduction results, intelligently recommend a treatment plan. According to the deduction model, the system may recommend adjusting the drug dosage or considering transferring to the intensive care unit (ICU) to provide higher-intensity treatment.

[0073] To further optimize the treatment strategy, the system will evaluate the potential impact of these measures on the condition by simulating different intervention measures. For example, after simulating an increase in the drug dosage, the system will predict whether the condition can be effectively controlled based on the new health trajectory. If the patient's health status has not been significantly improved, the system will recommend trying other intervention measures, such as adjusting the treatment plan or changing the drug.

[0074] Assume that Mr. Li's condition gradually deteriorates due to heart failure, and as the treatment progresses, the treatment decisions of doctors and the system will be continuously optimized. Reinforcement learning gradually finds the optimal treatment path by learning the feedback effects of the patient under different treatment plans, enabling Mr. Li to obtain the most suitable treatment in a shorter time.

[0075] In the process of implementing reinforcement learning, the expected reward formula for treatment decisions is first defined as follows:

[0076] J(θ) = E π(θ) [∫ 0 T γ t r t (θ, X t ) dt]

[0077] where J(θ) is the expected reward of the treatment strategy, representing the overall reward value calculated by time integration under the current strategy; π(θ) is the treatment strategy, and the optimal strategy obtained through reinforcement learning training; r t (θ, X t ) is the immediate reward function, which measures the reward value corresponding to the patient's health state X t when the strategy θ is adopted at time t; γ is the discount factor, used to balance the influence degrees of the current reward and future rewards; T is the time endpoint, representing the time span of the deduction process.

[0078] The immediate reward function r t (θ, X t ) is used to measure the improvement degree of each treatment decision on the patient's health state. It is set that Mr. Li's heart rate has dropped from 90 beats per minute to 80 beats per minute during the treatment, and the blood oxygen level has improved from 88% to 92%, which means that the treatment has achieved certain effects. Therefore, the treatment plan adopted at this time t can obtain a positive immediate reward. Defining the immediate reward as the improvement degree of the treatment effect, it can be expressed as:

[0079] r t (θ, X t ) = w 1 ·ΔH t + w 2 ·ΔO t

[0080] where ΔH t is the change in the patient's heart rate at time t, ΔO t is the change in the patient's blood oxygen level, w 1 and w 2 are weight coefficients, used to balance the influence of different physiological indicators on the reward. It is set that w 1 = 0.6, w 2 = 0.4, then the immediate reward can be calculated as:

[0081] r t (θ, X t ) = 0.6·(90 - 80) + 0.4·(92 - 88) = 0.6·10 + 0.4·4 = 6 + 1.6 = 7.6

[0082] Therefore, when Mr. Li received treatment, the system gave an immediate reward of 7.6, indicating that the treatment was effective.

[0083] Next, the expected reward formula is introduced to optimize the strategy. In the process of reinforcement learning, the goal is to maximize the expected reward J(θ) to obtain the optimal treatment strategy. The expected reward is calculated through time integration, and the formula is:

[0084] J(θ) = E π(θ) [∫ 0 T γ t r t (θ, X t )dt]

[0085] Among them, the discount factor γ is used to balance the influence of the current reward and future rewards. Usually, the value range of γ is from 0 to 1. The larger γ is, the more the reinforcement learning focuses on the accumulation of future rewards; t is the total duration of the treatment deduction, and it is set that T = 24 hours. In this example, it is set that the treatment process starts at time 0. After 24 hours, Mr. Li's condition has improved. It is set that at each moment of the treatment, the system calculates the immediate reward and adjusts the strategy using the discount factor, and finally obtains an expected reward value.

[0086] It is set that the immediate reward r t (θ, X t ) at each moment t of Mr. Li's treatment remains at 7.6 (that is, the treatment effect continues to improve), then the expected reward can be expressed as:

[0087] J(θ) = E π(θ) [∫ 0 24 0.95 t ·7.6dt]

[0088] Among them, the discount factor γ = 0.95, indicating that the influence of future rewards is relatively large. By calculating this integral, the expected reward can be obtained:

[0089]

[0090] Therefore, Mr. Li's expected reward is 116.51, indicating that under the current treatment strategy, the expected value of Mr. Li's overall treatment effect in the next 24 hours is 116.51. By optimizing this expected reward, the system can adjust the treatment plan and gradually find the most suitable treatment path for Mr. Li.

[0091] The core of reinforcement learning lies in continuously interacting with the environment (the patient's health status) to optimize the treatment strategy. In Mr. Li's case, the system adjusts the treatment plan through a reinforcement learning algorithm and updates the expected reward based on new data feedback. If Mr. Li's health deteriorates at a certain moment (such as the heart rate abnormally rising to 110 beats per minute), the immediate reward will decrease, and the expected reward will also be affected. At this time, the system may adjust the drug dosage, change the treatment strategy, or refer the patient to a higher-level hospital for treatment.

[0092] Based on real-time patient data and historical records, simulate the effects of various treatment measures and evaluate the long-term impact of each treatment plan. This can dynamically adjust the treatment strategy according to the simulation results and provide a personalized treatment path. In this plan, the optimization degree and evaluation mechanism of the treatment plan are very important core links. Mr. Li's treatment plan can be optimized by means of this method. First, use the following formula to optimize and evaluate the treatment plan:

[0093]

[0094] In this formula, represents the optimization evaluation of all treatment strategies by the system at time t, aiming to evaluate the overall effect of the current treatment plan; ρ k is the weight of the intervention measure T k , indicating the contribution degree of each treatment measure to the optimization goal; η k (X t ,T k ) is the effect evaluation function, indicating the treatment effect brought by taking the intervention measure T t under the current state X k ; e -λt is the decay factor, controlling the decay degree of the treatment effect over time; λ is the decay constant, used to adjust the decay rate in time.

[0095] First, the system defines a variety of intervention measures and assigns a weight to each measure. Mr. Li's treatment plan includes drug treatment, mechanical ventilation support, and physical therapy. The weight ρ k of each intervention measure reflects its importance in the overall treatment plan. Set:

[0096] The weight ρ 1 of drug treatment T 1 = 0.5

[0097] The weight ρ 2 of mechanical ventilation support T 2 = 0.3

[0098] The weight ρ 3 of physical therapy T 3= 0.2

[0099] Next, the system evaluates the effectiveness of each intervention through the patient's health status X t These effects are given by the effectiveness evaluation function η k (X t , T k ). For example, drug treatment may reduce Mr. Li's heart rate and improve his oxygenation level, while mechanical ventilation may help improve his respiratory function. In this setting, the following effectiveness evaluation functions are obtained:

[0100] Drug treatment: η 1 (X t , T 1 ) = 0.6·ΔH t + 0.4·ΔO t , where ΔH t and ΔO t represent the changes in heart rate and blood oxygen level respectively.

[0101] Mechanical ventilation: η 2 (X t , T 2 ) = 0.7·ΔB t + 0.3·ΔO t , where ΔB t represents the improvement in respiratory function.

[0102] Physical therapy: η 3 (X t , T 3 ) = ΔP t , where ΔP t represents the degree of recovery of Mr. Li's motor function.

[0103] For example, assume that in the first hour of treatment, Mr. Li's heart rate drops from 100 beats per minute to 90 beats per minute, and his blood oxygen level increases from 85% to 90%. The effectiveness of drug treatment can be calculated as follows:

[0104] η 1 (X t , T 1 ) = 0.6·(100 - 90)+ 0.4·(90 - 85)= 0.6·10 + 0.4·5 = 6 + 2 = 8

[0105] At the same time, the effect of mechanical ventilation may be: the respiratory function has increased by 5%, and the oxygenation has improved by 3%. Therefore, the effectiveness evaluation of mechanical ventilation is:

[0106] η 2 (X t , T 2 ) = 0.7·5+ 0.3·3 = 3.5 + 0.9 = 4.4

[0107] The therapeutic effect usually decays over time, so the decay factor e is introduced -λt , to control the decay of the effect over time. It is set that the impact of the treatment weakens gradually over time, so the decay constant λ = 0.05 is selected, that is, the therapeutic effect decays by 5% per hour. During Mr. Li's treatment, considering that at each moment, the effects of all interventions will decay gradually over time, the decay factor will be used for each effect evaluation value. It is set that within the first hour, the decay factor is:

[0108] e -λt = e -0.05·1 ≈ 0.951

[0109] Next, the system calculates the optimization degree of the treatment plan by performing weighted sum and time integration on the optimized evaluation of all treatment measures , that is, the optimization degree of the overall therapeutic effect. According to the above weights and effect evaluations, the value of the optimization degree can be calculated. For the first hour, the optimization degree of the treatment plan is:

[0110]

[0111] It is set that the effect of physical therapy is 3.2, and we get:

[0112]

[0113] According to the calculated optimization degree of the treatment plan The system will dynamically adjust the treatment strategy. If the optimization degree value is lower than the set threshold (for example, 6.0), the system will automatically adjust the treatment plan. For example, if the impact of drug treatment on heart rate and oxygenation is not obvious enough, the system may increase the dose of the drug, or transfer the patient to a medical institution with more resources for more intensive treatment. It is set that after the end of the first hour, the optimization degree does not reach the expectation, and the system may adjust the treatment path, such as strengthening drug treatment and reducing the dependence on mechanical ventilation, or introducing new interventions to improve the therapeutic effect.

[0114] Example 2:

[0115] Ms. Zhang is a critically ill patient in need of emergency treatment and has been referred by multiple hospitals due to acute heart failure. Her case is very complex and requires rapid coordination among multiple hospitals and medical resources to ensure the best treatment effect. In response to this situation, the system adopts a real-time adjustment method for cross-institutional treatment paths based on collaborative optimization. Through a multi-objective optimization algorithm model, it comprehensively considers Ms. Zhang's health status, doctor availability, occupancy of hospital resources, and cross-hospital collaboration capabilities to calculate the optimal treatment path. The objective function in the system is composed of the weighted sum of multiple objective functions and constraint functions. The treatment path is adjusted in real time through the optimization algorithm to ensure the most reasonable scheduling of the resources of each hospital and doctor. The following are the specific application steps and calculation processes.

[0116] First, the system defines the overall objective function F(x), which is the comprehensive treatment path index to be optimized and reflects the weighted comprehensive effect of each objective. The objective function is as follows:

[0117]

[0118] In this formula, F(x) represents the overall objective function, reflecting the weighted comprehensive effect of multiple objectives, aiming to provide the optimal treatment path for Ms. Zhang; f i (x) is the i-th objective function, representing the optimization objective related to the i-th objective, such as the improvement of the patient's health status; h j (x) is the j-th constraint function, representing the function related to resource allocation, hospital capacity, and doctor availability resource constraints; α i and γ j are the weight coefficients of the objective function and the constraint function respectively, used to adjust the importance of each objective and constraint in the overall optimization; x is the vector of optimization variables, representing specific treatment path selection, resource allocation, and scheduling decision factors of doctors and hospitals.

[0119] Step 1: Define the objective function:

[0120] First, the system needs to define multiple objective functions f i (x) to represent different treatment objectives. Taking Ms. Zhang's case as an example, the objective functions can include the following items:

[0121] This objective reflects the degree of improvement in the patient's health after treatment and can be measured by the patient's physiological indicators (such as blood pressure, heart rate, blood oxygen saturation, etc.). The objective function can be expressed as:

[0122] f 1 (x) = HealthImprovementScore = w 1 ·ΔBP + w 2 ·ΔHR + w 3·ΔSpO2

[0123] where w 1 , w 2 , w 3 are weight coefficients, representing the contributions of blood pressure, heart rate, and blood oxygen saturation to health improvement respectively. The weight coefficients are set as w 1 = 0.4, w 2 = 0.3, w 3 = 0.3.

[0124] This objective measures the workload and availability of each doctor. Assuming there are multiple doctors available for treatment, the objective function f 2 (x) represents the optimization of the doctor's workload. Through the scheduling algorithm, the goal is to minimize the workload of each doctor while ensuring sufficient available doctor resources:

[0125]

[0126] where DoctorWorkload k represents the workload of doctor k. Assuming there are 3 doctors in the hospital with workloads of 3, 4, and 5 hours respectively, then f 2 (x) = 3 + 4 + 5 = 12.

[0127] This objective function f 3 (x) represents the resource occupancy required for patient treatment, including hospital beds, drugs, equipment, etc. The optimization goal is to reduce resource occupancy and improve resource utilization efficiency:

[0128] f 3 (x) = ResourceUsage = w 4 ·ΔICUBeds + w 5 ·ΔVentilators

[0129] Assuming the change in ICU bed occupancy ΔICUBeds = 2, the change in mechanical ventilation occupancy ΔVentilators = 1, and the weight coefficients w 4 = 0.5, w 5 = 0.5, then f 3 (x) = 0.5·2 + 0.5·1 = 1.5.

[0130] Step 2: Define the constraint function:

[0131] The constraint function h j (x) is used to reflect the constraint conditions of various resources and capabilities in the treatment plan. Taking Ms. Zhang's treatment as an example, the constraint conditions can include:

[0132] The resources of each department in the hospital are limited, and the optimization of the treatment path must take into account the carrying capacity of the hospital. For example, the number of ICU beds limits the admission capacity of each hospital. Set the current ICU resource limit of the hospital to 5 beds. Set h 1 (x) represents the usage constraint of ICU beds:

[0133] h 1 (x) = ICUBedUsage - 5

[0134] If h 1 (x) > 0, it means that the hospital's capacity has been exceeded, and treatment needs to be arranged in other hospitals.

[0135] Ms. Zhang needs cross-hospital collaboration for treatment, and the collaboration ability between hospitals is an important constraint. Set h 2 (x) represents the collaboration ability constraint between two hospitals. The collaboration ability between Hospital A and Hospital B during the patient transfer process is 0.8 (the maximum is 1):

[0136] h 2 (x) = 1 - 0.8 = 0.2

[0137] A value lower than 0.5 means that the current collaboration level is insufficient, and information sharing and equipment support between hospitals need to be improved.

[0138] Step 3: Calculate the weighted sum of the objective function and the constraint function:

[0139] According to the above definitions, the system can calculate the comprehensive objective function F(x) to optimize the treatment path. For example, set the following weight coefficients:

[0140] α 1 = 0.4 (improvement of health status)

[0141] α 2 = 0.3 (availability of doctors)

[0142] α 3 = 0.3 (occupation of hospital resources)

[0143] γ 1 = 0.5 (hospital capacity constraint)

[0144] γ 2 = 0.5 (cross-hospital collaboration ability constraint)

[0145] The system can calculate the optimized objective function:

[0146] F(x) = 0.4·f 1 (x) + 0.3·f 2 (x) + 0.3·f 3 (x) + 0.5·h1 (x) + 0.5·h 2 (x)

[0147] Set:

[0148] f 1 (x) = 8 (Improvement in health status)

[0149] f 2 (x) = 12 (Doctor availability)

[0150] f 3 (x) = 1.5 (Hospital resource occupancy)

[0151] h 1 (x) = 0 (ICU constraint satisfaction)

[0152] h 2 (x) = 0.2 (Cross - hospital collaboration constraint)

[0153] Then:

[0154] F(x) = 0.4·8 + 0.3·12 + 0.3·1.5 + 0.5·0 + 0.5·0.2 = 3.2 + 3.6 + 0.45 + 0.1 = 7.35

[0155] Step 4: Adjust the treatment path in real - time:

[0156] When F(x) reaches the preset target value (e.g., greater than 7), the system will automatically adjust the treatment path. If the objective function value is below the threshold, the system will continue to optimize the objective function and adjust the treatment plan, which may involve adjusting the bed allocation, increasing the doctor's working hours, and improving the collaboration mechanism between hospitals until the optimal treatment path is achieved.

[0157] In this embodiment, Ms. Zhang is jointly treated by multiple hospitals. Facing the complex situation of acute heart failure and complications, the selection of the treatment path is crucial. According to the real - time data such as Ms. Zhang's condition, resource occupancy, and doctor availability, the treatment path is adjusted in real - time to ensure the best treatment effect in the environment of multiple - hospital collaboration.

[0158] In this process, the system dynamically adjusts the treatment path through a differential equation model. The core of this model lies in the real - time calculation and adjustment of the change in the patient's health status, and is modeled by the following formula:

[0159]

[0160] Among them, Represents the rate of change of the patient's health status, that is, the degree of change in the patient's condition at time t, reflecting the immediate effects of the treatment plan and resource allocation. Next, the specific meanings of each variable and coefficient will be explained one by one, and examples will be given to illustrate how to apply these formulas in actual operations.

[0161] 1. Rate of change of health status

[0162] This is the left - hand side of the equation, representing the rate of change of the patient's health status at time t. Ms. Zhang's health status may change with the fluctuations of her condition. The system dynamically calculates the change in the patient's health status by real - time monitoring of physiological data such as heart rate, blood pressure, and oxygen saturation. For example, if Ms. Zhang's heart failure symptoms worsen and her blood oxygen level drops, the system dynamically adjusts the rate of change of her health status through real - time feedback of the monitored data.

[0163] At a certain moment, assume Ms. Zhang's health status vector x(t) = [BP(t), HR(t), SpO2(t)] is [100, 90, 88%], where BP(t) is blood pressure, HR(t) is heart rate, and SpO2(t) is blood oxygen saturation. By calculating its rate of change, the system can track the changing trends of each parameter in real - time to ensure rapid adjustment of the treatment path.

[0164] 2. Treatment plan impact matrix A(t)

[0165] Matrix A(t) describes the impact of the treatment plan on the patient's health status, specifically representing the effect of different treatment measures over time. Ms. Zhang may have received multiple treatments such as anti - failure drug treatment, diuretic treatment, and mechanical ventilation. The system needs to evaluate the immediate impact of each treatment measure on her health status.

[0166] Assume Ms. Zhang's treatment plan includes using drug A (such as an ACE inhibitor) and diuretic treatment B. Then matrix A(t) can be expressed as:

[0167]

[0168] Among them, each element represents the degree of impact of the treatment plan on the corresponding health index (such as blood pressure, heart rate, blood oxygen saturation). For example, the first element A 11 (t)= - 0.02 indicates that this drug will cause the blood pressure to drop by 0.02 units per hour, and the negative value represents a decreasing effect.

[0169] 3. External factor matrix B(t)

[0170] The matrix B(t) describes the impact of external factors such as resource utilization and doctor availability on the patient's health status. For example, the occupancy of ICU resources in a hospital, the availability of doctors, and the availability of treatment equipment, etc., will all affect the patient's health status. For instance, it is assumed that the ICU resources in a certain hospital are approaching full capacity, while another hospital has more empty beds. The system adjusts the matrix B(t) according to the actual occupancy of hospital resources:

[0171]

[0172] Among them, the matrix elements represent the influence coefficients of each resource (such as ICU beds, doctor availability, etc.) on the change in Ms. Zhang's health status. It is assumed that u(t) represents the resource scheduling decision, and the system will dynamically adjust the treatment plan according to real-time data.

[0173] 4. Collaboration efficiency matrix C(t):

[0174] The matrix C(t) describes the collaboration efficiency between hospitals, reflecting the impact of collaborative work such as information sharing and equipment scheduling on the patient's treatment path. In the case of collaborative treatment in multiple hospitals, the collaboration efficiency is crucial for the treatment effect of patients.

[0175] For example, it is assumed that the collaboration efficiency coefficient between two hospitals is C(t) = 0.8, indicating that the collaboration ability between the two hospitals is strong, and information sharing and resource allocation can be carried out efficiently. This coefficient weighs the impact of the external disturbance d(t), thereby optimizing the treatment path. If the collaboration efficiency between hospitals is low, C(t) will be adjusted to a smaller value, resulting in more cautious resource transfer and treatment decisions.

[0176] 5. External disturbance d(t):

[0177] d(t) represents a variable of external disturbance or hospital collaboration, reflecting the real-time data of collaboration between different hospitals. For example, there may be a delay in sharing the treatment data of patients between hospitals, resulting in a postponement of the adjustment of the treatment path. It is assumed that at a certain time point t = 3, the disturbance d(t) caused by the delay in data sharing between hospitals is 0.05. This disturbance will affect the real-time adjustment of the treatment path, thereby postponing or advancing certain treatment decisions.

[0178] Calculation example of the dynamic adjustment process:

[0179] It is assumed that the vector of Ms. Zhang's health status at t = 3 is x(3) = [110, 92, 89%], the treatment plan matrix A(3) is the value given above, the resource matrix B(3) represents the availability of doctors and equipment, the collaboration efficiency matrix C(3) is 0.8, and the external disturbance d(3) = 0.05. The control variable u(3) represents the treatment decision. At this time, dynamic adjustment can be carried out through differential equations:

[0180]

[0181] Set u(3) to represent the ongoing treatment measure, and the optimization degree of resource allocation is 1 (fully effective). The calculation process is as follows:

[0182]

[0183] Through the above calculation, the system obtains the rate of change of Ms. Zhang's health status, and then adjusts the treatment path, such as increasing the drug dosage, adjusting the doctor's schedule, or providing more effective resource support in another hospital. The real-time adjustment method of the cross-institutional treatment path based on collaborative optimization combines a dynamic control model and a real-time data feedback mechanism, enabling the treatment path to adapt to multi-dimensional changes such as the patient's health status, hospital resources, and doctor availability.

[0184] The real-time adjustment method of the cross-institutional treatment path in the present invention introduces an optimization model, which can dynamically adjust the treatment path to ensure the efficient utilization of cross-hospital resources. Specifically, the collaboration efficiency between hospitals is modeled by a time-dependent collaboration efficiency function as follows:

[0185]

[0186] In this formula, C collab (R) represents the collaboration efficiency between hospitals, measuring the overall effect of cross-hospital resource sharing and cooperation. Next, each element in the formula will be explained one by one, and a practical case will be used to illustrate how to use this formula for dynamic adjustment of the treatment path.

[0187] This variable reflects the cooperation efficiency among multiple hospitals. During the treatment of Ms. Zhang, multiple hospitals are involved in the treatment at the same time, so it is necessary to coordinate the resource and information flow between hospitals. The level of collaboration efficiency directly determines the optimization of the treatment effect. For example, if the information sharing between Hospital A and Hospital B is rapid and the medical records and treatment records can be transmitted in real time, the treatment efficiency will be greatly improved and the collaboration efficiency will also increase. Conversely, if the collaboration efficiency between hospitals is low, the optimization of the treatment path will be affected, resulting in a worse treatment effect for the patient.

[0188] β k is the weight coefficient of the collaboration dimension, indicating the influence degree of each collaboration dimension on the overall collaboration efficiency. In practical applications, different collaboration dimensions may have different influences on the final treatment result. For example, the timeliness of information sharing, the allocation of doctor availability, and the rationality of resource scheduling may all have different importance. Suppose in the treatment of Ms. Zhang, information sharing and doctor availability are more important than equipment sharing, then larger weight coefficients β 1 and β 2, while a smaller weight β is set for equipment sharing 3 . Specifically, the possible range of values for the weight coefficient is:

[0189] β 1 = 0.4 (information sharing)

[0190] β 2 = 0.3 (doctor availability)

[0191] β 3 = 0.2 (equipment sharing)

[0192] The choice of the weight coefficient will affect the optimization result and determine which collaboration dimension is most critical for Ms. Zhang's treatment.

[0193] g k (R(t)) is the effectiveness function of the collaboration dimension, representing the impact of a specific collaboration on the treatment path during the optimization process at time t. For example, g 1 (R(t)) can represent the effectiveness of information sharing between Hospital A and Hospital B, depending on the real-time data transmission speed, interface interoperability, and information integrity. Similarly, g 2 (R(t)) can represent the effectiveness of doctor availability between Hospital A and Hospital C, reflecting the efficiency of doctor scheduling and response time.

[0194] The effectiveness function of the information sharing system between Hospital A and Hospital B can be described by the following formula:

[0195]

[0196] where R A (t) represents the resource usage of Hospital A at time t, such as the current bed occupancy rate of Hospital A. The parameter α 1 is a constant that controls the gradual decrease of effectiveness as resource occupancy increases. Setting α 1 = 0.3, when the resource utilization rate of Hospital A approaches 100%, the effectiveness of information sharing will decrease.

[0197] Similarly, the effectiveness function of doctor availability can be defined as:

[0198]

[0199] where R B (t) and R C (t) are the doctor availabilities of Hospital B and Hospital C respectively, and the parameter α 2 controls the change speed of collaboration effectiveness with the difference in doctor availability. Setting α 2= 0.5. When the doctor resources in Hospital B are sufficient and those in Hospital C are strained, the value of this function will change rapidly.

[0200] R(t) represents the allocation vector of cross-institutional resources, reflecting the resource utilization in different hospitals. Taking Ms. Zhang as an example, her treatment requires resource coordination among different hospitals, such as ICU beds, doctors, treatment equipment, etc. Suppose at time t = 2, the occupancy rate of ICU beds in Hospital A is 70%, the occupancy rate of beds in Hospital B is 50%, and the occupancy rate of beds in Hospital C is 30%. The resource allocation vector can be set as:

[0201]

[0202] τ 1 and τ 2 are the time intervals of collaboration, representing the duration of cross-institutional collaboration. In the case of Ms. Zhang, it is set that the time considered for collaboration is from the start of treatment to the 3rd day, i.e., τ 1 = 0, τ 2 = 3. Therefore, the calculation of collaboration effectiveness will consider the resource utilization and collaboration effects within this time range.

[0203] Suppose at time t = 2, the resource occupancy rate of Hospital A is 70%, the resource occupancy rate of Hospital B is 50%, and the resource occupancy rate of Hospital C is 30%. The collaboration effectiveness among hospitals can be calculated according to the aforementioned formula:

[0204] First, calculate the effectiveness function of each collaboration dimension:

[0205] 1. For information sharing:

[0206] g 1 (R(t)) = 1e -0.3·0.7 = 1e -0.21 ≈0.188

[0207] 2. For doctor availability:

[0208]

[0209] Then, calculate the weighted sum of the collaboration effectiveness function:

[0210] C collab (R) = 0.4·0.188 + 0.3·0.524 + 0.2·0.3 = 0.0752 + 0.1572 + 0.06 = 0.2924

[0211] Therefore, the collaboration efficiency among hospitals at this time point is 0.2924. This result indicates that the current inter-hospital collaboration is at a medium level, and it may be necessary to further optimize resource allocation and information flow to improve the overall treatment efficiency.

[0212] By introducing factors such as resource sharing, information flow, and doctor availability among hospitals, this method can provide Ms. Zhang with real-time optimization of the treatment path, ensuring the efficient utilization and coordinated work of resources in each hospital. During the treatment process, the system dynamically adjusts the collaboration efficiency, resource scheduling, and doctor allocation, so that Ms. Zhang can obtain the optimal treatment path.

Claims

1. A cross-institutional collaborative data processing method for critical care, characterized by The following steps are involved: S1. Cross-institutional data standardization and dynamic adaptation based on critical care characteristics: S1.

1. Define a hierarchical data model based on the field of critical care, and model various types of patient information including vital signs, clinical diagnosis, and laboratory test results in time series; S1.

2. Design adaptive standardized interfaces to identify and convert data formats from different hospitals or devices; S1.3, introduce real-time dynamic tags for patient data through tagging technology; S2. Dynamic disease deduction and intervention strategy based on the patient's health trajectory: S2.

1. By converting various health data of critically ill patients, including vital signs, medical history, and treatment feedback, into multidimensional health trajectories, the patient's condition is modeled continuously using time series analysis; S2.

2. Combine deep learning and reinforcement learning methods to establish a disease evolution deduction engine; predict the disease progression based on historical data, and intelligently recommend treatment strategies by simulating the impact of different intervention measures on the disease; S2.

3. Generate multiple treatment plans based on the dynamic deduction results, and dynamically adjust the intervention strategy based on the patient's real-time response; S3. Real-time adjustment of cross-institutional treatment pathways based on collaborative optimization: S3.

1. Design an algorithm model based on multi-objective optimization to calculate the health status, treatment options, doctor availability, hospital resources, and inter-institutional collaboration capabilities of critically ill patients to generate the optimal treatment pathway; S3.

2. Use real-time data including patient condition changes and resource usage to dynamically adjust treatment pathways; S4. Intelligent decision support system and cross-institutional collaborative decision-making framework: S4.

1. Establish a decision support model based on expert knowledge base and machine learning, integrate the patient's condition data, historical treatment records, and the treatment capabilities of medical institutions to generate personalized treatment recommendations; S4.

2. In a cross-institutional collaborative environment, by recommending the best treatment plan, each medical institution conducts cross-institutional decision-making collaboration in real time based on patient data input, treatment plan adjustment and feedback; S4.

3. Build a real-time collaborative decision-making platform with the participation of multiple parties, allowing doctors and nurses from different institutions to share patient information, treatment progress and risk assessment in real time to make collaborative decisions.

2. The critical care cross-institutional collaborative data processing method according to claim 1 is characterized in that The dynamic disease deduction and intervention strategy method based on the patient's health trajectory includes: Through multi-dimensional data fusion, multiple data dimensions from vital signs, patient medical history and treatment feedback are integrated to form a complete health trajectory; time series analysis technology is used for modeling, and long short-term memory networks are used to capture the time series information in the data; the calculation formula is: Among them, X t is the health state vector at time t, representing the health data composed of vital signs Hτ, medical history B(τ) and treatment feedback (F(τ)) information; the function ψ(·) is a multidimensional data fusion function, which aims to combine continuous physiological signals with discrete medical history information heterogeneous data; the symbol τ represents the time variable, and dτ represents the integration operation over time.

3. The critical care cross-institutional collaborative data processing method according to claim 2 is characterized in that The dynamic disease deduction and intervention strategy method based on the patient's health trajectory includes: Each treatment decision in reinforcement learning receives an immediate reward including the degree of improvement in the patient's condition, and the treatment strategy is optimized through long-term feedback; the expected reward formula is introduced to optimize the strategy and maximize the treatment effect; J(θ)=E π(θ) [∫0 T γ t r t (θ,X t )dt] Among them, J(θ) is the expected reward of the treatment strategy, which means the overall reward value calculated by time integration under the current strategy; π(θ) is the treatment strategy, the optimal strategy trained by reinforcement learning; r t (θ,X t ) is the immediate reward function, which measures the patient's health status X when the strategy θ is adopted at time t. t The corresponding reward value; γ is the discount factor, which is used to balance the impact of current rewards and future rewards; T is the end time, indicating the time span of the deduction process.

4. The critical care cross-institutional collaborative data processing method according to claim 3 is characterized in that The dynamic disease deduction and intervention strategy method based on the patient's health trajectory includes: through the combination of reinforcement learning and simulated intervention, according to real-time patient data and historical records, simulating the effects of multiple treatment measures, and evaluating the long-term impact of each treatment plan; based on the simulation results, dynamically adjusting the treatment strategy to provide a personalized treatment path.

5. The critical care cross-institutional collaborative data processing method according to claim 1 is characterized in that The method for real-time adjustment of cross-institutional treatment pathways based on collaborative optimization introduces a multi-objective optimization algorithm model, integrates the patient's health status, the availability of doctors, the occupancy of hospital resources, and the cross-hospital collaboration capabilities into a unified optimization framework, and calculates the optimal treatment pathway; the objective function integrates multiple objectives and constraints in a weighted summation manner, and determines the global optimal treatment pathway based on the weight of each objective.

6. The critical care cross-institutional collaborative data processing method according to claim 5 is characterized in that The real-time adjustment method of cross-institutional treatment pathway based on collaborative optimization: A real-time data feedback mechanism is introduced, combined with a dynamic control model, to adjust the treatment path based on the real-time data of the condition, resource usage, and doctor availability; the algorithm is dynamically adjusted through the following differential equation model: in: It represents the rate of change of the patient's health status, reflecting the change of the patient's condition at time t; A(t) is a dynamic matrix that describes the impact of the treatment plan on the patient's health status, and represents the effect of different treatment measures over time; x(t) is the patient's health status vector, which is a specific manifestation at time t; B(t) is a matrix that describes the impact of external factors such as resource utilization and doctor availability on the patient's health status; u(t) is a control variable, representing the decision variables of treatment plan, resource scheduling, and doctor arrangement; C(t) is a matrix that represents the impact of collaborative effectiveness, and determines the real-time impact of information sharing and resource coordination between hospitals; d(t) is a variable of external disturbance or hospital collaboration, reflecting the real-time data of collaboration between different hospitals and the impact on the patient's treatment path.

7. The critical care cross-institutional collaborative data processing method according to claim 6 is characterized in that The real-time adjustment method of cross-institutional treatment pathway based on collaborative optimization: The resource sharing, information flow, and physician availability factors among hospitals are incorporated into the optimization model to optimize cross-institutional resource scheduling. The collaboration efficiency among hospitals is modeled through the following time-dependent collaboration efficiency function: in: C collab (R) represents the collaboration effectiveness among hospitals, reflecting the overall effect of cross-hospital resource sharing and cooperation; β k is the weight coefficient of each collaboration dimension, indicating the impact of different collaboration factors on the overall collaboration effectiveness; g k (R(t)) is the effectiveness function of the kth collaboration dimension, which indicates the effectiveness of a specific collaboration at time t; R(t) is the configuration vector of cross-institutional resources, which reflects the resource usage of different hospitals at time t; τ1 and τ2 are the time intervals of collaboration, indicating the duration of collaboration.

Images