A tumor complication grading early warning and follow-up method

Abstract

The application discloses a tumor complication grading early warning and follow-up method, relates to the field of intelligent medical treatment and medical data processing technology, and introduces a risk momentum model constructed by using a physical dynamics principle, calculates a directed change rate and a deterioration acceleration of an index in real time, and generates a comprehensive risk value of a patient; meanwhile, a resource load model is constructed, and a resource response threshold value is dynamically calculated based on a real-time load rate by using an exponential function. Through game matching of the comprehensive risk value and the resource threshold value, a grading scheduling instruction is generated, and the gradient descent method is used to correct model parameters in the reverse direction according to a real severity label. The application breaks through the hysteresis of static threshold monitoring, realizes advanced perception of a disease deterioration trend and adaptive optimization of medical resources.

Description

Technical Field

[0001] This invention relates to the field of smart healthcare and medical data processing technology, and more specifically, to a method for graded early warning and follow-up of tumor complications. Background Technology

[0002] Chemotherapy for malignant tumors, as an important form of systemic anti-tumor therapy, can inhibit the proliferation of tumor cells, but it can also suppress the bone marrow hematopoietic system. Neutropenia caused by bone marrow suppression is a common dose-limiting toxicity after chemotherapy. Clinically, patients are usually discharged after completing chemotherapy and enter a 14 to 21-day outpatient interval. During this period, bone marrow function is in a fluctuating recovery state, and related complications have a certain time delay. With the development of hospital and laboratory information systems, existing remote follow-up systems typically collect blood routine and biochemical data from patients' follow-up examinations outside the hospital and make anomaly judgments based on unified reference ranges set by medical guidelines. When the test indicators are below or above preset thresholds, the system triggers an early warning and pushes the alarm information to the medical staff's terminals. This type of technical solution essentially uses a static threshold comparison mechanism to achieve anomaly identification; However, the monitoring mode based on static threshold judgment has obvious limitations in practical applications. First, such systems only focus on the absolute value of the indicator at a single point in time and do not calculate the rate of change and trend of the indicator over time. They cannot identify risk states that have not yet exceeded the threshold but have shown a continuous deterioration trend, thus resulting in insufficient early warning timeliness. Second, existing alarm push mechanisms usually do not consider the real-time load of medical resources. The triggering logic of each level of resources is relatively fixed and lacks a dynamic response threshold adjustment mechanism based on resource load rate, which may lead to uneven resource allocation during peak medical periods. To address the above problems, this invention proposes a solution. Summary of the Invention

[0003] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method for graded early warning and follow-up of tumor complications. This method constructs a risk momentum model by drawing on the concepts of velocity and acceleration in physical dynamics, calculates the rate and acceleration of changes in patient indicators in real time, and constructs a resource potential energy field by combining the real-time load rate of medical resources. Through dynamic game matching of risk momentum and resource potential energy, it achieves advanced perception of the deterioration trend of tumor complications and adaptive graded scheduling of medical resources, thereby solving the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: A method for graded early warning and follow-up of tumor complications includes the following steps: Step S1: At monitoring time t, collect the raw values ​​of the patient's objective indicators, and calculate the mean and standard deviation based on the data within the historical time window for standardization to construct the patient's current state vector; Step S2: Based on the standardized indicator values, calculate the directional change rate and deterioration acceleration of the indicators, and identify the deterioration trend by combining the directional coefficient; Step S3: Construct a risk momentum model based on the directional change rate and deterioration acceleration of the indicators, and calculate the risk momentum value of a single indicator by combining the indicator risk weight and trend enhancement term; Step S4: Aggregate the risk momentum of multi-dimensional indicators and integrate subjective symptom data to generate a comprehensive risk value for the patient; Step S5: Construct a resource load model, calculate the resource response threshold based on the real-time load rate of medical resources, wherein the resource response threshold increases non-linearly with the load rate; Step S6: Execute graded scheduling decisions, compare and match the patient's comprehensive risk value with the resource response thresholds at each level, and generate corresponding medical resource scheduling instructions; Step S7: After the follow-up, obtain the patient's true severity label, calculate the prediction error, and correct the parameters in the risk momentum model in reverse using the gradient descent method.

[0005] In a preferred embodiment, the directional change rate is calculated in step S2. The formula is: ;in, This represents the standardized value of the i-th index at time t. The sampling interval; This is a directional coefficient; a higher value indicates a higher risk. =1, when a lower indicator value represents a higher risk. =-1; by introducing To ensure that regardless of whether the indicators rise or fall, as long as they develop in a deteriorating direction, All are positive values.

[0006] In a preferred embodiment, step S3 involves calculating the single-index risk momentum value. The formula is: ;in, Let be the risk weight of the i-th indicator; Trend sensitivity coefficient; This is a trend enhancement term, and its calculation formula is: ,in This accelerates the deterioration of the indicators; For risk gating function, when hour =1, otherwise =0.

[0007] In a preferred embodiment, the indicator risk weight The construction method is as follows: ;in: This represents the inherent clinical weight of the i-th indicator; This represents the patient's individual vulnerability score, calculated based on the patient's ECOG score or age normalization. This represents the vulnerability amplification factor.

[0008] In a preferred embodiment, step S4 involves calculating the patient's overall risk value. The formula is: ;in, This represents the weighted sum of the risk momentum of all objective indicators; This represents the patient's subjective symptom score vector; The subjective rating mapping function is a linear mapping function or a piecewise linear function. This represents the weighting coefficient for subjective symptoms.

[0009] In a preferred embodiment, the resource response threshold is constructed in step S5. The formula is: Where j represents the resource level, This represents the real-time load rate of the j-th level resource at time t, with a value ranging from 0 to 1; Denotes the basic threshold constant for the j-th level resource, and different levels of resources satisfy... ; This represents the congestion sensitivity coefficient.

[0010] In a preferred embodiment, the hierarchical scheduling decision logic in step S6 is as follows: setting a safety margin as Iterate through all resource levels j and search for those that meet the conditions. maximum level If there exists a condition that satisfies the conditions Then generate the allocation of the first Scheduling instructions for level-one resources; if no matching conditions are found. If not, the scheduling command will not be triggered, and the monitoring status will be maintained.

[0011] In a preferred embodiment, the specific method for parameter correction in step S7 is as follows: defining a loss function. ,in The true severity labels were obtained from follow-up studies; gradient descent was used to assign risk weights to the indicators. and trend sensitivity coefficient Update: ; ;in This is the learning rate.

[0012] In a preferred embodiment, the tumor complication is chemotherapy-induced myelosuppression, and the objective indicators include at least the absolute neutrophil count (ANC) and the orientation coefficient. The value is -1.

[0013] In a preferred embodiment, a tumor complication grading early warning and follow-up system, used to execute the method according to any one of claims 1 to 9, includes: a data acquisition and standardization module for acquiring objective indicators and subjective symptom data of patients and performing standardization processing; a risk dynamic modeling module for calculating the directional change rate and deterioration acceleration of indicators, calculating risk momentum in combination with physical dynamics principles, and generating a comprehensive risk value for patients; a resource load modeling module for monitoring the medical resource load rate and dynamically calculating resource response thresholds using an exponential function; a tiered scheduling decision module for matching the comprehensive risk value with the resource response thresholds and generating scheduling instructions; and a parameter adaptive update module for calculating prediction errors based on the true severity labels and reversibly correcting the model parameters of the risk dynamic modeling module.

[0014] The technical effects and advantages of the present invention's method for grading, early warning, and follow-up of tumor complications are as follows: This invention standardizes objective indicators and calculates the directional rate of change and the acceleration of deterioration. It introduces a trend enhancement term to construct a risk momentum model, enabling risk assessment to simultaneously reflect the current state and trend of the indicators. Compared with anomaly identification methods based on static thresholds, this invention can identify a continuous deterioration trend before the indicators break through the critical value, thereby triggering early warnings and improving the timeliness of early warnings. This invention constructs an exponential resource response threshold function based on real-time resource load rate, which dynamically adjusts the resource response threshold as the load rate changes, thereby achieving a quantitative match between the patient's comprehensive risk value and the availability of medical resources. This mechanism avoids the resource congestion problem caused by fixed-level triggering and improves the efficiency of medical resource utilization and the rationality of scheduling. This invention constructs a loss function by introducing a true severity label and uses a gradient update mechanism to continuously correct the risk weights and trend sensitivity coefficients, so that the predicted risk value gradually approaches the true clinical severity, thereby achieving adaptive optimization of model parameters and improving the system's adaptability and prediction accuracy among different patients. Attached Figure Description

[0015] Figure 1 This is a block diagram of the overall structure of a tumor complication grading, early warning, and follow-up method according to the present invention.

[0016] Figure 2 This is a flowchart of a method for graded early warning and follow-up of tumor complications according to the present invention.

[0017] Figure 3This is a schematic diagram of the data acquisition and standardization module structure of the present invention.

[0018] Figure 4 This is a schematic diagram of the risk dynamic modeling module structure of the present invention.

[0019] Figure 5 This is a schematic diagram of the comprehensive risk formation process of the present invention.

[0020] Figure 6 This is a schematic diagram illustrating the relationship between the resource response threshold and load rate in this invention.

[0021] Figure 7 This is a flowchart of the hierarchical scheduling decision-making process of the present invention.

[0022] Figure 8 This is a flowchart illustrating the adaptive parameter update process of the present invention.

[0023] Figure 9 This is a timing diagram showing the matching of risk values ​​and resource thresholds in a specific example of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0025] Example 1, as Figure 1 As shown, the modules form a four-dimensional closed-loop structure of perception, cognition, game theory, and evolution through data flow and control flow. The risk output result also serves as the feedback input for the parameter update module, achieving continuous optimization. The system runs on a computer device, including a memory and a processor. The memory stores the computer program, and the processor executes the computer program to implement the functions of the following five core modules: The data acquisition and standardization module, used to construct digital holographic images of patients, specifically includes: The objective data unit is used to establish a communication connection with the hospital's LIS and HIS systems via the HL7 or FHIR standard interface, and to capture and analyze patients' blood routine, biochemical indicators and vital signs data in real time. The subjective data unit is used to provide an interactive interface to receive subjective symptom data such as pain scores and fatigue scores uploaded by patients; The standardization processing unit normalizes indicators of different dimensions to form a standardized objective indicator vector.

[0026] The risk dynamic modeling module is used to calculate risk values ​​based on the changing trends of indicators, and specifically includes: The indicator weight calculation unit is used to calculate the inherent clinical weight of the indicator. Patient individual vulnerability score and vulnerability amplification factor Calculate the risk weight of the i-th indicator. ; The rate of change calculation unit is used to set the sampling interval. And define the direction coefficient Calculate the directional rate of change for each indicator. and the rate of deterioration Define the risk gating function. ; Trend enhancement computing unit is used for deterioration acceleration Calculate trend enhancement terms The logic is ; The risk aggregation unit first combines the trend sensitivity coefficient. Construct a single-indicator risk value Then for all indicators The total risk momentum is obtained by weighted summation. Finally, the subjective symptom weighting coefficient is combined. and subjective rating mapping function Generate comprehensive risk score for patients .

[0027] The resource load modeling module is used for quantitative modeling of hospital resources, and specifically includes: The resource load monitoring unit is used to monitor the real-time load rate of the j-th level resource. Its value ranges from 0 to 1; The resource threshold calculation unit is used to calculate the basic threshold constant of the j-th level resource. and congestion sensitivity coefficient Construct a resource response threshold function and output the resource response threshold. .

[0028] The hierarchical scheduling decision module is used to perform game-theoretic matching. Its specific functions include: setting a safety margin. Receive patient comprehensive risk value With resource response thresholds at all levels Search for conditions maximum level If there exists a condition that satisfies the conditions If the resource allocation is successful, a scheduling instruction to allocate the corresponding resource will be generated; otherwise, monitoring will continue.

[0029] The parameter adaptive update module is used to adjust the model parameters based on the actual results, including: The actual severity acquisition unit is used to obtain the true severity label after the follow-up period ends. ; The loss function calculation unit is used to calculate the prediction error. and define the loss function. ; The gradient update unit is used to set the learning rate. The gradient is calculated using the gradient descent method, and the risk weights of the indicators in the risk dynamic modeling module are applied. and trend sensitivity coefficient A reverse correction is performed to make the risk momentum predicted by the system closer to the actual severity of the illness.

[0030] Example 2: This example provides a method for grading, early warning, and follow-up of tumor complications, such as... Figure 2 As shown: Step S1 is used to perform individualized and standardized modeling of the patient's multidimensional objective indicators at monitoring time t, such as... Figure 3 As shown, the system calculates the mean of each indicator based on the sampled data within the historical time window. with standard deviation This process is used to create a baseline model for each patient, and the currently collected raw measurements are standardized dimensionlessly to obtain a standardized objective index vector. This standardized objective index vector is then fused with the subjective rating vector to construct the patient's current state vector. This provides a unified data input basis for subsequent trend analysis and risk prediction; Step S2 is used to unify the trend direction and model the dynamic changes of standardized objective indicators. First, the directional coefficient is set according to the positive and negative correlation between the indicators and risks. This involves mapping indicators from different risk directions to a unified risk-oriented space, and then calculating the directional rate of change of each indicator. To characterize the rate at which the index deteriorates, the acceleration of deterioration is further calculated. This is used to characterize the degree of intensification of a deteriorating trend; at the same time, a risk gating function is constructed. It only participates in subsequent risk calculations when the indicator changes in a deteriorating direction, thereby suppressing the interference of invalid fluctuations and improving the accuracy of trend identification; Step S3 is used to construct a single-indicator risk quantification model that integrates individual vulnerability and trend enhancement mechanisms, first based on the inherent clinical weights of the indicators. And combined with the patient's individual vulnerability score Constructing dynamic risk weights This allows different individuals to exhibit differentiated risk responses under the same indicator changes, and then the positive component of the deterioration acceleration is extracted as a trend enhancement term. This is used to characterize the risk amplification trend. Finally, a gating function is used to filter out effective deterioration signals, and individualized weights, deterioration rates, and trend enhancement terms are fused to obtain a single-indicator risk value. This enables the joint quantification of risk intensity and risk development trends; Step S4 is used to summarize and merge the risk values ​​of each individual indicator, such as... Figure 5 As shown, and to construct the patient's current comprehensive risk value, the risk momentum corresponding to all objective indicators is first weighted and accumulated to obtain the patient's current total objective risk momentum. The subjective symptom scores were then converted into quantitative risk values ​​using a mapping function, and then weighted according to subjective weight coefficients. The patient's overall risk score is obtained by weighting and summing the results. This enables joint modeling of subjective and objective information, providing a unified decision-making basis for risk classification and early warning triggering; Step S5 is used to construct a dynamic response threshold model based on real-time resource load rate, such as... Figure 4 and Figure 6 As shown, the system sets up m-level medical resources and allocates incremental basic threshold constants for each level of resources. At the same time, real-time resource load rate is introduced. As a regulating factor, a resource response threshold is constructed using an exponential function. This allows the trigger thresholds of resources at all levels to be dynamically adjusted according to changes in the load rate. When the resource load rate increases, the response threshold is automatically raised, thereby achieving an adaptive match between the patient's risk level and the medical resource carrying capacity. Step S6 is used to compare and determine the patient's overall risk score with the response thresholds of various resource levels, such as... Figure 7 As shown, the system is configured with a safety margin. This is used to improve the stability of the judgment and to find the conditions that are met across all resource levels. maximum level If a suitable risk level exists, resources of that level will be allocated for intervention; otherwise, regular monitoring will be maintained. This mechanism enables graded matching and dynamic response between risk levels and resource levels. Step S7 is used to construct an error feedback optimization mechanism based on the true severity label, such as... Figure 8 As shown, the actual severity label of the patient was obtained after the follow-up period. Calculate the prediction error A squared loss function is constructed as the optimization objective; the risk weight parameters are adjusted using the gradient descent method. and trend sensitivity coefficient Iterative updates are performed to gradually bring the predicted risk momentum closer to the actual severity of the disease, thereby achieving adaptive optimization of model parameters and long-term accuracy improvement.

[0031] Furthermore, the specific steps include: Step S1: At any monitoring time t, collect the raw value of the i-th objective indicator. The mean and standard deviation were calculated based on patient data within a historical time window. ; ; Then, standardization is performed: ; in: This represents the original measurement value of the i-th index at time t; This represents the average value of the indicator over a historical window; This indicates the standard deviation of the indicator within the historical window; This represents the standardized dimensionless index value; N represents the number of samples within the historical window. Indicates the historical sampling time; Obtain the standardized objective index vector: The subjective rating vector is denoted as Construct the patient's current state vector: ; Step S2, set the sampling interval as... Define the direction coefficient If the higher the indicator, the higher the risk. If the lower the indicator, the higher the risk. ; Calculate the directional rate of change for each indicator. : By introducing To ensure that regardless of whether the indicators rise or fall, as long as they develop in a deteriorating direction, All are positive values; Further calculation of deterioration acceleration : ; Define the risk gating function: ; Step S3, first define the indicator weights: ;in: This represents the risk weight of the i-th indicator; This indicates the inherent clinical weight of the indicator; This represents the patient's individual vulnerability score, ranging from 0 to 1, calculated based on the ECOG score or age. Indicates the vulnerability amplification factor; Then define the trend enhancement term: ;in: This represents the trend enhancement value of the i-th indicator; Finally, a single-indicator risk value is constructed: ; in: This represents the risk value of the i-th indicator; Indicates the trend sensitivity coefficient; Step S4: Weighted summation of the risk momentum of all n indicators to obtain the patient's current total risk momentum. : ; After mapping subjective symptoms, the patient's overall risk score is obtained: ;in: This represents the subjective symptom weighting coefficient; This represents the subjective rating mapping function; Step S5, assume the system contains m-level resources and the resource load rate is... ; Construct a resource response threshold function: ;in: Let represent the resource base threshold constant of level j, and satisfy . ; Indicates the congestion sensitivity coefficient; This represents the real-time load rate of the j-th level resource; Step S6, set the safety margin as follows: Find the highest level that meets the conditions. : If there exists a condition that satisfies the conditions If the resource is available, allocate the corresponding resource; otherwise, continue monitoring. Step S7: Obtain the true severity label after the follow-up period ends. Define prediction error: Define the loss function: Gradient updates are performed on the weight parameters: ; ; Learning rate Through continuous iteration, the risk momentum predicted by the system can be made closer to the actual severity of the illness.

[0032] Furthermore, such as Figure 9 As shown, this embodiment uses the early warning of bone marrow suppression complications after lung cancer chemotherapy as an example to explain in detail the calculation process of the present invention. Scenario preset: Patient Information: Zhang San, 65 years old, ECOG score 2, with a history of severe infection; monitoring indicator: absolute neutrophil count (ANC); monitoring time point: As baseline, Day 3 after chemotherapy It was day 4 after chemotherapy; to simplify the calculation, we assume the patient's historical mean ANC. Standard deviation ; Step one: First, determine the direction factor. The lower the ANC, the more dangerous it is, as it is a negative indicator. Set the direction factor. Then calculate the indicator weight m and set the clinical baseline weight of ANC. This is a high-risk indicator used to calculate patient vulnerability. Set adjustment coefficient Calculate the indicator weights: ; Step 2, known time = 4.0、 time = 3.5, sampling interval Days; Calculation speed For the first monitoring point, the system defaults to an acceleration of 0, and calculates the acceleration. : Set the trend sensitivity coefficient. =2.0, calculate the patient's comprehensive risk score. ; Resource response threshold matching: The AI ​​customer service resource threshold is constant. 5.0; Specialist nurses are currently at 60% capacity, resource threshold. The attending physician is currently at 90% capacity, exceeding the resource threshold. Since 18.2 > 13.5 > 5.0, the system automatically triggers Level 1 intervention, and the AI ​​robot sends a reminder of precautions, without disturbing medical staff for the time being. Step 3, known time = 2.0; Calculation speed Calculate acceleration : ; Calculate the patient's overall risk score Nurses' workload rises to 80%, resource threshold reached. At this point, the doctor's workload drops to 70%, reaching the resource threshold. Since 40.3 > 31.5 > 22.2, the system directly sends an alert to the nurse in charge of the patient and displays a pop-up message on the nurse's workbench: Patient Zhang San's ANC is rapidly decreasing and is expected to fall below the threshold in 12 hours. Immediate manual intervention is recommended. Step 4, Assume After receiving the alert, the nurse manually followed up and found that the patient had a high fever of 39°C. The system then received the subjective symptom data. Adjust the overall risk value: Since 51.5 > 40.3, the system automatically locks the attending physician's appointment for the next day and notifies the patient via SMS that an expedited outpatient appointment for tomorrow has been booked for you. Step five: The patient was diagnosed with neutropenia-related fever the following day. Timely administration of white blood cell boosting injections brought the patient out of danger, and the actual severity was... Calculate the prediction error The results show that the error is extremely small, and the system maintains the current m and With parameters unchanged, the high-sensitivity model configuration for this patient was enhanced.

[0033] It should be noted that, for the sake of brevity, the foregoing method embodiments are described as a series of actions, but this does not mean that the application limits the order of the steps. Based on the ideas of this application, some steps can be executed in different orders or in parallel without affecting the functional implementation. Secondly, those skilled in the art should also understand that the specific embodiments described in the specification are preferred embodiments of the technical solutions of this application, and not limitations on the scope of protection of this application. All equivalent improvements or substitutions made within the spirit and principles of this application should be covered within the scope of protection of this application.

[0034] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for grading, early warning, and follow-up of tumor complications, characterized in that, Includes the following steps: Step S1: At monitoring time t, collect the raw values ​​of the patient's objective indicators, and calculate the mean and standard deviation based on the data within the historical time window for standardization processing to construct the patient's current state vector; Step S2: Based on the standardized index values, calculate the directional rate of change and the acceleration of deterioration of the index, and identify the deterioration trend by combining the directional coefficient; Step S3: Construct a risk momentum model based on the directional change rate and deterioration acceleration of the indicators, and calculate the risk momentum value of a single indicator by combining the indicator risk weights and trend enhancement terms. Step S4: Aggregate the risk momentum of multidimensional indicators and integrate subjective symptom data to generate a comprehensive risk value for the patient; Step S5: Construct a resource load model and calculate the resource response threshold based on the real-time load rate of medical resources. The resource response threshold increases non-linearly with the load rate. Step S6: Execute hierarchical scheduling decision-making, compare and match the patient's comprehensive risk value with the response thresholds of each level of resources, and generate corresponding medical resource scheduling instructions; Step S7: After the follow-up period, obtain the patient's true severity label, calculate the prediction error, and correct the parameters in the risk momentum model in reverse based on the gradient descent method.

2. The method for grading, early warning, and follow-up of tumor complications according to claim 1, characterized in that... In step S2, the directional rate of change is calculated. The formula is: ;in, This represents the standardized value of the i-th index at time t. The sampling interval; This is a directional coefficient; a higher value indicates a higher risk. =1, when a lower indicator value represents a higher risk. =-1; by introducing To ensure that regardless of whether the indicators rise or fall, as long as they develop in a deteriorating direction, All are positive values.

3. The method for grading, early warning, and follow-up of tumor complications according to claim 2, characterized in that, In step S3, the single-indicator risk momentum value is calculated. The formula is: ； in, Let be the risk weight of the i-th indicator; Trend sensitivity coefficient; This is a trend enhancement term, and its calculation formula is: ,in This accelerates the deterioration of the indicators; For risk gating function, when hour =1, otherwise =0.

4. The method for grading, early warning, and follow-up of tumor complications according to claim 3, characterized in that: Risk weight of the indicator The construction method is as follows: ； in: This represents the inherent clinical weight of the i-th indicator; This represents the patient's individual vulnerability score, calculated based on the patient's ECOG score or age normalization. This represents the vulnerability amplification factor.

5. The method for grading, early warning, and follow-up of tumor complications according to claim 1, characterized in that, In step S4, the patient's comprehensive risk value is calculated. The formula is: ； in, This represents the weighted sum of the risk momentum of all objective indicators; This represents the patient's subjective symptom score vector; The subjective rating mapping function is a linear mapping function or a piecewise linear function. This represents the weighting coefficient for subjective symptoms.

6. The method for grading, early warning, and follow-up of tumor complications according to claim 1, characterized in that, In step S5, a resource response threshold is constructed. The formula is: ； Where j represents the resource level, This represents the real-time load rate of the j-th level resource at time t, with a value ranging from 0 to 1; Denotes the basic threshold constant for the j-th level resource, and different levels of resources satisfy... ; This represents the congestion sensitivity coefficient.

7. The method for grading, early warning, and follow-up of tumor complications according to claim 6, characterized in that, The hierarchical scheduling decision logic in step S6 is as follows: Set the safety margin as... Iterate through all resource levels j and search for those that meet the conditions. maximum level If there exists a condition that satisfies the conditions Then generate the allocation of the first Scheduling instructions for level-one resources; if no matching conditions are found. If not, the scheduling command will not be triggered, and the monitoring status will be maintained.

8. The method for grading, early warning, and follow-up of tumor complications according to claim 3, characterized in that, The specific method for parameter correction in step S7 is as follows: Define loss function ,in For the true severity labels obtained during follow-up; Using gradient descent to assign risk weights to indicators and trend sensitivity coefficient Update: ； ； in This is the learning rate.

9. A method for grading, early warning, and follow-up of tumor complications according to claims 1 to 8, characterized in that, The tumor complication is bone marrow suppression following chemotherapy. The objective indicators include at least the absolute neutrophil count (ANC) and the orientation coefficient. The value is -1.

10. A tumor complication grading, early warning, and follow-up system, used to perform the method according to any one of claims 1 to 9, characterized in that, include: The data acquisition and standardization module is used to collect patients' objective indicators and subjective symptom data and perform standardization processing. The risk dynamic modeling module is used to calculate the directional rate of change and deterioration acceleration of indicators, and calculate the risk momentum by combining the principles of physical dynamics to generate the patient's comprehensive risk value. The resource load modeling module is used to monitor the load rate of medical resources and dynamically calculate the resource response threshold using an exponential function. The hierarchical scheduling decision module is used to match the comprehensive risk value with the resource response threshold and generate scheduling instructions; The parameter adaptive update module is used to calculate the prediction error based on the true severity label and then correct the model parameters of the risk dynamic modeling module.

Images