Infection risk grading early warning method and system based on clinical parameter dynamic fusion

Abstract

The invention discloses an infection risk grading early warning method and system based on clinical parameter dynamic fusion. The method comprises the following steps: acquiring patient data from a plurality of medical systems to obtain initial data; performing regularization preprocessing on the initial data to obtain a preprocessing result; quantifying the importance of the preprocessing result in combination with clinical expert opinions and a machine learning model, and generating a final risk assessment weight through dynamic weighted fusion; based on the final risk assessment weight and the knowledge graph, factors influencing the risk and weights of the factors are determined, and a dynamic assessment model used for calculating a risk score is established; and performing continuous monitoring by using the dynamic evaluation model, and automatically triggering an alarm when the risk score exceeds a threshold value. By implementing the method provided by the invention, the accuracy and reliability of infection judgment can be effectively improved, false alarms and missing alarms are reduced, and more timely and accurate infection risk early warning is realized.

Description

Technical Field

[0001] The present invention relates to a medical data processing method, and more specifically to an infection risk classification warning method and system based on dynamic fusion of clinical parameters. Background Art

[0002] Traditional infection early warning systems primarily rely on patients' medication records, pathogen culture results, and inflammatory markers to determine the presence of infection. However, this approach has limitations. First, medication is used not only to treat infection but also as a preventive measure; some pathogens may simply colonize rather than cause disease; and abnormal inflammatory markers may be caused by non-infectious factors. Therefore, relying solely on these single-dimensional data sets fails to accurately reflect a patient's true infection status. In recent years, a lung infectious disease prediction system based on multimodal data fusion has emerged. This system improves the accuracy and reliability of disease prediction by comprehensively analyzing multidimensional data such as clinical data, medical imaging, and environmental factors. It comprises a data acquisition module, a feature extraction module, a dynamic fusion module, and a predictive modeling module. Although this approach can provide relatively comprehensive information, it also has several significant drawbacks.

[0003] First, the effectiveness of the system is highly dependent on the quality and consistency of different data sources. Inconsistent standards between clinical and imaging data or differences in data collection methods may lead to inaccuracies in information fusion. Second, the system does not accurately consider the impact of medication behavior on infection prediction. Prophylactic and therapeutic medications are not treated differently, making it easy to misjudge prophylactic medication as infection treatment, thereby increasing the possibility of false positives. Furthermore, when dealing with the relationship between pathogens and infections, the system does not consider that pathogens may be harmless colonizing bacteria. This may lead to harmless bacteria being mistaken for pathogenic bacteria, triggering unnecessary warnings. Finally, although time series analysis can help capture long-term trends in diseases, the system may not respond quickly enough to patients with sudden acute infections or rapidly changing conditions, and may not be able to issue timely warnings, affecting the actual application effect.

[0004] In summary, although the existing multimodal data fusion prediction system has improved the comprehensiveness and efficiency of disease prediction, there is still room for improvement in the consistency of data fusion, accurate consideration of medication behavior, correct identification of pathogens, and timeliness of responding to emergencies.

[0005] Therefore, it is necessary to design a new method to effectively improve the accuracy and reliability of infection judgment, reduce false positives and missed reports, and achieve more timely and accurate infection risk warnings. Summary of the Invention

[0006] The purpose of the present invention is to overcome the defects of the prior art and provide an infection risk classification warning method and system based on dynamic fusion of clinical parameters.

[0007] To achieve the above objectives, the present invention adopts the following technical solutions: an infection risk classification warning method based on dynamic fusion of clinical parameters, comprising:

[0008] Obtain patient data from multiple healthcare systems to obtain initial data;

[0009] Performing regularized preprocessing on the initial data to obtain a preprocessing result;

[0010] Combining clinical expert opinions and machine learning models to quantify the importance of the pre-processing results, and generating final risk assessment weights through dynamic weighted fusion;

[0011] Based on the final risk assessment weights and knowledge graph, determine the factors affecting the risk and their weights, and establish a dynamic assessment model for calculating the risk score;

[0012] The dynamic assessment model is used for continuous monitoring and automatically triggers alerts when the risk score exceeds a threshold.

[0013] A further technical solution is: after continuously monitoring using the dynamic assessment model and automatically triggering an alarm when the risk score exceeds a threshold, it also includes:

[0014] Adjust and optimize the dynamic evaluation model.

[0015] A further technical solution is: performing regularized preprocessing on the initial data to obtain a preprocessing result, including:

[0016] The numerical data of the initial data is filled with the median of historical data of similar patients, and the categorical data is marked as unknown. The data range of the initial data is checked according to medical common sense to exclude unreasonable values. The validity of the categorical features of the initial data is confirmed through the established infection-related knowledge graph rule base to obtain an intermediate result;

[0017] The continuous medical parameters in the intermediate results are converted into classification labels with clear medical significance, and the surgical effectiveness and microbial inspection type characteristics are verified and mapped to standardized medical entities through clinical knowledge graphs to obtain preprocessing results.

[0018] A further technical solution is to combine clinical expert opinions and machine learning models to quantify the importance of the pre-processing results and generate the final risk assessment weights through dynamic weighted fusion, including:

[0019] Assign initial weights to high-risk data elements based on clinical guidelines;

[0020] Use XGBoost to train a binary classification model and quantify the importance of the preprocessing results to infection prediction using SHAP values ​​to obtain quantitative results;

[0021] The quantification results are linearly superimposed in a fixed proportion in combination with the expert rule weights and the model weights to generate the final dynamic weights. The data elements whose final dynamic weights meet the requirements are retained through threshold filtering to obtain the final risk assessment weights.

[0022] A further technical solution is: combining the expert rule weights and the model weights to linearly superimpose the quantification results in a fixed ratio to generate a final dynamic weight, and retaining the data elements whose final dynamic weights meet the requirements through threshold filtering to obtain the final risk assessment weight, including:

[0023] The final dynamic weight is generated by linearly superimposing the quantified results in a fixed proportion in combination with the expert rule weight and the model weight. The data elements whose final dynamic weight meets the requirements are retained through threshold filtering, and the proportion of the expert rule weight is adjusted online to determine the final risk assessment weight.

[0024] A further technical solution is: based on the final risk assessment weight and knowledge graph, determining the factors affecting the risk and their weights, and establishing a dynamic assessment model for calculating the risk score, including:

[0025] Based on the final risk assessment weight, the specific threshold for each data element to trigger different risk levels is determined, and the data elements that can trigger specific rules are determined. Knowledge graph technology is used to verify the validity and mutual exclusivity of the combination conditions to obtain a dynamic assessment model for calculating risk scores, and the rule base is automatically adjusted according to the dynamic assessment model updated monthly.

[0026] A further technical solution is: the dynamic assessment model is used for continuous monitoring, and an alarm is automatically triggered when the risk score exceeds a threshold, including:

[0027] Determining the final sum of the risk assessment weights of the data elements of the current trigger condition using the dynamic assessment model, and calculating the patient's risk score based on the final sum of the risk assessment weights;

[0028] Based on the risk level determined by the patient's risk score, alerts are sent through different channels.

[0029] A further technical solution is as follows: the adjusting and optimizing the dynamic evaluation model includes:

[0030] Every morning, new data is used to update the dynamic assessment model and adjust the final risk assessment weights involved. At the same time, inefficient rules are eliminated or new rules are launched through expert review and knowledge graph verification.

[0031] The present invention also provides an infection risk classification warning system based on dynamic fusion of clinical parameters, including:

[0032] an acquisition unit, configured to acquire patient data from a plurality of medical systems to obtain initial data;

[0033] A preprocessing unit, configured to perform regularized preprocessing on the initial data to obtain a preprocessing result;

[0034] a weight generation unit, configured to combine clinical expert opinions and a machine learning model to quantify the importance of the preprocessing results and generate a final risk assessment weight through dynamic weighted fusion;

[0035] A model generation unit, configured to determine the factors affecting the risk and their weights based on the final risk assessment weights and the knowledge graph, and to establish a dynamic assessment model for calculating the risk score;

[0036] A monitoring unit is used to continuously monitor using the dynamic assessment model and automatically trigger an alarm when the risk score exceeds a threshold.

[0037] Its further technical solution is: also includes:

[0038] An adjustment unit is used to adjust and optimize the dynamic evaluation model.

[0039] The beneficial effects of the present invention compared to the existing technology are as follows: the present invention first ensures the integrity and consistency of the data by integrating patient data from multiple medical systems and performing regularized preprocessing; combines the opinions of clinical experts with machine learning models to quantitatively analyze the importance of these data, and generates the final risk assessment weights through a dynamic weighted fusion method, thereby achieving an effective combination of medical experience and data analysis; based on this assessment weight and knowledge graph, the key factors affecting the risk of infection and their relative importance can be accurately identified, and a dynamically adjusted assessment model can be established to calculate an individualized risk score. The model can continuously monitor the health status of the patient, automatically trigger an alarm when it detects that the risk score exceeds the set threshold, and promptly notify medical staff to take measures. This method not only improves the accuracy and reliability of infection judgment, but also effectively reduces false positives and missed reports, achieving more timely and accurate infection risk warnings, and providing strong support for clinical decision-making.

[0040] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0042] Figure 1 A schematic diagram of an application scenario of the infection risk classification and early warning method based on dynamic fusion of clinical parameters provided by an embodiment of the present invention;

[0043] Figure 2 A schematic diagram of a flow chart of an infection risk classification and early warning method based on dynamic fusion of clinical parameters provided by an embodiment of the present invention;

[0044] Figure 3 A schematic block diagram of an infection risk grading and early warning system based on dynamic fusion of clinical parameters provided by an embodiment of the present invention;

[0045] Figure 4 A schematic block diagram of a computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0047] It will be understood that when used in this specification and the appended claims, the terms “comprises” and “comprising” indicate the presence of described features, integers, steps, operations, elements and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or groups thereof.

[0048] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the present invention. As used in the specification and appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms unless the context clearly indicates otherwise.

[0049] It should be further understood that the term "and / or" used in the present description and the appended claims refers to and includes any and all possible combinations of one or more of the associated listed items.

[0050] See also Figure 1 and Figure 2 , Figure 1 Schematic diagram of an application scenario of the infection risk classification and early warning method based on dynamic fusion of clinical parameters provided by an embodiment of the present invention. Figure 2 A schematic flow chart of an infection risk grading and early warning method based on dynamic fusion of clinical parameters, provided in an embodiment of the present invention, is provided. This infection risk grading and early warning method, based on dynamic fusion of clinical parameters, is applied to a server that interacts with a terminal. By integrating multi-source patient data and performing regularized preprocessing, it generates risk assessment weights using dynamic weighted fusion of clinical expert opinions and machine learning models. A dynamic assessment model for calculating risk scores is established based on a knowledge graph, achieving precise quantification of infection risk and graded early warning. It combines historical data infilling, medical knowledge verification, and clinical knowledge graph validation to ensure data validity and accuracy. It also employs an XGBoost model and SHAP value analysis to quantify data importance, further optimizing the risk assessment process. Through continuous monitoring and an automatic alert mechanism, relevant personnel are immediately notified when the risk score exceeds the threshold, and the model is continuously adjusted and optimized based on feedback, effectively improving the accuracy and reliability of infection judgments, reducing false positives and missed negatives, and achieving more timely and accurate infection risk warnings. Furthermore, regular updates to the model and rule base ensure the adaptability and advancement of the early warning system.

[0051] Figure 2 FIG. 1 is a flow chart of an infection risk classification warning method based on dynamic fusion of clinical parameters provided by an embodiment of the present invention. Figure 2 As shown, the method includes the following steps S110 to S160.

[0052] S110 , acquiring patient data from multiple medical systems to obtain initial data.

[0053] In this embodiment, initial data refers to unprocessed original patient information collected from multiple medical systems, including medical records, test results, medication history, and surgical details, etc., which provides a basis for subsequent data analysis and clinical decision-making.

[0054] Specifically, first, it is necessary to determine which medical systems will serve as data sources and ensure that these systems can stably provide the required information. Based on the above description, the main data sources include:

[0055] Hospital Information System (HIS): used to obtain patients' personal information, diagnosis records, surgical history, medication history, etc.

[0056] Laboratory Information System (LIS): Mainly used to collect key laboratory data such as blood test results, microbial culture results and other biochemical indicators.

[0057] Electronic medical records (EMR): Provides rich clinical information, such as disease progression, nursing details, and imaging reports.

[0058] Hand anesthesia system: specially designed to record the anesthesia process and surgical details.

[0059] By using efficient data middleware technology, real-time extraction of data from various sources can be achieved. This approach not only ensures the timeliness of data, but also effectively addresses the issue of inconsistent data formats between systems. For example:

[0060] Structured data (such as numerical or categorical data) can be directly extracted and integrated into a unified data warehouse.

[0061] For unstructured text data (such as imaging reports), natural language processing (NLP) technology is needed to parse and extract key medical entities and conclusions, such as descriptions such as "pulmonary inflammatory infiltration".

[0062] After completing initial data collection, these raw data need to be standardized to facilitate subsequent analysis. This may involve multiple steps, such as unit conversion, missing value imputation, and outlier detection. Furthermore, based on medical expertise and rule bases, the validity and rationality of the data can be further verified to eliminate obviously erroneous or unreasonable information.

[0063] In summary, the core of S110 is to build an information network that comprehensively covers the patient diagnosis and treatment process. By integrating multiple data sources, adopting advanced data processing technology and strict verification processes, it ensures that the collected data is both rich and reliable, thereby supporting the development of more accurate clinical decision support systems.

[0064] S120: Perform regularized preprocessing on the initial data to obtain a preprocessing result.

[0065] In this embodiment, the preprocessing result refers to a data set with high consistency, good accuracy and clear clinical significance generated by filling missing values, filtering outliers, validating the initial data based on the medical knowledge graph, and converting features into regularized forms.

[0066] In one embodiment, the aforementioned step S120 may include steps S121 - S122 .

[0067] S121. The numerical data of the initial data is filled with the median of historical data of similar patients, and the categorical data is marked as unknown. The data range of the initial data is checked based on medical common sense to exclude unreasonable values. The validity of the categorical features of the initial data is confirmed through the established infection-related knowledge graph rule base to obtain an intermediate result.

[0068] In this embodiment, the intermediate result refers to a phased data set generated after preliminary analysis and feature extraction in the data processing process, which is used before further complex analysis or model training. The data set has been cleaned and transformed as necessary, but the final algorithm or decision logic has not yet been applied.

[0069] First, the numerical features in the initial data are filled with the median of historical data for similar patients, while categorical features are marked as "unknown." Next, the data range is verified based on medical knowledge to exclude unreasonable values ​​(such as body temperature outside the normal human range). Finally, the validity of the categorical features in the initial data is confirmed through the established infection-related knowledge graph rule base to ensure that they conform to clinical reality (for example, "surgery" records must meet specific conditions), thereby generating intermediate results.

[0070] S122. Convert the continuous medical parameters in the intermediate results into classification labels with clear medical significance, verify and map the surgical effectiveness and microbial inspection type characteristics to standardized medical entities through clinical knowledge graphs to obtain preprocessing results.

[0071] Secondly, based on the intermediate results, the continuous medical parameters are further converted into classification labels. This includes converting continuous features into classification labels with clear medical significance (such as body temperature classification and abnormal white blood cell classification) according to clinical guidelines and expert consensus, so as to retain clinical semantic information and avoid information loss caused by pure statistical standardization. At the same time, the clinical knowledge graph is used to encode categorical features (such as surgical effectiveness and microbial inspection type) and map them to standardized medical entities to ensure the consistency and accuracy of the data, and ultimately form preprocessing results that can be used for subsequent analysis and decision support. This series of operations ensures the integrity, accuracy and practicality of the data, laying the foundation for high-quality clinical data analysis.

[0072] In this example, to ensure the integrity and accuracy of the data and provide a high-quality data foundation for subsequent analysis, the initial data was processed as follows:

[0073] Missing value processing: For numerical features, the median of historical data of similar patients is used to fill in the missing values ​​to maintain the data distribution characteristics; for categorical features, they are uniformly marked as "unknown" categories.

[0074] Outlier filtering: Based on medical knowledge, the data range is verified and data points that are obviously inconsistent with the actual situation are removed. For example, records with body temperatures outside the physiological limit will be considered invalid.

[0075] Knowledge graph-driven validity verification: A knowledge graph rule base for infection-related situations is built to check the validity of categorical features. For example, "surgery" records must exist in the surgery master record or anesthesia record and cannot be marked as invalid surgery.

[0076] Convert raw medical data into a form with clear clinical meaning to facilitate further analysis and model training.

[0077] According to clinical guidelines such as the Expert Consensus on Emergency Diagnosis and Treatment of Heat Stroke, Body temperature is graded. For example, a temperature below 38°C is considered normal, 38°C to 39°C is considered fever, and 39°C or higher is considered hyperpyrexia.

[0078] The number of white blood cells (WBC) was classified according to the NCCLS standard. Determine whether it is within the normal range, slightly abnormal, or severely abnormal, so as to identify potential local or systemic infection. Among them, Mild Abnormal may indicate local infection or inflammation; Severe Abnormal is highly suggestive of systemic infection or blood system disease.

[0079] Other biochemical indicators, such as urine leukocyte count and neutrophil percentage, are also classified according to specific standards to reflect different health statuses. For example, urine leukocyte count is: UrineWBCFull = I (urine sediment microscopy ≥ 50 cells / HP); abnormal neutrophil percentage is: NeutrophilAbnormal = I (percentage > 75% ∨ < 40%).

[0080] The clinical knowledge graph is used to verify categorical features and map them to standardized medical entities. For example, a surgical record must meet certain conditions to be considered valid, that is, the surgery must be in the main surgical record or in the hand numbness record and must not be an invalid surgery.

[0081] The classification of microbiological inspection types and antibiotic change types also relies on specific standards and logical judgment to ensure data accuracy.

[0082] Monitoring detected key fields such as sputum culture and urine culture sent for examination;

[0083] Classification of antibiotic change type:

[0084] Medication record cleaning: exclude temporary medical orders (such as single-dose medication) and preventive medication (marked by knowledge graph)

[0085] Continuous feature segmentation, direct segmentation based on medical standards (such as body temperature ≥39℃ is "high fever", blood white blood cell count ≥10×10 9 / L indicates abnormality).

[0086] This process not only improves data quality but also makes it more relevant to clinical application scenarios, providing solid data support for precision medicine. Furthermore, this processing method helps preserve important clinical semantic information and avoids the potential loss of information that can occur with simple statistical standardization.

[0087] S130. Combining clinical expert opinions and machine learning models to quantify the importance of the pretreatment results, and generating final risk assessment weights through dynamic weighted fusion.

[0088] In this embodiment, the final risk assessment weight refers to a comprehensive score that reflects the contribution of each data element to the infection risk, which is generated by linearly superimposing the importance quantification results of the clinical expert opinions and the machine learning model in a fixed proportion.

[0089] In one embodiment, the aforementioned step S130 may include steps S131 to S133 .

[0090] S131. Assign initial weights to high-risk data elements based on clinical guidelines.

[0091] In this example, at this stage, data elements considered to have a higher risk of infection are first assigned a preliminary weight based on existing clinical guidelines and expert consensus. These weights reflect experience and knowledge in the medical field and ensure that the basic risk assessment is consistent with medical logic. For example:

[0092] White blood cells>15×10 9 / L: Assigned a weight of 1.5 because a highly abnormal white blood cell count often indicates a bacterial infection.

[0093] Antibiotic change: Assigned a weight of 1.0, indicating that the treatment regimen was adjusted to address a new or unrecognized infection.

[0094] S132. Use XGBoost to train a binary classification model and quantify the importance of the preprocessing result to infection prediction through SHAP value to obtain a quantitative result.

[0095] In this embodiment, the quantitative result refers to the specific numerical value of the importance of each pre-processed data element for distinguishing between infection and non-infection, which is calculated by using the XGBoost model and the SHAP value.

[0096] Specifically, we next used machine learning methods to further enhance risk assessment capabilities. Specifically, we used the XGBoost algorithm to train a binary classification model to distinguish between infected and non-infected cases. We then used SHAP (SHapley Additive exPlanations) values ​​to quantify the contribution of each feature to the prediction. This step can capture new risk patterns or atypical infection scenarios that are not covered by expert rules.

[0097] XGBoost model: Utilizes its powerful nonlinear modeling capabilities and combines L2 regularization to prevent overfitting and improve model generalization capabilities.

[0098] SHAP value calculation: Based on the principles of cooperative game theory, it accurately measures the average marginal contribution of each feature to the model output, thereby obtaining a specific importance score for each feature.

[0099] S133. Combine the expert rule weight and the model weight to linearly superimpose the quantification results in a fixed ratio to generate a final dynamic weight, and retain the data elements whose final dynamic weight meets the requirements through threshold filtering to obtain the final risk assessment weight.

[0100] In this embodiment, the quantification results are linearly superimposed at a fixed ratio in combination with the expert rule weights and the model weights to generate the final dynamic weights. The data elements whose final dynamic weights meet the requirements are retained through threshold filtering, and the proportion of the expert rule weights is adjusted online to determine the final risk assessment weights.

[0101] Specifically, after obtaining the expert rule weights and machine learning model weights, they are linearly superimposed at a fixed ratio to generate the final dynamic weights for the data elements. By default, the expert rule weights account for 60% and the model weights for 40%. This combination takes into account the importance of clinical experience while also accommodating the flexibility and adaptability of a data-driven approach.

[0102] Dynamic weight formula: Where α is the ratio of the expert rule weight, and the default value is 0.6.

[0103] Only data elements with a final weight greater than or equal to 0.3 are retained, and those features with low contribution to risk assessment are removed.

[0104] The α value can be flexibly adjusted on the management interface according to actual application conditions and the needs of different departments to optimize the accuracy of risk assessment.

[0105] Through the above steps, the conversion process from raw medical data to clinically meaningful risk assessment weights is achieved, which not only improves the scientific nature and reliability of risk assessment, but also enhances its adaptability to different application scenarios. In addition, the transparent final weight formula makes it easier for clinicians to understand the relationship between each data element and infection risk, which helps to develop more accurate and effective diagnosis and treatment strategies.

[0106] In this example, based on clinical guidelines and expert consensus, initial weights are assigned to high-risk data elements to ensure that the basic risk assessment conforms to medical logic. For example, blood white blood cell count > 15×10 9 The initial weight of the serum creatinine level / L was set to 1.5, as its high abnormality indicates bacterial infection; the antibiotic change was given a weight of 1.0 to reflect the need for treatment adjustment.

[0107] XGBoost is used to train a binary classification model (distinguishing between infection and non-infection). The importance of features is quantified through a data-driven approach to capture risk patterns not covered by experts, such as new pathogens or atypical infections. The model is optimized using the L(θ) formula, and the contribution of each feature to the prediction result is calculated using the SHAP value to obtain the weight of the machine learning model. y i ∈{0,1}, represents the sample label (1 = infected, 0 = non-infected); represents the infection probability predicted by the model; f(x i ; θ), represents the output of the XGBoost tree model (accumulating the prediction results of multiple trees); λ represents the L2 regularization coefficient to prevent overfitting.

[0108] The contribution of each feature to the prediction result is quantified by the SHAP (SHapley Additive exPlanations) value:

[0109] Based on cooperative interaction theory, it measures the average marginal contribution of feature j to the model output among all feature subsets.

[0110] The specific calculation is approximated by the following formula: F is the set of all features; S is the subset without feature j; f(S) is the model output using subset S.

[0111] The SHAP value, based on cooperative game theory, measures the average marginal contribution of a feature to the model's output and is calculated using an approximate formula. Ultimately, the machine learning model enhancement process converts the importance of these features into specific weights.

[0112] The final dynamic weights are generated by linearly adding expert rule weights and model weights in a fixed ratio. By default, α is set to 0.6, meaning that expert opinion accounts for 60% and model weights account for 40%. This strategy aims to combine clinical experience and data-driven approaches to ensure that the final weights are both consistent with medical consensus and adaptive.

[0113] Specifically, the expert rule weight and the model weight are linearly added in a fixed ratio to generate the final dynamic weight:

[0114] α is the weight of expert rules (default 60%), reflecting the dominance of clinical experience;

[0115] 1-α is the model weight ratio (default 40%), which enhances data-driven adaptability;

[0116] α∈[0,1] guarantees exist Avoid extreme weights within the range.

[0117] Linear superposition intuitively reflects the contribution ratio of experts and data, which is easier for clinicians to understand.

[0118] The value of α ranges from [0, 1], ensuring that the final weights are reasonably distributed between the expert weights and the model weights. For features not covered by the expert rules, the model weights are fully relied upon. When conflicts arise between the model weights and the expert rules, a manual review process is triggered for adjustment.

[0119] A threshold filtering method was applied to retain data elements with a final weight ≥ 0.3, while low-contribution features were excluded. This step combined expert-led and data-driven approaches to ensure that weight assignments were consistent with medical consensus and captured potential risk patterns. Model weights were periodically updated with new data, supporting online adjustment of the α value to accommodate the needs of different departments.

[0120] When model weights conflict with expert rules:

[0121] This triggers a manual review process, where clinical experts confirm whether to adjust α or modify the rule. An example calculation is shown in Table 1.

[0122] Table 1. Calculation examples

[0123]

[0124] The data elements with a final weight ≥ 0.3 are retained, and low-contribution features are excluded. Retention conditions:

[0125] Dual-driven fusion mechanism: Expert rule-driven (60%): ensures that weight distribution is consistent with medical consensus (e.g., sterile culture positive has the highest weight). Data-driven supplement (40%): captures potential risk patterns (e.g., the implicit association between special inspection behaviors and infection).

[0126] Model weights are updated periodically with new data (e.g. monthly incremental training) to continuously optimize predictive capabilities.

[0127] The alpha value (through the management interface) is adapted to the needs of different departments. The SHAP value provides a global and individual interpretation of the feature contribution. The final weight formula is transparent and can be directly associated with clinical semantics (e.g., "high fever = 1.12" corresponds to sepsis risk).

[0128] The entire process directly links clinical semantics through a transparent final weight formula, making complex mathematical models easy for clinicians to understand. For example, "high fever = 1.12" corresponds to the risk of sepsis. Furthermore, SHAP values ​​provide global and individual-level explanations, enhancing the model's interpretability and practicality.

[0129] S140. Based on the final risk assessment weight and knowledge graph, determine the factors affecting the risk and their weights, and establish a dynamic assessment model for calculating the risk score.

[0130] In this embodiment, the dynamic evaluation model aims to evaluate the final weight of each data element. The model uses knowledge graph technology to verify the validity and mutual exclusivity of combined conditions, thereby building a system for calculating risk scores.

[0131] Specifically, the specific threshold for each data element to trigger different risk levels is determined based on the final risk assessment weight, and the data element that can trigger a specific rule is determined. The knowledge graph technology is used to verify the validity and mutual exclusivity of the combination conditions to obtain a dynamic assessment model for calculating the risk score, and the rule base is automatically adjusted according to the dynamic assessment model updated monthly.

[0132] Weight and rule binding

[0133] Single condition trigger: If the final weight of a data element If the set weight threshold is reached or exceeded, a risk warning of the corresponding level will be directly triggered.

[0134] Combination condition trigger: when the sum of the final weights of multiple data elements ∑ When the corresponding weight and threshold are reached or exceeded, a risk warning of the corresponding level will also be triggered.

[0135] According to the final weight of the output Determine which data elements can participate in rule generation (for example, only data elements with weight ≥ 0.5 participate).

[0136] The model automatically triggers the regeneration of the rule base after updating the weights every month, ensuring that emerging risk patterns (such as drug-resistant bacteria characteristics) are taken into consideration in a timely manner.

[0137] Use the Cypher query language to verify the validity of a combination of conditions by matching relationships between nodes such as imaging descriptions and antibiotic changes. For example, MATCH(img:Imaging {type:"Infection Description"})-[:Related Treatments]->(ant:Antibiotic Changes) WHERE EXISTS(img)-[:Supporting Diagnosis]->(inf:Infection) RETURN COUNT>0 AS valid. If valid = True is returned, the combination is allowed to trigger the rule.

[0138] Cypher queries are also used to detect potential mutually exclusive relationships to prevent false triggering of rules due to multiple anomalies for the same patient. For example, MATCH (f: fracture) - [: exclude association] -> (u: urine white blood cell abnormality) WHERE f.patient_id = u.patient_id RETURN COUNT > 0 AS conflict. If a conflict exists, further manual review or rule adjustment is required.

[0139] Based on the above logic, four levels of risk assessment standards are defined:

[0140] Level 4: Applicable to highly abnormal / critical events, single data element weight or multiple data element weights and It can be triggered.

[0141] Level 3: Involving inspection and medication adjustment, requiring a single data element weight or multiple data element weights and

[0142] Level 2: For cases with minor abnormalities, a single data element weight is required or multiple data element weights and

[0143] Level 1: covers single mild abnormalities, inspection or medication changes, requiring only a single data element weight or multiple data element weights and

[0144] The dynamic assessment model not only considers the importance of individual data elements but also focuses on the interactions between them. It can be regularly updated based on the latest clinical guidelines and expert consensus. This flexibility enables the model to maintain efficiency and accuracy in an ever-changing healthcare environment, while providing clinicians with a scientific basis for more precise diagnostic decisions. This approach achieves an integrated process from data collection and analysis to risk assessment, significantly improving the capabilities and efficiency of infection risk management.

[0145] Specifically, the rule triggering logic is as follows:

[0146] Single condition trigger: If the weight of a data element Level threshold, directly trigger the corresponding level.

[0147] Combined condition trigger: If the weights of multiple data elements and Level threshold, the corresponding level is triggered.

[0148] Output weight Determines whether a data element can trigger a rule (e.g., only elements with a weight ≥ 0.5 will participate in rule generation).

[0149] According to the final weight of the output data element The model determines which data elements are eligible for rule generation. For example, only data elements with a weight of 0.5 or greater are considered for rule generation. Furthermore, monthly weight updates automatically trigger the regeneration of the rule base, ensuring that emerging risk patterns (such as drug-resistant bacterial signatures) are promptly considered. Table 2 shows the specific risk levels, trigger condition types, and corresponding weight thresholds.

[0150] Table 2. Triggering rules

[0151]

[0152] Level 4 rule: Single condition with high weight: such as "positive sterile culture (weight = 2.0)" is directly triggered, reflecting absolute criticality.

[0153] The sum of the combined weights, such as "fever (1.2) + antibiotic change (1.8) = 3.0," meets the sepsis warning criteria.

[0154] Level 3 rule: The sum of the weights of inspection and medication: For example, "sputum culture inspection (0.9) + antibiotic change (1.0) = 1.9 ≥ 2.0" needs to meet the additional knowledge graph relevance (such as respiratory infection inspection) to be triggered.

[0155] A high weight of a single factor can trigger a fourth-level warning. If the weight of a single factor is not enough to trigger a fourth-level warning, the warning threshold can be reached by combining the weights with other relevant factors, but the combined factors need to meet the dual verification of the knowledge graph.

[0156] Combination validity check:

[0157] MATCH(img:imaging\{type:"infection description"})-[:associated treatment]->(ant:antibiotic change);

[0158] WHERE EXISTS (img)-[:supporting diagnosis]->(inf:infection);

[0159] RETURN COUNT>0AS valid;

[0160] If valid=True is returned, the combination is allowed to trigger the rule.

[0161] Mutually exclusive filtering: MATCH (f: fracture) - [: exclude association] - > (u: abnormal urine leukocytes);

[0162] WHERE f.patient_id=u.patient_id;

[0163] RETURN COUNT>0 AS conflict;

[0164] If there is an associated path in the knowledge graph (such as high fever-sepsis-level 4 risk), separate triggering is allowed.

[0165] If level 4 (positive blood culture) and level 3 (sputum culture + medication) are triggered at the same time, the final judgment will be level 4 according to the level coverage principle.

[0166] After data collection, it is standardized and classified, and the data is analyzed using machine learning algorithms to calculate the weight of each data element. The data elements with infection risks are mainly classified into abnormal indicators, inspection, and antibiotic use.

[0167] Among them, abnormal indicators are divided into mild abnormalities and severe abnormalities according to their severity, and are divided into physical signs, biochemistry, microbiology, diagnosis, and other clinical data according to their types. Data elements include: physical signs (fever, high fever), biochemistry (blood white blood cells, blood neutrophils, urine white blood cells, fecal white blood cells, cerebrospinal fluid white blood cells, joint fluid white blood cells, pleural effusion white blood cells, ascites white blood cells), microbiology (urine culture, sputum culture, stool culture, skin culture, other specimen culture), diagnosis (influenza A, influenza B, new museum, vaginitis, infection, otitis, Mycoplasma pneumonia, hepatitis A, hepatitis B, biscuits, Treponema pallidum, alanine aminotransferase, hepatitis A antibody, hepatitis B antibody, hepatitis C antibody, Treponema pallidum TPPA, Helicobacter pylori), other clinical data (pediatrics, surgery, delivery surgery, imaging infection description, imaging inflammation description, imaging request, fracture).

[0168] The types of samples submitted for examination are divided into microbiological examination and biochemical examination. Biochemical examination (blood leukocytes, blood neutrophils, urine leukocytes, fecal leukocytes, cerebrospinal fluid leukocytes, joint fluid leukocytes, pleural effusion leukocytes, ascites leukocytes) and microbiological examination (urine culture, sputum culture, fecal culture, skin culture, and other specimen cultures) are available.

[0169] Antibiotic use is categorized by grade as either antibiotics or high-grade antibiotics. This is primarily described as an antibiotic change. An antibiotic change refers to a change in a patient's antibiotic use (e.g., a new antibiotic or a replacement antibiotic).

[0170] The specific conditions for determining the fourth level of risk are:

[0171] Level 1 risk: laboratory microbiological testing, abnormal body temperature, abnormal erythrocyte sedimentation rate, abnormal interleukin-6, abnormal lymphocytes, abnormal eosinophils, abnormal procalcitonin, abnormal C-reactive protein, abnormal blood white blood cells, abnormal blood neutrophils, description of inflammation on imaging examinations, abnormal urine white blood cells, abnormal stool white blood cells, abnormal cerebrospinal fluid white blood cells, abnormal joint fluid white blood cells, abnormal pleural effusion white blood cells, abnormal ascites white blood cells, and changes in non-prophylactic antibiotics.

[0172] Level 2 risk: abnormal body temperature of microorganisms submitted for examination, abnormal erythrocyte sedimentation rate of microorganisms submitted for examination, abnormal interleukin-6 of microorganisms submitted for examination, abnormal lymphocytes of microorganisms submitted for examination, abnormal eosinophils of microorganisms submitted for examination, abnormal procalcitonin of microorganisms submitted for examination, abnormal C-reactive protein of microorganisms submitted for examination, abnormal white blood cells of microorganisms submitted for examination, abnormal neutrophils of microorganisms submitted for examination, abnormal white blood cells of pleural effusion + pleural effusion or sputum submitted for examination, abnormal white blood cells of ascites + ascites submitted for examination, abnormal white blood cells of stool + stool submitted for examination, abnormal white blood cells of urine + urine submitted for examination, abnormal white blood cells of cerebrospinal fluid + cerebrospinal fluid submitted for examination, abnormal white blood cells of joint fluid + joint fluid submitted for examination.

[0173] Level 3 risk: changes in microorganisms and antibiotics submitted for testing;

[0174] Level 4 risk: fever + antibiotic change, high fever, abnormal erythrocyte sedimentation rate + antibiotic change, abnormal interleukin-6 + antibiotic change, abnormal lymphocytes + antibiotic change, abnormal eosinophils + antibiotic change, abnormal procalcitonin + antibiotic change, abnormal C-reactive protein + antibiotic change, abnormal blood white blood cells + antibiotic change, sputum culture detection of colonizing bacteria + antibiotic change, abnormal blood neutrophils + antibiotic change, abnormal pleural effusion white blood cells + antibiotic change, abnormal ascites white blood cells + antibiotic change, abnormal fecal white blood cells + antibiotic change, abnormal urine white blood cells + antibiotic change, multiple urine white blood cells full of visual field, abnormal cerebrospinal fluid white blood cells + antibiotic change, joint fluid white blood cells Abnormal + antibiotic change, full field of view of synovial fluid white blood cells, cultured pathogens + antibiotic change, imaging description of infection, imaging description of inflammation + antibiotic change, highly abnormal erythrocyte sedimentation rate, highly abnormal interleukin-6, highly abnormal lymphocytes, highly abnormal eosinophils, highly abnormal procalcitonin, highly abnormal C-reactive protein, highly abnormal blood white blood cells, highly abnormal blood neutrophils, moderate bacterial smear, large bacterial smear, small bacterial smear + antibiotic change, pathogens cultured from other specimens, non-colonized bacteria detected in sputum culture, abnormal Mycoplasma pneumoniae, infectious diseases detected, diagnosed infectious diseases, infectious diseases detected, infectious diseases (microorganisms), Helicobacter pylori infection, etc.

[0175] S150: Continuously monitor using the dynamic assessment model and automatically trigger an alarm when the risk score exceeds a threshold.

[0176] In this embodiment, the aforementioned step S150 may include steps S151 - S152 .

[0177] S151. Determine the final sum of the risk assessment weights of the data elements of the current trigger condition using the dynamic assessment model, and calculate the patient's risk score based on the final sum of the risk assessment weights;

[0178] S152. Send an alert through different channels based on the risk level determined by the patient's risk score.

[0179] Specifically, to ensure timely response to patients' risk status, the system has designed a multi-level early warning mechanism based on real-time scoring calculations. The specific implementation steps are as follows:

[0180] The weight of the data element according to the current trigger condition By formula Calculate the real-time score S.

[0181] The push strategy is as follows:

[0182] Level 4 Risk: When the real-time score reaches or exceeds the Level 4 risk threshold, the information will be pushed to the medical mobile app and message notification system, and displayed on the medical workstation. Action is required within 15 minutes.

[0183] Level 3 Risk: If the real-time score is Level 3, a notification will be sent to the nurses' workstation and a pop-up window will pop up in the electronic medical record to alert medical staff. A review must be completed within one hour.

[0184] Level 1 and 2 risks: For level 1 and 2 risks, only a warning pop-up window needs to be displayed in the electronic medical record, and it is recommended to pay attention within 24 hours.

[0185] For example, suppose patient ID is P001 and its real-time score is 3.2, which triggers a Level 4 alert. The system then pushes this information to the attending physician's mobile app and simultaneously alerts the medical workstation in the ward so that swift action can be taken.

[0186] To help medical staff more intuitively understand changes in patients' health status and risk trends, the system provides a visual traceability dashboard. The dashboard includes the following main parts:

[0187] Risk Overview Panel: Displays the current patient's risk level, using color labels (purple / red / pink / yellow / green) corresponding to Level 4 / Level 3 / Level 2 / Level 1 / No risk respectively.

[0188] Trend Analysis Charts: These charts display time series graphs of key indicators such as body temperature, white blood cell count (WBC), and antibiotic usage history. These graphs support zooming, making it easy to view trends over a specific time period. Warning events are specifically marked on the chart (for example, a red vertical line marks the triggering time of a Level 4 warning), allowing for quick identification of important events.

[0189] Original Data Traceability: Users can click on any data point in the trend analysis chart to jump to the corresponding original test report or medication record page for further details. This interactive design makes it easy to track information from macro trends to micro details.

[0190] In summary, S401 and S402 together form an efficient dynamic monitoring and feedback system that not only promptly identifies potential high-risk patients but also helps medical staff gain a deeper understanding of their specific conditions, enabling more precise treatment plans. Furthermore, the introduction of visualization tools has greatly improved the efficiency and accuracy of data analysis, contributing to improved healthcare quality.

[0191] S160: Adjust and optimize the dynamic evaluation model.

[0192] In this embodiment, the dynamic assessment model is updated with new data every morning and the final risk assessment weights involved are adjusted. At the same time, inefficient rules are eliminated or new rules are put online through expert review and knowledge graph verification.

[0193] To ensure the accuracy and effectiveness of the system while promoting its self-improvement and adaptation to new clinical situations, the following mechanisms were designed:

[0194] Quality control includes:

[0195] Restrictions on annotation permissions: Only medical personnel with the title of attending physician or above are allowed to manually annotate false positives. This restriction ensures the quality and professionalism of annotations.

[0196] Double review mechanism: After the annotation is completed, it must be confirmed by another doctor to reduce subjective errors and improve the accuracy of data annotation.

[0197] Every morning, the system uses the newly added data from the previous day (including the labeled results) to update the XGBoost model. The update formula is The learning rate η is 0.01. In addition, the system retains historical model versions for rollback when necessary.

[0198] If a feature j appears frequently in false positive cases, its weight is reduced. The adjustment formula is

[0199] High-frequency effective feature reward: For those features that play a key role in accurate warning, increase their weight in the model. Specifically, through the formula α new =α old -0.05·Effective hit rate to achieve.

[0200] Existing rules will be phased out when they prove to be inefficient.

[0201] New rules must go through a rigorous review process: first, doctors submit candidate rules, then the infectious disease expert committee votes, and finally, they are verified through a knowledge graph to ensure their scientificity and practicality before they are officially put into use.

[0202] In summary, S5 not only emphasizes the importance of continuous monitoring of system performance but also introduces a comprehensive feedback and optimization mechanism to ensure system accuracy and reliability. Through professional labeling of false positives, rigorous auditing, and continuous iterative updates of models and rules, the system is able to effectively self-optimize, thereby better serving clinical decision support.

[0203] This embodiment uses a linear fusion method based on expert rules and machine learning feature importance, weighted with a fixed ratio (α=0.6), to achieve a dynamic balance between medical experience and data analysis, overcoming the limitations of the static nature of traditional models.

[0204] The medical knowledge base is used to set the surgical effectiveness standards, and the graph database causal path retrieval is used to test the clinical logical rationality of the rule combination, filter out contradictory items, and ensure that the generated rules are in line with the diagnosis and treatment consensus.

[0205] A real-time computing engine based on time windows was built, using a dynamic threshold judgment strategy and combining multiple channels such as mobile pop-up windows, medical workstations, and electronic medical record tags to achieve rapid response to infection risks, which is superior to the traditional batch processing mode.

[0206] A manual labeling process for false positive cases is established, and new rules must be reviewed by an expert committee and verified by the knowledge graph, forming a complete optimization closed loop of "clinical feedback-model iteration-rule update".

[0207] Compared with the traditional multimodal data fusion method, the method of this embodiment can accurately distinguish between preventive and therapeutic medications through precise analysis of the patient's medication behavior, thereby avoiding false alarms caused by preventive medication. In addition, the method has been refined in the judgment of pathogen culture results, effectively identifying the difference between colonizing bacteria and sources of infection, and improving the accuracy of the system. At the same time, in terms of inflammatory index analysis, multi-dimensional inflammatory indicators and clinical data are comprehensively considered to reduce the possibility of misdiagnosis caused by abnormalities in a single indicator. This method not only improves the accuracy and real-time nature of the early warning, but also enhances the effectiveness and reliability of clinical diagnosis. Through these improvement measures, the early warning system can more accurately reflect the patient's true infection status and provide strong support for medical decision-making.

[0208] The above-mentioned infection risk grading and early warning method based on dynamic fusion of clinical parameters first ensures the integrity and consistency of the data by integrating patient data from multiple medical systems and performing regular preprocessing. It then quantifies the importance of this data by combining the opinions of clinical experts with machine learning models, and generates final risk assessment weights through a dynamic weighted fusion method, thus effectively combining medical experience and data analysis. Based on these assessment weights and knowledge graphs, it can accurately identify the key factors affecting infection risk and their relative importance, and establish a dynamically adjusted assessment model to calculate an individualized risk score. This model can continuously monitor the patient's health status and automatically trigger an alarm when it detects that the risk score exceeds the set threshold, promptly notifying medical staff to take action. This method not only improves the accuracy and reliability of infection judgments, but also effectively reduces false positives and missed positives, achieving more timely and accurate infection risk warnings and providing strong support for clinical decision-making.

[0209] Figure 3 FIG is a schematic block diagram of an infection risk classification warning system 300 based on dynamic fusion of clinical parameters provided by an embodiment of the present invention. Figure 3 As shown, corresponding to the above infection risk classification warning method based on dynamic fusion of clinical parameters, the present invention also provides an infection risk classification warning system 300 based on dynamic fusion of clinical parameters. The infection risk classification warning system 300 based on dynamic fusion of clinical parameters includes a unit for executing the above infection risk classification warning method based on dynamic fusion of clinical parameters, and the system can be configured in a server. Specifically, please refer to Figure 3 The infection risk grading warning system 300 based on dynamic fusion of clinical parameters includes an acquisition unit 301, a preprocessing unit 302, a weight generation unit 303, a model generation unit 304, a monitoring unit 305 and an adjustment unit 306.

[0210] The acquisition unit 301 is used to acquire patient data from multiple medical systems to obtain initial data. The preprocessing unit 302 is used to perform regularized preprocessing on the initial data to obtain preprocessing results. The weight generation unit 303 is used to quantify the importance of the preprocessing results by combining clinical expert opinions and machine learning models, and to generate final risk assessment weights through dynamic weighted fusion. The model generation unit 304 is used to determine the factors affecting risk and their weights based on the final risk assessment weights and knowledge graph, and to establish a dynamic assessment model for calculating risk scores. The monitoring unit 305 is used to continuously monitor using the dynamic assessment model and automatically trigger an alarm when the risk score exceeds a threshold. The adjustment unit 306 is used to adjust and optimize the dynamic assessment model.

[0211] In one embodiment, the pre-processing unit 302 includes:

[0212] The processing subunit is used to fill the numerical data of the initial data with the median of historical data of similar patients, mark the categorical data as unknown, perform data range verification on the initial data based on medical common sense to exclude unreasonable values, and confirm the validity of the categorical features of the initial data through the established infection-related knowledge graph rule base to obtain intermediate results; the standardization subunit is used to convert the continuous medical parameters in the intermediate results into classification labels with clear medical significance, verify and map the surgical effectiveness and microbial inspection type features to standardized medical entities through the clinical knowledge graph to obtain preprocessing results.

[0213] In one embodiment, the weight generating unit 303 includes:

[0214] The assignment subunit is used to assign initial weights to high-risk data elements based on clinical guidelines; the quantification subunit is used to use XGBoost to train a binary classification model and quantify the importance of the pretreatment results to infection prediction through SHAP values ​​to obtain quantitative results; the weight determination subunit is used to combine the expert rule weights and model weights to linearly superimpose the quantitative results in a fixed proportion to generate a final dynamic weight, and retain the data elements whose final dynamic weights meet the requirements through threshold filtering to obtain the final risk assessment weight.

[0215] In one embodiment, the weight determination subunit is used to combine the expert rule weight and the model weight to linearly superimpose the quantification results in a fixed proportion to generate a final dynamic weight, retain the data elements whose final dynamic weight meets the requirements through threshold filtering, and adjust the expert rule weight ratio online to determine the final risk assessment weight.

[0216] In one embodiment, the model generation unit 304 is used to determine the specific threshold for each data element to trigger different risk levels based on the final risk assessment weight, and decide the data element that can trigger a specific rule, use knowledge graph technology to verify the validity and mutual exclusivity of the combination conditions to obtain a dynamic assessment model for calculating the risk score, and automatically adjust the rule base according to the dynamic assessment model updated monthly.

[0217] In one embodiment, the monitoring unit 305 includes:

[0218] A scoring calculation subunit is used to use the dynamic assessment model to determine the final sum of the risk assessment weights of the data elements of the current trigger condition, and calculate the patient's risk score based on the final sum of the risk assessment weights; an alarm subunit is used to send alarms through different channels based on the risk level determined by the patient's risk score.

[0219] In one embodiment, the adjustment unit 306 is used to update the dynamic assessment model using new data every morning and adjust the final risk assessment weights involved, while eliminating inefficient rules or launching new rules through expert review and knowledge graph verification.

[0220] It should be noted that technical personnel in the relevant field can clearly understand that the specific implementation process of the above-mentioned infection risk grading warning system 300 based on dynamic fusion of clinical parameters and each unit can refer to the corresponding description in the aforementioned method embodiment. For the convenience and conciseness of the description, it will not be repeated here.

[0221] The infection risk classification warning system 300 based on dynamic fusion of clinical parameters can be implemented in the form of a computer program. Figure 4 Runs on the computer equipment shown.

[0222] See also Figure 4 , Figure 4 5 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device 500 may also be a server, wherein the server may be an independent server or a server cluster composed of multiple servers.

[0223] See Figure 4 The computer device 500 includes a processor 502 , a memory, and a network interface 505 connected via a system bus 501 , wherein the memory may include a non-volatile storage medium 503 and an internal memory 504 .

[0224] The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. The computer program 5032 includes program instructions, which, when executed, can enable the processor 502 to execute an infection risk classification warning method based on dynamic fusion of clinical parameters.

[0225] The processor 502 is used to provide computing and control capabilities to support the operation of the entire computer device 500.

[0226] The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute an infection risk classification warning method based on dynamic fusion of clinical parameters.

[0227] The network interface 505 is used to communicate with other devices through the network. Figure 4 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0228] The processor 502 is configured to execute a computer program 5032 stored in the memory to implement the following steps:

[0229] Acquire patient data from multiple medical systems to obtain initial data; perform regularized preprocessing on the initial data to obtain preprocessing results; quantify the importance of the preprocessing results by combining clinical expert opinions and machine learning models, and generate final risk assessment weights through dynamic weighted fusion; based on the final risk assessment weights and knowledge graph, determine the factors affecting the risk and their weights, and establish a dynamic assessment model for calculating the risk score; use the dynamic assessment model for continuous monitoring and automatically trigger an alarm when the risk score exceeds a threshold.

[0230] In one embodiment, after implementing the step of continuously monitoring using the dynamic assessment model and automatically triggering an alarm when the risk score exceeds a threshold, the processor 502 further implements the following steps:

[0231] Adjust and optimize the dynamic evaluation model.

[0232] In one embodiment, when the processor 502 implements the step of performing regularized preprocessing on the initial data to obtain a preprocessing result, the processor 502 specifically implements the following steps:

[0233] The numerical data of the initial data are filled with the median of historical data of similar patients, and the categorical data are marked as unknown. The initial data are checked for data range based on medical common sense to exclude unreasonable values. The validity of the categorical features of the initial data is confirmed through the established infection-related knowledge graph rule base to obtain intermediate results; the continuous medical parameters in the intermediate results are converted into classification labels with clear medical significance, and the surgical effectiveness and microbial inspection type features are verified and mapped to standardized medical entities through the clinical knowledge graph to obtain preprocessing results.

[0234] In one embodiment, when implementing the step of combining clinical expert opinions and machine learning models to quantify the importance of the preprocessing results and generating the final risk assessment weights through dynamic weighted fusion, the processor 502 specifically implements the following steps:

[0235] Based on clinical guidelines, high-risk data elements are assigned initial weights; a binary classification model is trained using XGBoost and the importance of the pretreatment results to infection prediction is quantified using SHAP values ​​to obtain quantitative results; the quantitative results are linearly superimposed at a fixed ratio based on expert rule weights and model weights to generate final dynamic weights, and data elements with the final dynamic weights that meet the requirements are retained through threshold filtering to obtain the final risk assessment weights.

[0236] In one embodiment, when the processor 502 implements the step of linearly superimposing the quantification results in a fixed ratio by combining the expert rule weights and the model weights to generate the final dynamic weights, and retaining the data elements whose final dynamic weights meet the requirements through threshold filtering to obtain the final risk assessment weights, the processor 502 specifically implements the following steps:

[0237] The final dynamic weight is generated by linearly superimposing the quantified results in a fixed proportion in combination with the expert rule weight and the model weight. The data elements whose final dynamic weight meets the requirements are retained through threshold filtering, and the proportion of the expert rule weight is adjusted online to determine the final risk assessment weight.

[0238] In one embodiment, when implementing the step of determining the factors affecting the risk and their weights based on the final risk assessment weights and the knowledge graph, and establishing a dynamic assessment model for calculating the risk score, the processor 502 specifically implements the following steps:

[0239] Based on the final risk assessment weight, the specific threshold for each data element to trigger different risk levels is determined, and the data elements that can trigger specific rules are determined. Knowledge graph technology is used to verify the validity and mutual exclusivity of the combination conditions to obtain a dynamic assessment model for calculating risk scores, and the rule base is automatically adjusted according to the dynamic assessment model updated monthly.

[0240] In one embodiment, the processor 502 implements the following steps when implementing the step of continuously monitoring using the dynamic assessment model and automatically triggering an alarm when the risk score exceeds a threshold:

[0241] The dynamic assessment model is used to determine the final sum of the risk assessment weights of the data elements of the current trigger condition, and the patient's risk score is calculated based on the final sum of the risk assessment weights; based on the risk level determined by the patient's risk score, an alert is sent through different channels.

[0242] In one embodiment, when implementing the step of adjusting and optimizing the dynamic evaluation model, the processor 502 specifically implements the following steps:

[0243] Every morning, new data is used to update the dynamic assessment model and adjust the final risk assessment weights involved. At the same time, inefficient rules are eliminated or new rules are launched through expert review and knowledge graph verification.

[0244] It should be understood that in the embodiment of the present application, the processor 502 may be a central processing unit (CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0245] Those skilled in the art will appreciate that all or part of the steps in the method of the above-described embodiment can be implemented by instructing the relevant hardware through a computer program. The computer program includes program instructions, which can be stored in a storage medium that is computer-readable. The program instructions are executed by at least one processor in the computer system to implement the steps in the method of the above-described embodiment.

[0246] Therefore, the present invention also provides a storage medium. The storage medium may be a computer-readable storage medium. The storage medium stores a computer program, wherein when the computer program is executed by a processor, the processor performs the following steps:

[0247] Acquire patient data from multiple medical systems to obtain initial data; perform regularized preprocessing on the initial data to obtain preprocessing results; quantify the importance of the preprocessing results by combining clinical expert opinions and machine learning models, and generate final risk assessment weights through dynamic weighted fusion; based on the final risk assessment weights and knowledge graph, determine the factors affecting the risk and their weights, and establish a dynamic assessment model for calculating the risk score; use the dynamic assessment model for continuous monitoring and automatically trigger an alarm when the risk score exceeds a threshold.

[0248] In one embodiment, after the processor executes the computer program to implement the step of continuously monitoring using the dynamic assessment model and automatically triggering an alarm when the risk score exceeds a threshold, the processor further implements the following steps:

[0249] Adjust and optimize the dynamic evaluation model.

[0250] In one embodiment, when the processor executes the computer program to implement the step of performing regularized preprocessing on the initial data to obtain a preprocessing result, the processor specifically implements the following steps:

[0251] The numerical data of the initial data are filled with the median of historical data of similar patients, and the categorical data are marked as unknown. The initial data are checked for data range based on medical common sense to exclude unreasonable values. The validity of the categorical features of the initial data is confirmed through the established infection-related knowledge graph rule base to obtain intermediate results; the continuous medical parameters in the intermediate results are converted into classification labels with clear medical significance, and the surgical effectiveness and microbial inspection type features are verified and mapped to standardized medical entities through the clinical knowledge graph to obtain preprocessing results.

[0252] In one embodiment, when the processor executes the computer program to implement the step of combining clinical expert opinions and machine learning models to quantify the importance of the preprocessing results and generating final risk assessment weights through dynamic weighted fusion, the processor specifically implements the following steps:

[0253] Based on clinical guidelines, high-risk data elements are assigned initial weights; a binary classification model is trained using XGBoost and the importance of the pretreatment results to infection prediction is quantified using SHAP values ​​to obtain quantitative results; the quantitative results are linearly superimposed at a fixed ratio based on expert rule weights and model weights to generate final dynamic weights, and data elements with the final dynamic weights that meet the requirements are retained through threshold filtering to obtain the final risk assessment weights.

[0254] In one embodiment, when the processor executes the computer program to implement the steps of linearly superimposing the quantification results in a fixed proportion by combining the expert rule weights and the model weights to generate the final dynamic weights, and retaining the data elements whose final dynamic weights meet the requirements through threshold filtering to obtain the final risk assessment weights, the processor specifically implements the following steps:

[0255] The final dynamic weight is generated by linearly superimposing the quantified results in a fixed proportion in combination with the expert rule weight and the model weight. The data elements whose final dynamic weight meets the requirements are retained through threshold filtering, and the proportion of the expert rule weight is adjusted online to determine the final risk assessment weight.

[0256] In one embodiment, when the processor executes the computer program to implement the steps of determining the factors affecting the risk and their weights based on the final risk assessment weights and the knowledge graph, and establishing a dynamic assessment model for calculating the risk score, the processor specifically implements the following steps:

[0257] Based on the final risk assessment weight, the specific threshold for each data element to trigger different risk levels is determined, and the data elements that can trigger specific rules are determined. Knowledge graph technology is used to verify the validity and mutual exclusivity of the combination conditions to obtain a dynamic assessment model for calculating risk scores, and the rule base is automatically adjusted according to the dynamic assessment model updated monthly.

[0258] In one embodiment, when the processor executes the computer program to implement the step of continuously monitoring using the dynamic assessment model and automatically triggering an alarm when the risk score exceeds a threshold, the processor specifically implements the following steps:

[0259] The dynamic assessment model is used to determine the final sum of the risk assessment weights of the data elements of the current trigger condition, and the patient's risk score is calculated based on the final sum of the risk assessment weights; based on the risk level determined by the patient's risk score, an alert is sent through different channels.

[0260] In one embodiment, when the processor executes the computer program to implement the step of adjusting and optimizing the dynamic evaluation model, the processor specifically implements the following steps:

[0261] Every morning, new data is used to update the dynamic assessment model and adjust the final risk assessment weights involved. At the same time, inefficient rules are eliminated or new rules are launched through expert review and knowledge graph verification.

[0262] The storage medium may be any computer-readable storage medium that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a magnetic disk, or an optical disk.

[0263] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the composition and steps of each example according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0264] In the several embodiments provided herein, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the various units is merely a logical functional division, and actual implementations may employ other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be omitted or not implemented.

[0265] The steps in the method of the embodiment of the present invention may be adjusted in order, combined, or deleted as needed. The units in the system of the embodiment of the present invention may be combined, divided, or deleted as needed. In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit.

[0266] If this integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the existing technology, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes a number of instructions for causing a computer device (which can be a personal computer, terminal, or network device, etc.) to execute all or part of the steps of the method described in various embodiments of the present invention.

[0267] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. An infection risk classification and early warning method based on dynamic fusion of clinical parameters, characterized by: include: Obtain patient data from multiple healthcare systems to obtain initial data; Performing regularized preprocessing on the initial data to obtain a preprocessing result; Combining clinical expert opinions and machine learning models to quantify the importance of the pre-processing results, and generating final risk assessment weights through dynamic weighted fusion; Based on the final risk assessment weights and knowledge graph, determine the factors affecting the risk and their weights, and establish a dynamic assessment model for calculating the risk score; The dynamic assessment model is used for continuous monitoring and automatically triggers alerts when the risk score exceeds a threshold.

2. The infection risk classification and early warning method based on dynamic fusion of clinical parameters according to claim 1 is characterized in that: After continuously monitoring using the dynamic assessment model and automatically triggering an alert when the risk score exceeds a threshold, the method further includes: Adjust and optimize the dynamic evaluation model.

3. The infection risk classification and early warning method based on dynamic fusion of clinical parameters according to claim 2 is characterized in that: The performing regularized preprocessing on the initial data to obtain a preprocessing result includes: The numerical data of the initial data is filled with the median of historical data of similar patients, and the categorical data is marked as unknown. The data range of the initial data is checked according to medical common sense to exclude unreasonable values. The validity of the categorical features of the initial data is confirmed through the established infection-related knowledge graph rule base to obtain an intermediate result; The continuous medical parameters in the intermediate results are converted into classification labels with clear medical significance, and the surgical effectiveness and microbial inspection type characteristics are verified and mapped to standardized medical entities through clinical knowledge graphs to obtain preprocessing results.

4. The infection risk classification and early warning method based on dynamic fusion of clinical parameters according to claim 1 is characterized in that: The combination of clinical expert opinions and machine learning models to quantify the importance of the pre-processing results and generate the final risk assessment weights through dynamic weighted fusion includes: Assign initial weights to high-risk data elements based on clinical guidelines; Use XGBoost to train a binary classification model and quantify the importance of the preprocessing results to infection prediction using SHAP values ​​to obtain quantitative results; The quantification results are linearly superimposed in a fixed proportion in combination with the expert rule weights and the model weights to generate the final dynamic weights. The data elements whose final dynamic weights meet the requirements are retained through threshold filtering to obtain the final risk assessment weights.

5. The infection risk classification and early warning method based on dynamic fusion of clinical parameters according to claim 4 is characterized in that: The combination of expert rule weights and model weights generates a final dynamic weight by linearly superimposing the quantified results in a fixed ratio, and retains data elements that meet the requirements of the final dynamic weight through threshold filtering to obtain the final risk assessment weight, including: The final dynamic weight is generated by linearly superimposing the quantified results in a fixed proportion in combination with the expert rule weight and the model weight. The data elements whose final dynamic weight meets the requirements are retained through threshold filtering, and the proportion of the expert rule weight is adjusted online to determine the final risk assessment weight.

6. The infection risk classification and early warning method based on dynamic fusion of clinical parameters according to claim 1 is characterized in that: The method of determining the factors affecting the risk and their weights based on the final risk assessment weights and the knowledge graph, and establishing a dynamic assessment model for calculating the risk score, includes: Based on the final risk assessment weight, the specific threshold for each data element to trigger different risk levels is determined, and the data elements that can trigger specific rules are determined. Knowledge graph technology is used to verify the validity and mutual exclusivity of the combination conditions to obtain a dynamic assessment model for calculating risk scores, and the rule base is automatically adjusted according to the dynamic assessment model updated monthly.

7. The infection risk classification and early warning method based on dynamic fusion of clinical parameters according to claim 1 is characterized in that: The dynamic assessment model is used to continuously monitor and automatically trigger an alert when the risk score exceeds a threshold, including: Determining the final sum of the risk assessment weights of the data elements of the current trigger condition using the dynamic assessment model, and calculating the patient's risk score based on the final sum of the risk assessment weights; Based on the risk level determined by the patient's risk score, alerts are sent through different channels.

8. The infection risk classification and early warning method based on dynamic fusion of clinical parameters according to claim 2 is characterized in that: The adjusting and optimizing the dynamic evaluation model includes: Every morning, new data is used to update the dynamic assessment model and adjust the final risk assessment weights involved. At the same time, inefficient rules are eliminated or new rules are launched through expert review and knowledge graph verification.

9. An infection risk grading warning system based on dynamic fusion of clinical parameters is characterized by: include: an acquisition unit, configured to acquire patient data from a plurality of medical systems to obtain initial data; A preprocessing unit, configured to perform regularized preprocessing on the initial data to obtain a preprocessing result; a weight generation unit, configured to combine clinical expert opinions and a machine learning model to quantify the importance of the preprocessing results and generate a final risk assessment weight through dynamic weighted fusion; A model generation unit, configured to determine the factors affecting the risk and their weights based on the final risk assessment weights and the knowledge graph, and to establish a dynamic assessment model for calculating the risk score; A monitoring unit is used to continuously monitor using the dynamic assessment model and automatically trigger an alarm when the risk score exceeds a threshold.

10. The infection risk classification warning system based on dynamic fusion of clinical parameters according to claim 9 is characterized in that: Also includes: An adjustment unit is used to adjust and optimize the dynamic evaluation model.

Images