A method of predicting the risk of infection from an infection caused by gastrointestinal bleeding
By constructing clinical event propagation pathways and quantifying the risk contribution of each pathway, the problem of difficulty in characterizing causal propagation relationships in existing technologies has been solved. This enables dynamic prediction and interpretation of infection risk in patients with acute gastrointestinal bleeding, thereby improving the practical value of clinical decision support systems.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- 四川互慧软件有限公司
- Filing Date
- 2026-05-25
- Publication Date
- 2026-07-14
AI Technical Summary
Existing methods for predicting infection risk cannot effectively characterize causal transmission relationships, lack interpretability, and are difficult to handle situations involving multiple overlapping pathways, resulting in insufficient practical value for clinical decision support systems.
We construct clinical event propagation pathways, quantify the risk contribution of each pathway by building initial state maps and clinical state maps, and achieve pathway-level analysis of the risk of secondary infection in patients with acute gastrointestinal bleeding, outputting the results of risk source explanation.
It enables dynamic characterization and quantification of infection risk, provides reliable risk prediction basis, supports clinicians in developing targeted intervention measures, and reduces the incidence of secondary infections and sepsis in patients.
Smart Images
Figure CN122392973A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of medical data processing, and specifically discloses a method for predicting the risk of infection caused by gastrointestinal bleeding. Background Technology
[0002] Acute gastrointestinal bleeding (AGIB) is a common acute and critical illness in clinical practice. Patients often require blood transfusions, endoscopic hemostasis, interventional therapy, or monitoring support during treatment. Due to factors such as continuous bleeding, increased invasive procedures, and changes in immune function, some patients may further develop secondary infections or even sepsis, significantly increasing patient mortality and medical burden.
[0003] Clinical practice shows that the infection risk in patients with acute gastrointestinal bleeding is not determined by a single indicator, but rather by the combined effect of multiple clinical events. For example, persistent bleeding may lead to an increase in the number of blood transfusions, while massive transfusions may affect the body's immune status; at the same time, invasive procedures such as central venous catheterization, mechanical ventilation, or endoscopic procedures may also increase the risk of infection. These factors often affect the patient's infection risk gradually through a chain of consecutive clinical events.
[0004] However, most existing methods for predicting infection risk employ simple scoring models or machine learning classification models, such as prediction models based on logistic regression, random forests, or neural networks. While these methods can improve prediction accuracy to some extent, they suffer from drawbacks such as failing to characterize causal transmission relationships, lacking interpretability, and being unable to handle situations involving overlapping multiple pathways.
[0005] In view of this, the present invention provides a method for predicting the risk of infection caused by gastrointestinal bleeding, which realizes path-level analysis of the risk of secondary infection in patients with acute gastrointestinal bleeding. This method enables risk prediction to not only provide probability results but also explain the source of risk, thereby significantly improving the practical value of clinical decision support systems. Summary of the Invention
[0006] The purpose of this invention is to provide a method for predicting the infection risk caused by gastrointestinal bleeding, addressing the problem of how to construct clinical event transmission pathways, quantify the risk contribution of each pathway, and thus predict the infection risk. The specific solution is as follows: A method for predicting the risk of infection caused by gastrointestinal bleeding includes the following steps: Step 1: Collect multi-source clinical data from patients with acute gastrointestinal bleeding and construct an infection risk feature set; the infection risk feature set includes various indicator data used to indicate whether a patient has an infection; Step 2: Construct an initial state graph, and determine the node states, activated nodes, and edges of the graph nodes in the initial state graph based on the infection risk feature set to obtain the clinical state graph; Step 3: Extract paths from the graph nodes and edges in the clinical status graph to obtain a set of infection transmission paths; Step 4: Quantify the risk of each infection transmission path based on the node status to obtain the path risk value of the infection transmission path; Step 5: Aggregate the risk values of multiple paths in the infection transmission path set to obtain the total infection risk.
[0007] Furthermore, the construction of the initial state diagram includes: Graph nodes are constructed based on anomaly categories; Graph edges are constructed based on the propagation relationships between anomaly categories to obtain the initial state graph; The obtained clinical status map includes: The indicator data corresponding to the graph node category in the infection risk set is used as the graph node status; When the indicator data is abnormal, the corresponding graph node is activated; otherwise, the corresponding graph node is not activated. When all adjacent graph nodes are activated, the edges connecting the adjacent graph nodes are activated. A clinical state graph is constructed based on activated graph nodes and edges.
[0008] Furthermore, the graph nodes include bleeding state nodes, coagulation state nodes, and infection state nodes; The bleeding status nodes include bleeding nodes, abnormal bleeding volume nodes, rebleeding nodes, abnormal hemoglobin nodes, and transfusion nodes; The coagulation status nodes include INR abnormality nodes, PT abnormality nodes, APTT abnormality nodes, and platelet abnormality nodes. The infection status nodes include fever nodes, abnormal WBC nodes, inflammatory response nodes, and abnormal PCT nodes.
[0009] Furthermore, step 3 includes: Use the bleeding state node or the coagulation state node as the starting node; Use the infected state node as the termination node; All transmission paths from the starting node to the ending node in the clinical status diagram are considered as the set of infection transmission paths.
[0010] Furthermore, step 4 includes: Obtain the node status value of each graph node along the infection transmission path; Based on historical infection data, obtain the causal relationship weights between adjacent graph nodes; The path risk value is calculated based on the node state value and the causal relationship weight.
[0011] Furthermore, the path risk value is: ; in, The risk value for path p; The symbol is a multiplication sign; i represents a node variable in the infection transmission path; The weight of the causal relationship between node i and node i+1; Let i be the node state value of node i.
[0012] Furthermore, step 6 is included: based on the path risk value, output the explanation results of the risk source.
[0013] Furthermore, step 6 includes: The path risk values of each transmission path are weighted and summed to obtain the total infection risk, and the product of the path risk value and the weight of the transmission path is used as the path risk contribution. The infection risk level is determined based on the magnitude of the total infection risk and the infection risk threshold; The risk contribution is ranked, and the top-ranked preset transmission paths are taken as the main transmission paths of infection risk. Infection risk level and main transmission route are used as the explanation of risk sources.
[0014] Furthermore, step 7 is included: updating the transmission pathway weights based on the total infection risk and actual clinical outcomes.
[0015] Furthermore, the updated propagation path weights are:
[0016] in, The updated propagation path weights; The weights of the original propagation path; The learning rate; This indicates the infection status, with 1 indicating an infection and 0 indicating no infection. Total risk of infection.
[0017] The present invention has the following advantages and beneficial effects: This invention is the first to construct a clinical event propagation path and apply it to risk prediction, overcoming the technical limitation of traditional models that treat each clinical indicator as an independent feature. By mapping the patient's discrete clinical indicators to nodes and edges in a causal graph structure, a clinical event network is constructed, enabling the complete identification of the entire propagation path from the onset of bleeding to the occurrence of infection, thus reconstructing the formation logic of the patient's infection risk.
[0018] Unlike existing technologies that obtain risk scores solely through static feature weighting, this invention can simulate a continuous causal transmission process such as "continuous bleeding → increased blood transfusion → immunosuppression → increased infection risk," thus achieving a dynamic characterization of the infection risk formation process, rather than a simple calculation of superimposed indicators.
[0019] This invention establishes a standardized path risk calculation model by defining node state values and edge propagation weights. It can quantify the risk intensity of each clinical event propagation path and obtain the patient's final infection risk probability through weighted aggregation, making the risk assessment results more in line with clinical reality.
[0020] This invention, while outputting the probability of infection risk, can automatically identify and output the core transmission path that contributes the most to the final infection risk, and explain the source of risk in a way that conforms to cognitive logic, providing clinicians with a more reliable basis for risk prediction.
[0021] This invention provides the final infection risk and core transmission pathway through clinical event transmission pathways, enabling physicians to develop targeted interventions to reduce the incidence of secondary infections and sepsis in patients and reduce unnecessary infection treatments. Attached Figure Description
[0022] Figure 1 An exemplary flowchart of an infection risk prediction method for infections caused by gastrointestinal bleeding provided by the present invention. Detailed Implementation
[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.
[0024] Example 1 This embodiment details the complete infection risk prediction process. For example... Figure 1 As shown, the process includes the following steps: Step 1: Collect multi-source clinical data and construct an infection risk feature set.
[0025] The system retrieves multi-source data on patients with acute gastrointestinal bleeding from the hospital information system within a specified time window (e.g., the past 24 hours). Data categories include basic information, bleeding markers, coagulation markers, infection markers, vital signs, and procedural records. For example, basic information might include: age 65 years, history of chronic liver disease; bleeding markers: esophageal and gastric variceal rupture bleeding, classified as "moderate to large" bleeding volume, hemoglobin decreasing from 110 g / L to 78 g / L within 24 hours, with 2 units of red blood cells transfused during this period; and coagulation markers: INR 1.8 (abnormal), PT 20 seconds (prolonged), PLT 45 × 10⁻⁶. 9 / L (decreased); Infection indicators: body temperature 38.5℃, WBC 12×10 9 / L, CRP 50mg / L, PCT 1.2ng / mL; Vital signs: heart rate 110 bpm, blood pressure 95 / 60 mmHg; Procedure record: previously underwent emergency endoscopic hemostasis, currently has a central venous catheter in place.
[0026] The system cleans, standardizes, and aligns the data over time to construct a set of infection risk characteristics X for the patient. For example, this set includes the following features: (Decrease in hemoglobin) = 32 g / L (Number of blood transfusions) = 2 times. (INR)=1.8, (CRP) = 50 mg / L, etc. These characteristics comprehensively reflect the patient's current bleeding, coagulation, infection, and intervention status, providing structured input for the subsequent construction of graphical models.
[0027] Step 2: Construct a clinical status map.
[0028] First, an initial state graph is constructed. Based on the pathophysiological knowledge of acute gastrointestinal bleeding, the system predefines the causal graph structure G=(V,E) for this domain. The nodes V of the graph cover all the abnormality categories involved in step 1, including bleeding state nodes, coagulation state nodes, and infection state nodes. Bleeding state nodes include bleeding nodes, abnormal bleeding volume nodes, rebleeding nodes, abnormal hemoglobin nodes, and transfusion nodes. Coagulation state nodes include abnormal INR nodes, abnormal PT nodes, abnormal APTT nodes, and abnormal platelet nodes. Infection state nodes include fever nodes, abnormal WBC nodes, inflammatory response nodes, and abnormal PCT nodes. Edges E represent causal propagation relationships based on clinical consensus. For example: “bleeding node → transfusion node,” “transfusion node → inflammatory response node,” “abnormal hemoglobin node → rebleeding node,” and “abnormal platelet node → rebleeding node,” etc.
[0029] Then, a clinical state graph is generated. The system inputs the risk feature set X of patient A from step 1 into the initial state graph. The state values of each node are calculated using predefined rules and a standardization function. .For example, This indicates that the node is activated; the bleeding level is classified as moderate to heavy. The node was activated; because INR > 1.5, CRP > 10 mg / L and PCT > 0.5 ng / mL, therefore, a comprehensive judgment was made. A node is activated. A node with a state value of 1 is activated. An edge is activated when both ends of a directed edge are activated. The system extracts all activated nodes and edges from the original graph to form a clinical state graph reflecting patient A's current individualized clinical state. The picture at this moment. It dynamically and in real time presents the various abnormal events that the patient is experiencing and their transmission network.
[0030] Step 3: Extract the set of infection transmission paths.
[0031] The clinical status diagram generated by the system in step 2 The system uses a graph traversal algorithm (such as Depth-First Search, DFS) to search for all paths from the risk start node to the infection termination node. The start node is set to all bleeding and coagulation state nodes. The termination node is set to all infection state nodes. The path length is limited to no more than 5 nodes to avoid excessively long paths that lose clinical interpretability. For example, the system might find the following 3 infection transmission paths on patient A's clinical status graph: Path 1 ( ): Bleeding node → Transfusion node → Inflammatory response node; Path 2 ( ): Platelet abnormality node → Rebleeding node → Inflammatory response node. Path 3 ( The pathways are: bleeding point → blood transfusion point → fever point → inflammatory response point. These pathways together constitute the set of infection transmission pathways. It depicts from different perspectives the process by which patients gradually develop from problems such as bleeding and coagulation abnormalities to the risk of infection.
[0032] Step 4: Quantify the risk of each path and calculate the path risk value.
[0033] The system calculates the risk value for each path in the path set P. Get node status value The system obtains the standardized state value of each node on the path from step 2. For example, for path 1, the bleeding node state value is 1.0, the transfusion node state value is calculated as 0.8 based on the number and amount of transfusions, and the inflammatory response node state value is calculated as 0.9 based on CRP and PCT levels. The system obtains a weight matrix through pre-training with massive historical case data. For example, w(bleeding node, transfusion node) = 0.9 indicates that continuous bleeding has a high probability of leading to transfusion; w(transfusion node, inflammatory response node) = 0.7 indicates that the impact of transfusion on the immune system is an important driver of infection; and w(platelet abnormality node, rebleeding node) = 0.6. Based on the node state values and causal relationship weights, the path risk value corresponding to each path is calculated. The path risk value is: ; in, The risk value for path p; The symbol is a multiplication sign; i represents a node variable in the infection transmission path; The weight of the causal relationship between node i and node i+1; Let be the node state value of node i. This calculation quantifies the specific strength of each causal chain in shaping the final risk of infection.
[0034] Step 5: Calculate the total risk of infection.
[0035] The system collects risk values from all pathways and calculates the patient's total infection risk through weighted aggregation. The final overall infection risk for patients is calculated as follows: ; in, The patient's total risk of infection; This represents the path risk value. This represents the path weight, which can be determined based on path length, clinical experience rules, or historical case statistics.
[0036] Step 6: Output the explanation results of the risk sources.
[0037] To provide decision support for physicians, the system generates interpretable risk analysis reports. Risk contribution is calculated. ; in, Risk contribution of path k; Indicates path weight; This represents the path risk value; k is the main propagation path variable.
[0038] Set the thresholds to 0.2 and 0.5. When When the probability is 0.2, it is considered a low-risk probability; when 0.2 < A score of 0.5 indicates a moderate risk of infection; otherwise, a high risk of infection. The pathways are ranked in descending order of contribution, and the top two are selected as the primary risk pathways. When the current patient... At that time, the system outputs the following to the doctor's interface: Current patient infection risk level: Moderate; Main sources of risk: 1. Continuous bleeding leads to increased blood transfusions, which in turn triggers an inflammatory response; 2. Thrombocytopenia leads to rebleeding, which in turn triggers an inflammatory response.
[0039] The results clearly inform doctors of the origin of the risk, rather than just providing an empty number, and directly support doctors in taking measures such as strictly assessing transfusion indications, using thrombopoietin, and strengthening hemostasis.
[0040] Step 7: Update the model weights based on the feedback from the actual outcome.
[0041] After a patient is discharged or the observation period ends, the system tracks their actual clinical outcome. Assume the patient eventually developed an infection (Y=1). The system compares the predicted value with the actual value and finds... The risk was underestimated. Therefore, the system compares the actual outcome with the predicted result and adjusts the propagation path weights based on the prediction error. The updated propagation path weights are: ; in, The updated propagation path weights; The weights of the original propagation path; The learning rate; Indicates the infection status. This indicates that an infection has occurred. This indicates that no infection has occurred; This represents the total risk of infection. When a transmission route appears frequently in actual cases, its weight gradually increases; conversely, if the route contributes little, its weight decreases.
[0042] Through this feedback learning, the weights of paths 1 and 2, which played a dominant role in this case, were increased, enabling the model to more accurately predict higher risks when encountering similar patients. This feedback mechanism allows the model to continuously self-optimize in clinical applications.
[0043] Example 2 This embodiment describes a scenario involving low-risk patients to demonstrate the discriminative ability of the present invention.
[0044] Patient B had mild ulcer bleeding, received only one blood transfusion, and had no coagulation abnormalities or fever. CRP was slightly elevated (8 mg / L). The only activated nodes were the bleeding node, the transfusion node, and the inflammatory response node.
[0045] A unique path was extracted: bleeding node → transfusion node → inflammatory response node. Its calculation yielded... <0.2. Low risk of infection. Output interpretation: Risk source: Post-bleeding transfusion, but the current inflammatory response is mild; routine monitoring is recommended. The patient is not infected (Y=0), consistent with the prediction, with minimal weight adjustment.
[0046] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.
Claims
1. A method for predicting the risk of infection caused by gastrointestinal bleeding, characterized in that, Includes the following steps: Step 1: Collect multi-source clinical data from patients with acute gastrointestinal bleeding and construct an infection risk characteristic set; The infection risk feature set includes various indicator data used to indicate whether a patient has been infected; Step 2: Construct an initial state graph, and determine the node states, activated nodes, and edges of the graph nodes in the initial state graph based on the infection risk feature set to obtain the clinical state graph; Step 3: Extract paths from the graph nodes and edges in the clinical status graph to obtain a set of infection transmission paths; Step 4: Quantify the risk of each infection transmission path based on the node status to obtain the path risk value of the infection transmission path; Step 5: Aggregate the risk values of multiple paths in the infection transmission path set to obtain the total infection risk.
2. The method for predicting the risk of infection caused by gastrointestinal bleeding according to claim 1, characterized in that, The construction of the initial state diagram includes: Graph nodes are constructed based on anomaly categories; Graph edges are constructed based on the propagation relationships between anomaly categories to obtain the initial state graph; The obtained clinical status map includes: The indicator data corresponding to the graph node category in the infection risk set is used as the graph node status; When the indicator data is abnormal, the corresponding graph node is activated; otherwise, the corresponding graph node is not activated. When all adjacent graph nodes are activated, the edges connecting the adjacent graph nodes are activated. A clinical state graph is constructed based on activated graph nodes and edges.
3. The method for predicting the risk of infection caused by gastrointestinal bleeding according to claim 2, characterized in that, The graph nodes include bleeding status nodes, coagulation status nodes, and infection status nodes; The bleeding status nodes include bleeding nodes, abnormal bleeding volume nodes, rebleeding nodes, abnormal hemoglobin nodes, and transfusion nodes; The coagulation status nodes include INR abnormality nodes, PT abnormality nodes, APTT abnormality nodes, and platelet abnormality nodes. The infection status nodes include fever nodes, abnormal WBC nodes, inflammatory response nodes, and abnormal PCT nodes.
4. The method for predicting the risk of infection caused by gastrointestinal bleeding according to claim 3, characterized in that, Step 3 includes: Use the bleeding state node or the coagulation state node as the starting node; Use the infected state node as the termination node; All transmission paths from the starting node to the ending node in the clinical status diagram are considered as the set of infection transmission paths.
5. The method for predicting the risk of infection caused by gastrointestinal bleeding according to claim 1, characterized in that, Step 4 includes: Obtain the node status value of each graph node along the infection transmission path; Based on historical infection data, obtain the causal relationship weights between adjacent graph nodes; The path risk value is calculated based on the node state value and the causal relationship weight.
6. The method for predicting the risk of infection caused by gastrointestinal bleeding according to claim 5, characterized in that, The path risk value is: ； in, The risk value for path p; The symbol is a multiplication sign; i represents a node variable in the infection transmission path; The weight of the causal relationship between node i and node i+1; Let i be the node state value of node i.
7. The method for predicting the risk of infection caused by gastrointestinal bleeding according to claim 1, characterized in that, It also includes step 6: Based on the path risk value, output the explanation results of the risk source.
8. The method for predicting the risk of infection caused by gastrointestinal bleeding according to claim 7, characterized in that, Step 6 includes: The path risk values of each transmission path are weighted and summed to obtain the total infection risk, and the product of the path risk value and the weight of the transmission path is used as the path risk contribution. The infection risk level is determined based on the magnitude of the total infection risk and the infection risk threshold; The risk contribution is ranked, and the top-ranked preset transmission paths are taken as the main transmission paths of infection risk. Infection risk level and main transmission route are used as the explanation of risk sources.
9. The method for predicting the risk of infection caused by gastrointestinal bleeding according to claim 8, characterized in that, It also includes step 7: updating the transmission pathway weights based on the total infection risk and actual clinical outcomes.
10. The method for predicting the risk of infection caused by gastrointestinal bleeding according to claim 9, characterized in that, The update propagation path weight is: in, The updated propagation path weights; The weights of the original propagation path; The learning rate; This indicates the infection status, with 1 indicating an infection and 0 indicating no infection. Total risk of infection.