A decision-making method and system for prophylactic antibiotic use in animal wounds
By acquiring the clinical characteristics of patients injured by animals, and utilizing clinical clustering models and data analysis, we can provide precise antibiotic use decisions, solve the problem of lacking unified standards, achieve scientific and rational antibiotic use, and reduce the risk of abuse.
Patent Information
- Application Number
- CN202611178865.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-05
- Publication Date
- 2026-11-17
AI Technical Summary
The lack of unified judgment standards makes it impossible to accurately determine whether animal bite wounds require preventative antibiotic use, resulting in a lack of scientific rigor and rationality in the treatment of dog bite wounds.
By acquiring the clinical characteristics of patients injured by animals, a trained target clinical clustering model is used to output antibiotic use decisions. Patients are grouped and risk stratified by combining retrospective data analysis and clustering methods to construct an initial clinical clustering model. The model is then optimized through prospective validation to provide specific antibiotic types, administration methods, and durations of administration.
This approach enables differentiated antibiotic use recommendations for wounds in different animals, eliminating the need for physicians' subjective experience, improving the scientific rigor, rationality, and cost-effectiveness of antibiotic use, and reducing the risk of antibiotic abuse.
Smart Images

Figure CN123120286A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of intelligent decision-making technology in medical clinical practice, and relates to, but is not limited to, a decision-making method and system for the use of prophylactic antibiotics in animal wounds. Background Technology
[0002] Wounds from mammal bites carry a high risk of infection and are prone to poor healing. Dog bite wounds have an even higher risk of infection because dogs' oral cavity is home to a variety of bacteria, such as Pasteurella multocida, Staphylococcus epidermidis, hemolytic streptococci, and anaerobic bacteria.
[0003] If a wound is primarily sutured, necrotic tissue and exudate may not be able to drain, leading to the spread of infection. Therefore, multiple studies emphasize the importance of prophylactic antibiotic use after primary suturing of high-risk wounds. High-risk factors for infection include: cat bites, limb wounds, large areas of skin and soft tissue loss, elderly patients, and patients with underlying conditions such as diabetes.
[0004] For dog bites, prophylactic antibiotic use does not reduce the risk of wound infection; it only benefits patients in specific situations such as injuries caused by non-canine mammals or head and facial injuries. Therefore, prophylactic antibiotic use for dog bite wounds is only necessary in specific circumstances. Regardless of whether antibiotics are used, wound irrigation, thorough debridement, and drainage remain the primary measures in the treatment of dog bites.
[0005] Currently, there is no consensus on whether prophylactic antibiotics should be used for dog bite wounds, and large-scale randomized controlled trials and retrospective cohort studies are lacking. It remains unclear which patients with specific clinical characteristics require prophylactic antibiotic use, and there is an urgent need to establish clinical clustering models to guide the rational use of antibiotics in these special patients. Summary of the Invention
[0006] In view of this, embodiments of this application provide a method and system for decision-making regarding the use of prophylactic antibiotics in animal wounds, which at least solves the problem that after primary suturing of wounds, there is a lack of unified judgment criteria, making it impossible to accurately determine whether prophylactic antibiotic use is necessary.
[0007] The technical solution of this application embodiment is implemented as follows: In a first aspect, embodiments of this application provide a method for making decisions regarding the use of prophylactic antibiotics in animal wounds, the method comprising: Obtain the first clinical characteristics of the patient who was injured by the first animal; The first clinical feature is input into the trained target clinical clustering model to obtain the antibiotic use decision output by the target clinical clustering model; The antibiotic use decision-making process includes various medication decisions and non-medication decisions, with each type of medication decision including the corresponding antibiotic type, administration method, and duration of administration; the method further includes: Based on retrospective data on the clinical characteristics and actual healing outcomes of patients with animal-induced injuries, the differences in healing outcomes under different antibiotic use decisions in these patients were compared, and analytical conclusions were drawn. Based on the analysis results, clustering methods were used to segment and stratify the retrospective animal-injured patients, and an initial clinical clustering model was constructed based on the patient segmentation and risk stratification results. Based on prospective cases of patients with animal-induced injuries, the initial clinical clustering model is externally validated, and the trained target clinical clustering model is obtained after successful validation.
[0008] Secondly, embodiments of this application provide a decision-making system for the prophylactic use of antibiotics in animal wounds, the system comprising: The acquisition module is used to acquire the first clinical characteristics of the patient injured by the first animal. The determination module is used to input the first clinical feature into the trained target clinical clustering model to obtain the antibiotic use decision output by the target clinical clustering model; The antibiotic use decision-making process includes various medication decisions and non-medication decisions. Each type of medication decision includes the corresponding antibiotic type, administration method, and duration of administration. The system also includes: The comparison module is used to compare the differences in healing effects of various antibiotic use decisions in retrospective animal-injured patients based on their clinical characteristics and actual healing effect data, and to obtain analytical conclusions. A construction module is used to perform patient grouping and risk stratification of the retrospective animal injury patients based on the analysis conclusions using clustering methods, and to construct an initial clinical clustering model based on the patient grouping and risk stratification results; The validation module is used to perform external validation on the initial clinical clustering model based on prospective cases of patients with animal-induced injuries. After successful validation, the trained target clinical clustering model is obtained.
[0009] The beneficial effects of the technical solutions provided in this application include at least the following: This application embodiment completes intelligent antibiotic use determination based on patient clinical characteristics and clinical clustering model. It can simultaneously cover various situations where antibiotics are used and not needed. The output results are refined to the specific type of antibiotic, administration method and medication cycle. It can get rid of the judgment bias caused by the physician's personal subjective experience and give differentiated antibiotic use suggestions for different animal wounds, so as to realize the scientific, rational and economical selection of antibiotics in diagnosis and treatment. Attached Figure Description
[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 A flowchart illustrating a decision-making method for the use of prophylactic antibiotics in animal wounds, provided as an embodiment of this application; Figure 2 This is a schematic diagram illustrating a method for determining the optimal number of clusters provided in an embodiment of this application; Figure 3 This is a schematic diagram illustrating the steps of a clustering algorithm provided in an embodiment of this application; Figure 4 A two-dimensional multi-scale analysis diagram of a clustering algorithm provided in this application embodiment; Figure 5 This is a contour diagram of a clustering result provided in an embodiment of this application. Detailed Implementation
[0011] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0012] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.
[0013] It should be noted that the terms "first, second, and third" used in the embodiments of this application are merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.
[0014] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of this application pertain. It should also be understood that terms such as those defined in general dictionaries should be understood to have a meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.
[0015] This application provides a method for making decisions regarding the use of prophylactic antibiotics for animal wounds, applied to an electronic device. The electronic device includes, but is not limited to, mobile phones, laptops, tablets, handheld internet devices, multimedia devices, streaming media devices, mobile internet devices, wearable devices, or other types of electronic devices. The functions implemented by this method can be achieved by a processor in the electronic device calling program code. The program code can be stored in a computer storage medium; therefore, the electronic device includes at least a processor and a storage medium. The processor can be used to perform processing based on decisions regarding the use of prophylactic antibiotics for animal wounds, and the storage medium can be used to store data required and generated during the decision-making process.
[0016] Figure 1 This application provides a flowchart illustrating a decision-making method for the prophylactic use of antibiotics in animal wounds, as illustrated in the embodiments of this application. Figure 1 As shown, the method includes at least the following steps: Step S110: Obtain the first clinical characteristics of the patient injured by the first animal; The first animal-injured patient refers to a patient who has been bitten or scratched by an animal such as a dog or cat and is currently seeking medical attention at a hospital for wound debridement and assessment of healing effectiveness. The first clinical characteristics include patient baseline characteristics and surgical treatment-related characteristics. The patient baseline characteristics include gender, age, whether it is the first exposure, type of injury, whether there is a medical history, time from exposure to the first visit, number of wounds, wound location, wound length, wound depth grading, and damage to other tissues besides the skin. The surgical treatment-related characteristics include the type of wound irrigation solution, wound irrigation time, whether compression ligation is used for hemostasis, whether tissue repair is performed, and whether rabies immunoglobulin is administered.
[0017] Step S120: Input the first clinical feature into the trained target clinical clustering model to obtain the antibiotic use decision output by the target clinical clustering model; The antibiotic use decision includes various medication decisions and non-medication decisions. Each type of medication decision includes the corresponding antibiotic type, administration method, and duration of administration. A non-medication decision means not using antibiotics.
[0018] In some embodiments, the method further includes: Step S121: Based on the clinical characteristics and actual healing effect data of retrospective animal-injured patients, compare the differences in healing effects of various antibiotic use decisions in the retrospective animal-injured patients, and obtain the analysis conclusions. The retrospective analysis of animal-injured patients refers to those previously treated and whose medical records are archived. The preset timeframe can be two years. This can be achieved by retrospectively analyzing animal-injured cases treated in the emergency department of this treatment center within the past two years, comparing the differences in healing effects of various antibiotic use decisions in the treatment of animal-injured wounds, and obtaining analytical conclusions. The clinical characteristics of the retrospective animal-injured patients include preoperative basic characteristics and surgical treatment-related characteristics.
[0019] The preoperative basic characteristics include the patient's gender, age, whether it is the first exposure, whether the injury was caused by a pet or stray dog / cat, past medical history (such as diabetes), time from exposure to the first visit, number of wounds, wound location (such as limbs, head, face, neck, trunk, multiple sites), wound length, wound depth grading (modified Lackmann grading), and damage or defects in other tissues (skin, subcutaneous fat, muscle, tendon, blood vessels, nerves, bones, etc.) other than the skin.
[0020] The surgical treatment-related characteristics include: the type of wound irrigation solution (e.g., benzalkonium chloride, normal saline, benzalkonium chloride combined with normal saline); the duration of wound irrigation; the method of deep wound irrigation (syringe or dedicated irrigation equipment); whether small blood vessels are compressed or ligated for hemostasis; whether tissue defect repair is performed; the type of sutures used for wound closure; whether plastic repair techniques (local flap or transfer flap) are used; whether drainage strips or tubes are placed in the wound; whether negative pressure wound therapy (VSD) is used; whether prophylactic antibiotics are used, the type of antibiotic, the method of administration (i.e., administration method), and the duration of antibiotic administration (i.e., duration of administration); whether rabies immunoglobulin is infiltrated into the wound; whether tetanus immunoglobulin and rabies vaccine are administered; and the type of rabies vaccine used (Vero cells, diploid cells, or no vaccination).
[0021] The actual healing effect data includes wound healing time (days from suture to suture removal), whether infection occurs, wound cosmetic effect score, number of days since infection occurred, and other complications (such as functional impairment, ischemic necrosis, subcutaneous effusion).
[0022] The diagnostic criteria for wound infection include primary and secondary criteria; meeting the corresponding conditions is sufficient to diagnose wound infection. One of the main criteria must be met: fever (body temperature greater than 38℃), local abscess, or lymphangitis; The secondary criteria must meet at least four of the following: local erythema extending more than 2 cm from the wound edge, wound pain, wound edema, drainage of pus, and white blood cell count greater than 12,000.
[0023] The Vancouver Scar Scale (VSS), as shown in Table 1, can be used to assess the cosmetic outcome of the wound during the period from 4 weeks to 3 months after surgical treatment. Table 1. Vancouver Scar Scale Assessment Criteria
[0024] Step S122: Based on the analysis conclusions, clustering methods are used to group and stratify the retrospective animal-injured patients, and an initial clinical clustering model is constructed based on the patient grouping and risk stratification results. This process involves combining the conclusions from previous data analysis with clustering algorithms to segment the collected retrospective animal injury case samples into groups. Patients are categorized based on various clinical characteristics, wound conditions, and prognoses. Risk stratification is then performed on each patient group based on their corresponding prognoses, assigning each group a corresponding risk level label. After risk stratification, sample data carrying patient group and risk stratification information is used to train the clinical clustering model. The risk level label serves as prior information from the sample data, guiding the model to identify combinations of clinical characteristics that differentiate between individuals with different prognostic risks. This step, using the patient grouping and risk stratification results as the foundational dataset and modeling basis, establishes an initial clinical clustering model, laying the groundwork for subsequent model training, optimization, and clinical application.
[0025] Step S123: Based on retrospective cases of patients with prospective animal-induced injuries, the initial clinical clustering model is externally validated. After successful validation, the trained target clinical clustering model is obtained.
[0026] Among them, prospective animal injury patients are newly admitted animal injury patients who are observed and recorded throughout the entire process and followed up until the end of healing. The generalization and accuracy of the initial clinical decision tree model can be verified by randomly sampling the validation set multiple times in prospective cases. If the validation fails, the initial clinical decision tree model is iteratively optimized until the validation passes, and the target clinical decision tree model is obtained after training.
[0027] In the above embodiments, the healing differences of different antibiotic use methods were analyzed based on real clinical retrospective cases. The patient groups were clustered and classified according to the characteristics of the injuries to build a model that fits the disease characteristics of different injured individuals. At the same time, external validation was used to test the model performance, which effectively ensured that the constructed model ran stably and reliably and had high clinical reference value.
[0028] In some embodiments, the antibiotic types include amoxicillin-clavulanate potassium, second-generation cephalosporins, third-generation cephalosporins, clindamycin, quinolones, aminoglycosides, cephalosporins combined with metronidazole, and multi-mechanism antibiotic combination regimens; the administration methods include oral administration, intravenous administration, and combined intravenous and oral administration; the duration of treatment is divided into three levels: 3 to 5 days, 6 to 7 days, and 8 to 14 days. This application comprehensively includes single-agent and combination antibiotic regimens commonly used in the clinical treatment of animal bites, adaptable to diverse pathogenic bacterial types in bite wounds. Multiple administration methods can match different injury severity levels, and the tiered duration of treatment can balance wound treatment effectiveness while reasonably controlling treatment costs and reducing drug resistance problems caused by long-term inappropriate use.
[0029] This step is the clinical application prediction step of the target clinical clustering model. The first clinical characteristics of the first animal-injured patient are input into the trained target clinical clustering model to obtain the antibiotic use decision output by the target clinical clustering model, so as to guide doctors to select the most appropriate antibiotic use decision based on patients with different clinical characteristics.
[0030] In the above embodiments, intelligent antibiotic use determination is completed by combining patient clinical characteristics with a clinical clustering model. This can simultaneously cover various situations where antibiotics are used and not needed. The output results are refined to the specific type of antibiotic, administration method, and medication cycle. This can eliminate the judgment bias caused by the physician's personal subjective experience and provide differentiated antibiotic use recommendations for different animal wounds, thus achieving the scientific, rational, and economical selection of antibiotics in diagnosis and treatment.
[0031] In some embodiments, the first clinical feature includes categorical clinical features, ordered hierarchical clinical features, and numerical clinical features. Step S120, "inputting the first clinical feature into the trained target clinical clustering model to obtain the antibiotic use decision output by the target clinical clustering model," includes: Step S1201: Apply one-hot encoding to the categorical clinical features, apply hierarchical encoding to the ordered hierarchical clinical features, and perform normalization on the numerical clinical features to obtain the preprocessed first clinical features. For categorical clinical features, one-hot encoding is used for transformation. These features are discrete categorical attributes (such as injury site, animal species, wound infection history, underlying diseases, etc.). One-hot encoding transforms each independent category into an independent binary feature dimension, eliminating the implicit order relationship between categories and adapting to the requirements of subsequent model operations. For ordered hierarchical clinical features, hierarchical encoding is used for transformation. These features have a natural hierarchical order of high / low, severity, and strength (such as wound severity, pain level, infection risk level, etc.). Hierarchical encoding assigns corresponding values according to the inherent order of the features, preserving the hierarchical logic and order information within the features. For numerical clinical features, normalization is performed. These features are continuous numerical data (such as age, wound area, detection index values, etc.). Normalization maps values of different dimensions and value ranges to a fixed interval, eliminating the interference of numerical magnitude differences on model training. After the above classification encoding and numerical transformation processing, the preprocessed first clinical feature is finally generated, which serves as the input data for the subsequent clinical clustering model.
[0032] Step S1202: Input the preprocessed first clinical feature into the trained target clinical clustering model, and use the target clinical clustering model to determine the corresponding first predicted healing effect data; The current clinical indicators of patients injured by animals are input into the target clinical clustering model, and the model calculates the first predicted healing effect data based on the built-in learning rules.
[0033] Step S1203: Based on the first predicted healing effect data, determine the antibiotic use decision output by the target clinical clustering model; In this process, the recovery quality and risk of complications are assessed by comparing the corresponding first predicted healing effect data, and the target antibiotic with the best healing effect is selected for use. In some embodiments, the first predicted healing effect data includes three predictive indicators: wound infection rate, wound cosmetic effect score, and wound healing time. Clinical diagnosis and treatment data of dog bites, industry standards, and clinical experience can be combined to pre-define judgment thresholds and grade ranges for each of the three indicators. The wound infection rate represents the probability of wound infection and is divided into high, medium, and low risk thresholds. The wound cosmetic effect score uses a quantitative scoring system; the higher the score, the better the appearance and scar condition after healing, with acceptable and risk lines set. The wound healing time represents the predicted time required for complete wound healing, and is divided into normal, longer, and significantly prolonged ranges based on the conventional healing time.
[0034] The three predictive indicators output by the clinical clustering model are compared one by one with preset thresholds and intervals to comprehensively assess the overall prognostic risk: If the wound infection rate is higher than the high-risk threshold, or the wound cosmetic outcome score is lower than the risk line, or the wound healing time significantly exceeds the normal duration, it indicates a poor prognosis and a high risk of adverse events. After comprehensive review based on cluster grouping characteristics, a decision to recommend prophylactic antibiotic use is output. If the wound infection rate is in the low-risk range, and the wound cosmetic outcome score is satisfactory and the wound healing time is within the normal range, it indicates a good wound prognosis, and a decision not to recommend prophylactic antibiotic use is output. If the indicators are in the borderline range (moderate infection rate, slightly longer healing time, and cosmetic outcome score close to the acceptable line), a compromise decision to use antibiotics at the discretion of the patient is output, based on the clinical characteristics of the patient's cluster.
[0035] In the above embodiments, one-hot encoding, hierarchical encoding, and normalization are used to standardize data for different types of clinical features, ensuring the quality of input data and the accuracy of model calculation. The preprocessed features are input into the trained clinical clustering model, and healing effect prediction data is obtained by combining multi-dimensional indicators such as wound infection incidence, wound cosmetic effect score, and wound healing time. Based on this, antibiotic use decisions are generated. This scheme not only fully adapts to the characteristics of diverse clinical data and realizes a comprehensive and objective assessment of wound prognosis, but also establishes a unified and quantitative standard for medication judgment, reduces the bias of subjective human judgment, accurately distinguishes the medication population to avoid the risk of antibiotic abuse, and the overall process is highly automated and efficient.
[0036] In some embodiments, step S122, "based on the analysis conclusions, using clustering methods to perform patient grouping and risk stratification on the retrospective animal injury patients, and constructing an initial clinical clustering model based on the patient grouping and risk stratification results," includes: Step S1221: Based on the analysis conclusions, perform univariate and multivariate analyses to screen out clinical characteristic factors related to the actual healing effect, such as wound infection rate, wound cosmetic effect score, and wound healing time. Determine the union of each clinical characteristic factor as the target clinical characteristic. Standardize the target clinical characteristic into a standard format to obtain the standard clinical characteristic. This application allows for the grouping of all patients injured by animals in the past. In this embodiment, clinical characteristic factors related to wound healing effects are selected and clustered after data preprocessing. During data preprocessing, all quantitative indicators in the clinical characteristic factors are normalized to eliminate the influence of dimensions and improve the generalization ability of the algorithm. Qualitative indicators in the clinical characteristic factors are independently encoded and converted into numerical variables to meet the input requirements of the clustering algorithm. Ordered variables are encoded to preserve their order relationship and ensure the compatibility and accuracy of multiple types of data.
[0037] Step S1222: Based on the standard clinical characteristics, calculate the dissimilarity matrix among patients with historical animal-induced injuries using Gower distance; Since the standard clinical features include both numerical and categorical variables, Gower distance is used to calculate pairwise dissimilarity and generate a dissimilarity matrix. Numerical variables are standardized by range before distance calculation, while categorical variables are processed using matching coefficients according to the Gower formula.
[0038] Step S1223: Based on the dissimilarity matrix, an agglomerative hierarchical clustering algorithm is used to perform initial clustering of patients and generate a clustering tree; based on the dissimilarity matrix, the average contour width and the dissimilarity curve within the cluster are calculated respectively; based on the peak value of the average contour width and the inflection point of the curve corresponding to the elbow rule, the optimal number of clusters is determined; based on the optimal number of clusters, the clustering tree is pruned to obtain the target group of patients. The initial clustering of patients can be performed using an agglomerative hierarchical clustering (AHC) algorithm. To determine the number of clusters k, the average contour width is calculated based on the Gower dissimilarity matrix for internal validation. The final value of k is determined by balancing internal metrics (peak contour width, curve inflection point) with the clinical interpretability of the cluster contours.
[0039] like Figure 2 As shown, Figure 2 The graph represents the intra-cluster dissimilarity curve, with the horizontal axis representing the number of candidate clusters and the vertical axis representing the total intra-cluster dissimilarity of the clustering results at the corresponding number of candidate clusters. As the number of clusters increases, the intra-cluster dissimilarity continuously decreases. The inflection point where the intra-cluster dissimilarity curve changes from a rapid decrease to a slow decrease is the optimal number of clusters determined by the elbow rule. Figure 2 As shown, the curve shows a sudden change in slope at the location marked by the green box (around the number of candidate clusters is 3), and the dissimilarity curve within the cluster changes from a rapid decrease to a gradual decrease. This number of candidate clusters is the optimal number of clusters under the elbow rule. The slope of the curve at the location marked by the red box does not change significantly.
[0040] In some embodiments, the optimal number of clusters can also be determined based on the peak value of the average profile width in the average profile coefficient curve. The higher the average profile width value, the better the intra-cluster compactness and inter-cluster separation of the current cluster. The closer the average profile width is to 1, the higher the intra-cluster sample similarity and the greater the inter-cluster sample difference, and the better the clustering effect. The number of candidate clusters corresponding to the peak value of the average profile width is the optimal number of clusters under the average profile coefficient criterion.
[0041] The optimal number of clusters can be determined by combining the optimal number of clusters under the average profile coefficient criterion and the optimal number of clusters selected by the elbow rule. When the optimal number of clusters under the two criteria is the same, the optimal number of clusters is determined as the best number of clusters.
[0042] When the inflection point of the intra-cluster dissimilarity curve does not align with the optimal number of clusters indicated by the peak value of the average contour width, the following comprehensive judgment strategy can be adopted: First, using the peak value of the average contour width as the core criterion, prioritize the number of clusters with higher contour coefficients to ensure both intra-cluster compactness and inter-cluster separation. Simultaneously, constrain the selected number of clusters by considering the changing trend of the intra-cluster dissimilarity curve, ensuring that the selected number of clusters lies within a stable range where the curve's descent flattens out, avoiding over-subdivision. Based on this, clinical validity is validated on the grouping results corresponding to the candidate cluster numbers. Differences in outcome indicators such as wound infection rate, cosmetic effect score, and healing time among clusters are compared. The number of clusters with statistically significant inter-group differences, reasonable sample distribution, and clinical interpretability is prioritized as the final optimal number of clusters.
[0043] Step S1224: Perform multidimensional scaling analysis and dimensionality reduction on the dissimilarity matrix to obtain two-dimensional spatial projection data. Label the two-dimensional spatial projection data based on the patient target group to obtain a two-dimensional visualization map with clustering group labels. Step S1225: Based on the two-dimensional visualization map, evaluate the clustering separation effect; if the clustering separation effect meets the standard, compare the healing effect outcome index among each cluster group to obtain the grouping effectiveness verification result; In order to visualize the high-dimensional mixed data and the obtained clustering results, the classic multidimensional scaling (MDS) analysis was applied to the Gower dissimilarity matrix, and the observation points were projected into a two-dimensional space. The clustering classification was superimposed on the MDS map to evaluate the separation effect.
[0044] like Figure 3 As shown, Figure 3The clustering tree (dendriogram) generated for the agglomerative hierarchical clustering algorithm is used to visually represent the merging hierarchy and clustering process of samples: the horizontal axis represents individual samples (e.g., P1~P6); the vertical axis represents the dissimilarity (Gower distance) scale, indicating the dissimilarity level when two branches (samples or subclusters) are merged; each branch of the tree represents a merging operation: from bottom to top, samples with the lowest dissimilarity are merged into subclusters first, and then gradually merged with other subclusters until all samples are classified into one large class; the different colored branches of the tree correspond to the independent patient target groups divided after horizontal pruning of the dendritic diagram at the selected optimal number of clusters: the green branch on the left corresponds to the first group of patient samples, and the red branch on the right corresponds to the second group of patient samples. The two branches are merged only at the top of the dendritic diagram at a high dissimilarity level, visually indicating that the dissimilarity between the two groups of patients is high, the inter-group discrimination is significant, and the clustering structure is clear and stable.
[0045] Agglomerative hierarchical clustering was performed based on Gower distance, and the dimensionality was reduced to two-dimensional space using classical multidimensional scaling analysis. Figure 4 The two-dimensional multi-scale analysis plot, namely the hierarchical clustering on Gower distance (MDS 2D) distribution plot, is shown below. Figure 4 As shown, the horizontal axis represents MDS Dim 1 (Multidimensional Scaling Dimension 1), and the vertical axis represents MDS Dim 2 (Multidimensional Scaling Dimension 2). Each point represents a wound. Figure 4 The patient target group numbers (1 to 4, a total of 4 subgroups) represent the results of the cohesive hierarchical clustering algorithm. Scatter points with the same number and the same color belong to the same cluster (patient subgroup, representing patients with the same type of injury). The coloring is assigned according to the cluster, showing obvious spatial separation between most clusters.
[0046] Figure 5 This displays the silhouette coefficient visualization results of the four clusters obtained based on Gower distance. Figure 5 Each bar chart segment represents the contour coefficient of a single wound, and is colored according to the clustering assignment results. Figure 5 The red dashed line represents the average profile width of all samples. Most bars exceed the average profile width (red dashed line), indicating good cluster cohesion and separation, and few misclassified samples.
[0047] This study allows for comparison of wound healing outcomes and outcome indicators across different clusters. For wound infection, the incidence rate within each cluster is summarized, and comparisons between clusters are performed using the chi-square test or Fisher's exact test. For wound healing time and cosmetic outcome scores, their distributions are summarized using the median and interquartile range, and comparisons between clusters are performed using the Kruskal-Wallis test. If the overall difference between groups is statistically significant, pairwise comparisons are conducted using the Dunn test, and multiple tests are used to correct for Type I errors, thus providing clinical differential value for cluster grouping from a clinical outcome perspective.
[0048] Step S1226: If the grouping validity verification result characterization verification is passed, risk stratification is performed on each patient target group based on the patient target group, the standard clinical characteristics and the actual healing effect data to obtain patient clustering and risk stratification results, and an initial clinical clustering model is constructed based on the patient clustering and risk stratification results.
[0049] This application, based on a retrospective analysis of multidimensional mixed clinical characteristics of animal-inflicted wound patients, conducts phenotypic clustering analysis on all dog bite wound samples. Through clustering, wound subtypes with different clinical characteristics are obtained, and the core prognostic differences among the subtypes are compared. Before model construction, standardized data preprocessing is performed, distinguishing between numerical and categorical variables and standardizing their formats. Ordinal variables are rank-coded, and only clinical characteristic factors are strictly used in cluster distance calculations; healing effect data are excluded. Given the mixed data characteristics of this dataset, which includes both numerical and categorical variables, this application uses the more adaptable Gower distance as the distance metric. Numerical variables are standardized for range, and categorical variables are quantified using matching coefficients. An agglomerative hierarchical clustering algorithm is selected. The average contour width is calculated based on the constructed Gower distance matrix, and the optimal number of clusters is determined by combining statistically optimal indicators with clinical application significance. To visually validate the clustering effect, classical multidimensional scaling (MDS) analysis was performed on the Gower distance matrix, projecting the high-dimensional clustering results onto a two-dimensional plane to clearly display the inter-group separation of each cluster subtype. After clustering, statistical differences in prognostic outcomes among different subtypes were assessed. For binary outcomes, wound infection rate was evaluated using the chi-square test or Fisher's exact test; for numerical outcomes such as healing time and scar VSS score, the Kruskal-Wallis test was used; and post-hoc pairwise comparisons between groups were performed using the Dunn test with multiple comparison corrections. To further validate the effectiveness and stability of the unsupervised clinical clustering model, a supervised learning control model was trained using the same clinical characteristic variables. A weighted random forest classification model was constructed for infection outcomes, and its classification performance was evaluated using AUROC and calibration curves. For numerical outcomes such as healing time and VSS score, a random forest regression model was constructed, and its regression fit was evaluated using the RMSE index. Based on this, for the wound subgroup of prophylactic antibiotic use (PUA), three new medication characteristic variables were added to the above basic clinical factor system: antibiotic type, administration method, and duration of use. Using completely consistent data preprocessing, distance measurement, clustering algorithm, and validation evaluation system, a subgroup clinical clustering model adapted to antibiotic use decision-making was independently constructed.
[0050] In the above embodiments, target clinical features highly correlated with healing outcomes were screened and standardized through univariate and multivariate analyses, providing a high-quality data foundation for clustering. The dissimilarity matrix was calculated using Gower distance, adapted to mixed-type data, combined with an agglomerative hierarchical clustering algorithm. The optimal number of clusters was determined using both the average profile width and the elbow rule criteria, ensuring the stability and statistical reliability of patient grouping results. Multidimensional scaling analysis visualization and inter-group outcome index comparisons provided dual validation, effectively guaranteeing the clustering separation effect and the clinical effectiveness of grouping. The final initial clinical clustering model provides a data-driven decision-making basis for the precise stratification of animal-injured patients and the subsequent development of differentiated treatment plans, helping to improve individualized treatment levels and reduce the risk of adverse healing events.
[0051] In some embodiments, the method further includes: The retrospective cases of patients with animal-induced injuries were divided into multiple subsets; Initialize the current model as the initial clinical clustering model; Repeat the following iterative process until the performance metrics meet the preset standards, confirming successful verification: Select a single subset as the test set and the remaining subset as the training set. Based on the current model, determine the second predicted healing effect data for each retrospective case in each test set. Based on the second predicted healing effect data for each retrospective case, determine the priority antibiotic use decision for the corresponding retrospective case. Based on the priority antibiotic use decision and the actual healing effect data for the corresponding retrospective case, calculate the model performance index for the current iteration. After all subsets have been used as test sets to complete one round of full cross-validation, the parameters of the current model are adjusted based on the model performance metrics of this round to obtain the optimized clinical clustering model. Update the current model to the optimized clinical clustering model and proceed to the next iteration.
[0052] The generalization and accuracy of the clinical clustering model were evaluated using cross-validation. This involved randomly dividing all retrospective cases into subsets, using each subset sequentially as the test set, and the remainder as the training set, repeating this process multiple times to obtain stable performance metrics. Model performance metrics included accuracy, area under the curve (AUC), sensitivity, and specificity. Accuracy reflects the proportion of correct overall judgments by the model; AUC reflects the model's ability to distinguish between good and bad wound healing outcomes; sensitivity indicates the model's ability to detect patients with poor wound healing; and specificity measures the accuracy in identifying patients with good wound healing. By adjusting model parameters and thresholds, an optimal balance between sensitivity and specificity can be achieved. Ultimately, the obtained target clinical clustering model not only provides scientific evidence for clinical practice but also assists clinicians in providing early warnings and individualized treatment for high-risk patients, further improving wound management and prognosis.
[0053] Based on the patient target groups obtained from the above cohesive hierarchical clustering, an initial clinical clustering model for determining antibiotic use decisions is constructed, with each patient's group and corresponding standard clinical characteristics as independent variables and actual healing effect data (wound infection status and wound cosmetic effect score) as dependent variables.
[0054] In the above embodiments, multi-fold cross-validation is used to iteratively optimize model parameters, continuously correct model bias, improve model generalization ability and robustness, avoid overfitting problems caused by single training and testing, and ensure the stability of model clinical prediction.
[0055] In some embodiments, the method further includes: Step S101: Based on preset inclusion criteria and preset exclusion criteria, screen patients with a history of animal injury to obtain the retrospective animal injury patients; The inclusion criteria include: screening candidate cases based on inclusion criteria and excluding cases that do not meet the requirements according to preset exclusion criteria. The preset inclusion criteria include: time from injury to first medical visit less than or equal to 24 hours; clear record of the number and partial details of the wound; wound length greater than or equal to the preset length; modified Lackmann classification greater than the preset level; primary suturing of the wound after irrigation and thorough debridement; placement of a drainage device after suturing; complete record of antibiotic type, administration method, and duration; complete record of tetanus and rabies post-exposure prophylaxis; and complete record of actual healing effect data after a preset follow-up period following surgery. The pre-defined exclusion criteria include: the time from injury to the first medical visit is greater than 24 hours, or there was an infection before treatment; no primary suturing was performed; initial treatment was received at another hospital; skin or subcutaneous tissue defects requiring plastic surgery repair; severe fractures or major arterial and venous injuries; and loss to follow-up.
[0056] It should be noted that in the modified Lackmann classification, grade 0 indicates only the epidermis is involved; grade 1 indicates the wound involves the entire thickness of the skin and subcutaneous tissue but not the muscle layer; grade 2 indicates damage to the entire thickness of the skin and subcutaneous tissue, and involves the deep fascia and muscle layer; grade 3 indicates damage to the entire thickness of the skin and subcutaneous tissue, and involves the deep fascia and muscle layer, with tissue loss; grade 4 indicates damage to major blood vessels and nerves in addition to grade 3; and grade 5 indicates damage to bone or organs in addition to grade 3.
[0057] In the above embodiments, by setting standardized and rigorous case inclusion and exclusion criteria, case samples that do not meet research standards or have interfering factors are filtered out. This standardizes the entire process of case injury treatment, wound management, epidemic prevention and control, and postoperative follow-up, controls sample quality from the data source, avoids the influence of invalid data, and consolidates the foundational reliability of subsequent model training and intelligent judgment.
[0058] Based on the foregoing embodiments, this application provides a decision system for the use of prophylactic antibiotics in animal wounds. The system includes modules and sub-modules, and each unit of each sub-module can be implemented by a processor in an electronic device; of course, it can also be implemented by specific logic circuits. In the implementation process, the processor can be a central processing unit (CPU), a microprocessor (MPU), a digital signal processor (DSP), or a field programmable gate array (FPGA), etc.
[0059] This application provides a decision-making system for the prophylactic use of antibiotics in animal wounds, the system comprising: The acquisition module is used to acquire the first clinical characteristics of the patient injured by the first animal. The determination module is used to input the first clinical feature into the trained target clinical clustering model to obtain the antibiotic use decision output by the target clinical clustering model; The antibiotic use decision-making process includes various medication decisions and non-medication decisions. Each type of medication decision includes the corresponding antibiotic type, administration method, and duration of administration. The system also includes: The comparison module is used to compare the differences in healing effects of various antibiotic use decisions in retrospective animal-injured patients based on their clinical characteristics and actual healing effect data, and to obtain analytical conclusions. A construction module is used to perform patient grouping and risk stratification of the retrospective animal injury patients based on the analysis conclusions using clustering methods, and to construct an initial clinical clustering model based on the patient grouping and risk stratification results; The validation module is used to perform external validation on the initial clinical clustering model based on prospective cases of patients with animal-induced injuries. After successful validation, the trained target clinical clustering model is obtained.
[0060] In some possible embodiments, the first clinical feature includes categorical clinical features, ordinal clinical features, and numerical clinical features, and the determining module includes: The processing submodule is used to perform one-hot encoding on the categorical clinical features, hierarchical encoding on the ordered hierarchical clinical features, and normalization on the numerical clinical features to obtain the preprocessed first clinical features. The first determining submodule is used to input the preprocessed first clinical feature into the trained target clinical clustering model, and use the target clinical clustering model to determine the corresponding first predicted healing effect data. The second determining submodule is used to determine the antibiotic use decision output by the target clinical clustering model based on the first predicted healing effect data.
[0061] In some possible embodiments, the building module includes: The third determination submodule is used to perform univariate and multivariate analyses based on the analysis conclusions, screen out clinical feature factors related to the actual healing effect, such as the wound infection rate, wound cosmetic effect score, and wound healing time, and determine the union of each clinical feature factor as the target clinical feature; and unify the target clinical feature into a standard format to obtain the standard clinical feature. The calculation submodule is used to calculate the dissimilarity matrix between patients with historical animal-induced injuries based on the standard clinical characteristics and using Gower distance. The fourth determination submodule is used to perform initial patient clustering using an agglomerative hierarchical clustering algorithm based on the dissimilarity matrix to generate a clustering tree; calculate the average contour width and intra-cluster dissimilarity curve based on the dissimilarity matrix; determine the optimal number of clusters based on the peak value of the average contour width and the inflection point of the curve corresponding to the elbow rule; and prune the clustering tree based on the optimal number of clusters to obtain the target patient group. The analysis submodule is used to perform multidimensional scaling analysis and dimensionality reduction on the dissimilarity matrix to obtain two-dimensional spatial projection data. Based on the patient target group, the two-dimensional spatial projection data is labeled to obtain a two-dimensional visualization map with clustering group labels. The comparison submodule is used to evaluate the clustering separation effect based on the two-dimensional visualization map; if the clustering separation effect meets the standard, it compares the healing effect outcome index among the cluster groups to obtain the grouping effectiveness verification result. A submodule is constructed to perform risk stratification on each patient target group based on the patient target group, the standard clinical characteristics, and the actual healing effect data, after the grouping validity verification result characterization verification is passed, to obtain patient clustering and risk stratification results, and to construct an initial clinical clustering model based on the patient clustering and risk stratification results.
[0062] In some possible embodiments, the system further includes: A segmentation module is used to divide the retrospective cases of patients with animal-induced injuries into multiple subsets; An initialization module is used to initialize the current model as the initial clinical clustering model; The iterative optimization module is used to repeatedly execute the following iterative process until the performance indicators meet the preset standards and the verification is confirmed to be successful. Select a single subset as the test set and the remaining subset as the training set. Based on the current model, determine the second predicted healing effect data for each retrospective case in each test set. Based on the second predicted healing effect data for each retrospective case, determine the priority antibiotic use decision for the corresponding retrospective case. Based on the priority antibiotic use decision and the actual healing effect data for the corresponding retrospective case, calculate the model performance index for the current iteration. After all subsets have been used as test sets to complete one round of full cross-validation, the parameters of the current model are adjusted based on the model performance metrics of this round to obtain the optimized clinical clustering model. Update the current model to the optimized clinical clustering model and proceed to the next iteration.
[0063] In some possible embodiments, the system further includes: The screening module is used to screen patients with a history of animal injury based on preset inclusion criteria and preset exclusion criteria to obtain the retrospective animal injury patients; The pre-defined inclusion criteria include: time from injury to first medical visit less than or equal to 24 hours; clear record of the number and partial number of wounds; wound length greater than or equal to the pre-defined length; modified Lackmann classification greater than the pre-defined level; primary closure of the wound after irrigation and thorough debridement; placement of a drainage device after suturing; complete record of antibiotic type, administration method, and duration of administration; complete record of tetanus and rabies post-exposure prophylaxis; and complete record of actual healing effect data after a pre-defined postoperative follow-up period. The pre-defined exclusion criteria include: the time from injury to the first medical visit is greater than 24 hours, or there was an infection before treatment; no primary suturing was performed; initial treatment was received at another hospital; skin or subcutaneous tissue defects requiring plastic surgery repair; severe fractures or major arterial and venous injuries; and loss to follow-up.
[0064] In some possible embodiments, the antibiotic types include amoxicillin-clavulanate potassium, second-generation cephalosporins, third-generation cephalosporins, clindamycin, quinolones, aminoglycosides, cephalosporins combined with metronidazole, and multi-mechanism antibiotic combination regimens. The administration methods include oral administration, intravenous administration, and combined intravenous and oral administration; The duration of medication is divided into three levels: 3 to 5 days, 6 to 7 days, and 8 to 14 days.
[0065] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.
[0066] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.
[0067] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.
[0068] The units described above as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units; some or all of the units may be selected to achieve the purpose of the embodiments of this application according to actual needs. In addition, each functional unit in the embodiments of this application may be fully integrated into one processing unit, or each unit may be a separate unit, or two or more units may be integrated into one unit; the integrated unit may be implemented in hardware or in the form of hardware plus software functional units.
[0069] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause the device automatic test line to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.
[0070] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined to obtain new method embodiments without conflict. The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined to obtain new method embodiments or device embodiments without conflict.
[0071] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A decision-making method for the prophylactic use of antibiotics in animal wounds, characterized in that, The method includes: Obtain the first clinical characteristics of the patient who was injured by the first animal; The first clinical feature is input into the trained target clinical clustering model to obtain the antibiotic use decision output by the target clinical clustering model; The antibiotic use decision-making process includes various medication decisions and non-medication decisions, with each type of medication decision including the corresponding antibiotic type, administration method, and duration of administration; the method further includes: Based on retrospective data on the clinical characteristics and actual healing outcomes of patients with animal-induced injuries, the differences in healing outcomes under different antibiotic use decisions in these patients were compared, and analytical conclusions were drawn. Based on the analysis results, clustering methods were used to segment and stratify the retrospective animal-injured patients, and an initial clinical clustering model was constructed based on the patient segmentation and risk stratification results. Based on prospective cases of patients with animal-induced injuries, the initial clinical clustering model is externally validated, and the trained target clinical clustering model is obtained after successful validation.
2. The method according to claim 1, characterized in that, The first clinical features include categorical clinical features, ordinal clinical features, and numerical clinical features. The first clinical features are input into a trained target clinical clustering model to obtain the antibiotic use decision output by the target clinical clustering model, including: One-hot encoding is applied to the categorical clinical features, hierarchical encoding is applied to the ordered hierarchical clinical features, and normalization is performed on the numerical clinical features to obtain the preprocessed first clinical features. The preprocessed first clinical feature is input into the trained target clinical clustering model, and the target clinical clustering model is used to determine the corresponding first predicted healing effect data. Based on the first predicted healing effect data, the antibiotic use decision output by the target clinical clustering model is determined.
3. The method according to claim 1, characterized in that, Based on the analytical conclusions, clustering methods were used to segment and stratify the retrospective animal-injured patients. An initial clinical clustering model was constructed based on the patient segmentation and risk stratification results, including: Based on the above analysis conclusions, univariate and multivariate analyses were performed to screen out clinical characteristic factors related to the actual healing effect, such as the wound infection rate, wound cosmetic effect score, and wound healing time. The union of each clinical characteristic factor was determined as the target clinical characteristic. The target clinical characteristic was then standardized into a standard format to obtain the standard clinical characteristic. Based on the aforementioned standard clinical characteristics, the dissimilarity matrix among patients with historical animal-induced injuries was calculated using Gower distance. Based on the dissimilarity matrix, an agglomerative hierarchical clustering algorithm is used to initially cluster patients and generate a cluster tree. Based on the dissimilarity matrix, the average contour width and the dissimilarity curve within the cluster are calculated respectively. The optimal number of clusters is determined based on the peak value of the average contour width and the inflection point of the curve corresponding to the elbow rule. The cluster tree is pruned based on the optimal number of clusters to obtain the target group of patients. Multidimensional scaling analysis and dimensionality reduction are performed on the dissimilarity matrix to obtain two-dimensional spatial projection data. The two-dimensional spatial projection data are labeled based on the patient target group to obtain a two-dimensional visualization map with clustering group labels. Based on the two-dimensional visualization map, the clustering separation effect is evaluated; if the clustering separation effect meets the standard, the healing effect outcome index among each cluster group is compared to obtain the grouping effectiveness verification result. If the grouping validity verification results are successful, risk stratification is performed on each patient target group based on the patient target group, the standard clinical characteristics, and the actual healing effect data to obtain patient clustering and risk stratification results. An initial clinical clustering model is then constructed based on the patient clustering and risk stratification results.
4. The method according to claim 1, characterized in that, The method further includes: The retrospective cases of patients with animal-induced injuries were divided into multiple subsets; Initialize the current model as the initial clinical clustering model; Repeat the following iterative process until the performance metrics meet the preset standards, confirming successful verification: Select a single subset as the test set and the remaining subset as the training set. Based on the current model, determine the second predicted healing effect data for each retrospective case in each test set. Based on the second predicted healing effect data for each retrospective case, determine the priority antibiotic use decision for the corresponding retrospective case. Based on the priority antibiotic use decision and the actual healing effect data for the corresponding retrospective case, calculate the model performance index for the current iteration. After all subsets have been used as test sets to complete one round of full cross-validation, the parameters of the current model are adjusted based on the model performance metrics of this round to obtain the optimized clinical clustering model. Update the current model to the optimized clinical clustering model and proceed to the next iteration.
5. The method according to claim 1, characterized in that, The method further includes: Based on preset inclusion and exclusion criteria, patients with a history of animal injury were screened to obtain the retrospective animal injury patients. The pre-defined inclusion criteria include: time from injury to first medical visit less than or equal to 24 hours; clear record of the number and partial number of wounds; wound length greater than or equal to the pre-defined length; modified Lackmann classification greater than the pre-defined level; primary closure of the wound after irrigation and thorough debridement; placement of a drainage device after suturing; complete record of antibiotic type, administration method, and duration of administration; complete record of tetanus and rabies post-exposure prophylaxis; and complete record of actual healing effect data after a pre-defined postoperative follow-up period. The pre-defined exclusion criteria include: the time from injury to the first medical visit is greater than 24 hours, or there was an infection before treatment; no primary suturing was performed; initial treatment was received at another hospital; skin or subcutaneous tissue defects requiring plastic surgery repair; severe fractures or major arterial and venous injuries; and loss to follow-up.
6. The method according to claim 1, characterized in that, The types of antibiotics include amoxicillin-clavulanate potassium, second-generation cephalosporins, third-generation cephalosporins, clindamycin, quinolones, aminoglycosides, cephalosporins combined with metronidazole, and multi-mechanism antibiotic combination regimens. The administration methods include oral administration, intravenous administration, and combined intravenous and oral administration; The duration of medication is divided into three levels: 3 to 5 days, 6 to 7 days, and 8 to 14 days.
7. A decision-making system for the prophylactic use of antibiotics in animal wounds, characterized in that, The system includes: The acquisition module is used to acquire the first clinical characteristics of the patient injured by the first animal. The determination module is used to input the first clinical feature into the trained target clinical clustering model to obtain the antibiotic use decision output by the target clinical clustering model; The antibiotic use decision-making process includes various medication decisions and non-medication decisions. Each type of medication decision includes the corresponding antibiotic type, administration method, and duration of administration. The system also includes: The comparison module is used to compare the differences in healing effects of various antibiotic use decisions in retrospective animal-injured patients based on their clinical characteristics and actual healing effect data, and to obtain analytical conclusions. A construction module is used to perform patient grouping and risk stratification of the retrospective animal injury patients based on the analysis conclusions using clustering methods, and to construct an initial clinical clustering model based on the patient grouping and risk stratification results; The validation module performs external validation on the initial clinical clustering model based on prospective cases of patients with animal-induced injuries. Once the validation is successful, the trained target clinical clustering model is obtained.
8. The system according to claim 7, characterized in that, The first clinical feature includes categorical clinical features, ordinal clinical features, and numerical clinical features. The determining module includes: The processing submodule is used to perform one-hot encoding on the categorical clinical features, hierarchical encoding on the ordered hierarchical clinical features, and normalization on the numerical clinical features to obtain the preprocessed first clinical features. The first determining submodule is used to input the preprocessed first clinical feature into the trained target clinical clustering model, and use the target clinical clustering model to determine the corresponding first predicted healing effect data. The second determining submodule is used to determine the antibiotic use decision output by the target clinical clustering model based on the first predicted healing effect data.
