An antibiotic prescription auditing method and device based on drug resistance trend analysis, a terminal device, and a storage medium

Abstract

The application discloses a kind of antibiotic prescription auditing method, device, terminal equipment and storage medium based on drug resistance trend analysis, belong to medical information processing technical field.Method includes obtaining bacterial drug resistance monitoring data, patient medical record data and infection test index;According to bacterial drug resistance monitoring data, construct heterogeneous feature matrix, in combination with the long short-term memory network of pre-set, the drug resistance trend of different pathogenic bacteria to each type of antibiotic is time series prediction, obtain the drug resistance rate prediction result, according to the drug resistance rate prediction result, update the pre-set antibiotic sensitivity atlas;According to patient medical record data and infection test index, infection assessment is carried out to obtain patient infection state vector, in combination with the antibiotic prescription to be audited, the updated antibiotic sensitivity atlas and the preset medical constraint rule are carried out multidimensional prescription matching analysis, and the comprehensive audit score obtained is carried out auditing to generate prescription audit record.Through implementation of the present application, the problem of lagging of existing drug auditing basis is solved.

Description

Technical Field

[0001] This invention relates to the field of medical information processing technology, and in particular to an antibiotic prescription review method, apparatus, terminal equipment, and storage medium based on drug resistance trend analysis. Background Technology

[0002] Antimicrobial resistance has become a major challenge in the field of public health. Antimicrobial prescription review is a key means for medical institutions to curb antimicrobial resistance, standardize anti-infection treatment, and implement the requirements for graded management of antimicrobial drugs. The timeliness of the review determines the rationality of antimicrobial drug use and is also related to the effectiveness of infection treatment.

[0003] Current mainstream clinical antibiotic review techniques primarily rely on bacterial susceptibility monitoring data as the core reference. The conventional approach involves offline statistical analysis, summarizing bacterial susceptibility data on a fixed monthly or quarterly basis to output outdated resistance statistics reports. This forms a static antibiotic susceptibility reference standard, which is then combined with basic rules such as prescription classification and indication matching to verify prescription compliance. However, because susceptibility data depends on periodic offline summarization, the generated resistance statistics only reflect the overall resistance level of past periods and cannot synchronously match the real-time dynamic changes in pathogen resistance in the ward. When the prevalence of drug-resistant bacteria changes within the ward, the sensitivity standards used for prescription review cannot be dynamically updated. The review reference is severely outdated compared to the actual clinical resistance environment, easily leading to irrational drug use where drug selection is mismatched with the current resistance situation. Summary of the Invention

[0004] This invention provides an antibiotic prescription review method, device, terminal equipment, and storage medium based on drug resistance trend analysis, which can effectively solve the problem that the static antibiotic sensitivity of the prior art cannot adapt to the changing trend of drug-resistant bacteria, resulting in the lag in drug review and unreasonable review.

[0005] One embodiment of the present invention provides a method for antibiotic prescription review based on drug resistance trend analysis, comprising: Obtain bacterial resistance monitoring data, patient medical record data, infection test indicators, and antibiotic prescriptions pending review; Based on the bacterial resistance monitoring data, a heterogeneous feature matrix integrating spatial and temporal dimensions is constructed. Based on the preset long short-term memory network and the heterogeneous feature matrix, the resistance trends of different pathogens to various antibiotics are predicted in time series to obtain the resistance rate prediction results. The preset antibiotic sensitivity map is updated based on the resistance rate prediction results. Infection assessment is performed based on the patient's medical record data and the infection test indicators to obtain the patient's infection status vector; A multi-dimensional prescription matching analysis is performed based on the antibiotic prescription to be reviewed, the updated antibiotic sensitivity map, the patient's infection status vector, and the preset medical constraint rules to obtain a comprehensive review score. Prescriptions are reviewed based on the comprehensive review score, and prescription review records are generated.

[0006] Furthermore, based on the bacterial resistance monitoring data, a heterogeneous feature matrix integrating spatial and temporal dimensions is constructed, including: Using the distribution of departments, wards, and beds as the spatial dimension of the matrix, and the sampling calendar time, patient hospitalization duration, and antibiotic frequency as the temporal dimension of the matrix, the pathogen species code, specimen source characteristics, antimicrobial drug ATC code, inhibition zone continuity value, sampling timestamp, and ward spatial coordinates in the bacterial resistance monitoring data are extracted to generate the original feature tensor. Based on the pathogen species code and antimicrobial drug ATC code in the original feature tensor, determine the pathogen sample size and antimicrobial drug sample size; When the number of pathogen samples or the number of antimicrobial drug samples is less than a preset sample quantity threshold, the proportion of pathogens or the proportion of antimicrobial drugs is corrected based on the Bayesian smoothing algorithm to obtain the corrected feature tensor, and the corrected feature tensor is used as a heterogeneous feature matrix.

[0007] Furthermore, based on the preset long short-term memory network and the heterogeneous feature matrix, time-series predictions are made on the resistance trends of different pathogens to various antibiotics to obtain resistance rate prediction results. The preset antibiotic susceptibility map is then updated based on these resistance rate prediction results, including: The heterogeneous feature matrix is ​​input into a preset long short-term memory network. Through a multi-head attention mechanism, weights are assigned to the features corresponding to recent drug-sensitive samples in the same disease area. Based on the weights, time-series prediction is performed to obtain the drug resistance rate prediction result. The drug resistance rate prediction result includes the predicted drug resistance rate of each pathogen to various antibiotics. The predicted drug resistance rate is compared with the preset single drug warning threshold to obtain the first comparison result, and the antimicrobial drug marker warning label in the preset antibiotic susceptibility map is updated according to the first comparison result. The predicted drug resistance rate is compared with the preset drug use restriction threshold to obtain a second comparison result. The drug use priority of antimicrobial drugs in the preset antibiotic susceptibility map is updated according to the second comparison result to obtain the updated antibiotic susceptibility map.

[0008] Furthermore, infection assessment is performed based on the patient's medical record data and the infection test indicators to obtain a patient infection status vector, including: Entity extraction is performed on the unstructured medical history text in the patient's medical record data to obtain the semantic features of the case. The time-series values, the slope of index changes, and the fluctuation range are extracted from the infection test indicators as time-series test features. The patient's infection status vector is obtained by fusing features based on the semantic features of the case and the temporal test features.

[0009] Furthermore, based on the antibiotic prescription to be reviewed, the updated antibiotic susceptibility profile, the patient's infection status vector, and preset medical constraint rules, a multi-dimensional prescription matching analysis is performed to obtain a comprehensive review score, including: Extract the drug's antibacterial spectrum, antibiotic grade, dosage, type of combined medication, and pathogen identification from the antibiotic prescription to be reviewed to generate the basic characteristics of the prescription. Based on the basic characteristics of the prescription and the updated antibiotic susceptibility profile, a resistance risk score is calculated. Based on the basic features of the prescription and the patient's infection status vector, the infection indicator matching score is calculated; The basic characteristics of the prescription are verified for safety according to preset medical constraint rules to obtain a medication safety score; The antibiotic classification score is determined based on the antibiotic classification level in the basic characteristics of the prescription. Based on the corresponding weighting coefficients, the infection indication matching score, the medication safety score, and the antibiotic classification score are positively weighted, and the drug resistance risk score is negatively weighted to obtain the comprehensive review score. Specifically, when performing security verification on the basic features of the prescription according to preset medical constraint rules, if a hard medical rule in the medical constraint rules is triggered, the medication safety score will be directly set to 0.

[0010] Furthermore, prescriptions are reviewed based on the comprehensive review score, and prescription review records are generated, including: The comprehensive review score is compared with a preset first interception threshold and a second interception threshold. If the overall review score is less than the first interception threshold, a prescription interception operation is performed and a serious alarm message is generated; If the overall review score is greater than the second interception threshold, the prescription review is approved. If the overall review score is not less than the first interception threshold and not greater than the second interception threshold, a manual prescription review operation will be performed. Based on the comprehensive review score, the corresponding prescription operation, and the operation time, a prescription review record is generated.

[0011] As an improvement to the above solution, another embodiment of the present invention provides an antibiotic prescription review device based on drug resistance trend analysis, comprising: The data acquisition module is used to acquire bacterial resistance monitoring data, patient medical record data, infection test indicators, and antibiotic prescriptions pending review. The drug resistance trend analysis module is used to construct a heterogeneous feature matrix that integrates spatial and temporal dimensions based on the bacterial drug resistance monitoring data. Based on the preset long short-term memory network and the heterogeneous feature matrix, it performs time-series prediction of the drug resistance trends of different pathogens against various antibiotics, obtains the drug resistance rate prediction results, and updates the preset antibiotic susceptibility map based on the drug resistance rate prediction results. The infection assessment module is used to assess the infection based on the patient's medical record data and the infection test indicators, and obtain the patient's infection status vector. The comprehensive review score analysis module is used to perform multi-dimensional prescription matching analysis based on the antibiotic prescription to be reviewed, the updated antibiotic sensitivity map, the patient infection status vector, and preset medical constraint rules to obtain a comprehensive review score. The prescription review record generation module is used to review prescriptions based on the comprehensive review score and generate prescription review records.

[0012] Furthermore, the drug resistance trend analysis module is used to perform time-series prediction of the drug resistance trends of different pathogens against various antibiotics based on a preset long short-term memory network and the heterogeneous feature matrix, to obtain drug resistance rate prediction results, and to update the preset antibiotic susceptibility map based on the drug resistance rate prediction results, including: The heterogeneous feature matrix is ​​input into a preset long short-term memory network. Through a multi-head attention mechanism, weights are assigned to the features corresponding to recent drug-sensitive samples in the same disease area. Based on the weights, time-series prediction is performed to obtain the drug resistance rate prediction result. The drug resistance rate prediction result includes the predicted drug resistance rate of each pathogen to various antibiotics. The predicted drug resistance rate is compared with the preset single drug warning threshold to obtain the first comparison result, and the antimicrobial drug marker warning label in the preset antibiotic susceptibility map is updated according to the first comparison result. The predicted drug resistance rate is compared with the preset drug use restriction threshold to obtain a second comparison result. The drug use priority of antimicrobial drugs in the preset antibiotic susceptibility map is updated according to the second comparison result to obtain the updated antibiotic susceptibility map.

[0013] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements an antibiotic prescription review method based on drug resistance trend analysis as described in the above embodiments.

[0014] Another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform an antibiotic prescription review method based on drug resistance trend analysis as described in the above embodiment.

[0015] By implementing this invention, at least the following beneficial effects are achieved: This invention provides an antibiotic prescription review method, device, terminal equipment, and storage medium based on drug resistance trend analysis. The method constructs a heterogeneous feature matrix integrating spatial and temporal dimensions, combines it with a long short-term memory network to predict the time-series resistance trends of various antibiotics in different pathogens, outputs the predicted resistance rate, and dynamically updates the antibiotic sensitivity map accordingly. This upgrades traditional static statistics to dynamic prediction. It can not only predict drug resistance evolution trends but also ensure that the sensitivity standards used for prescription review are synchronously matched with the dynamic changes in drug resistance in the ward, fundamentally avoiding irrational drug use problems caused by outdated review criteria and mismatch between drug selection and actual drug resistance, significantly improving the timeliness of review references. It also simultaneously acquires patient data. By combining medical record data and infection test indicators to obtain patient infection status vectors through infection assessment, and integrating population-level drug resistance trend data with individual patient infection characteristics, the limitations of single drug susceptibility data are overcome. This allows prescription review to take into account both the overall drug resistance prevalence in the ward and the individual patient infection status, effectively improving the accuracy of prescription review. Based on antibiotic prescriptions to be reviewed, dynamically updated antibiotic susceptibility maps, patient infection status vectors, and preset medical constraint rules, multi-dimensional prescription matching analysis is conducted and a comprehensive review score is output. This incorporates multiple dimensions such as drug resistance suitability, patient infection suitability, and medical rule compliance into a unified evaluation system, enabling comprehensive and timely review of prescription rationality and improving the efficiency and accuracy of drug management. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating an antibiotic prescription review method based on drug resistance trend analysis, provided in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of an antibiotic prescription review device based on drug resistance trend analysis provided in an embodiment of the present invention. Detailed Implementation

[0017] 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. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] See Figure 1 To address the problem that static antibiotic sensitivity in existing technologies cannot adapt to the changing trends of drug-resistant bacteria, leading to delayed and unreasonable drug review, an embodiment of the present invention provides a flowchart of an antibiotic prescription review method based on drug resistance trend analysis, including: S1. Obtain bacterial resistance monitoring data, patient medical record data, infection test indicators, and antibiotic prescriptions pending review; Specifically, bacterial resistance monitoring data comprises the complete set of test results obtained by medical institution microbiology laboratories through antimicrobial susceptibility testing. This includes pathogen identification information, types of antimicrobial drugs, quantitative and qualitative antimicrobial susceptibility results, and sampling information, serving as the fundamental data source for assessing bacterial resistance levels. Patient medical record data consists of complete patient diagnosis and treatment data stored in the electronic medical record system, including basic information, medical history, diagnosis, progress notes, and medication history. This is core data reflecting the individual patient's treatment background. Infection testing indicators represent laboratory test items related to infectious diseases, including white blood cell count, C-reactive protein, procalcitonin, and neutrophil percentage. These are biochemical and blood routine indicators that reflect the degree and trend of infection.

[0019] To illustrate, four types of core data are collected synchronously through a standardized interface with the hospital information system: bacterial resistance monitoring data from the microbiology laboratory information system, patient medical record data from the electronic medical record system, infection testing indicators from the laboratory information system, and antibiotic prescriptions submitted by doctors at their workstations for review. The data collection process simultaneously performs format standardization and de-identification processing to ensure data consistency and privacy. All data undergoes missing value imputation, outlier removal, and standardized coding. Pathogen names and antimicrobial drugs use clinical standard codes and ATC codes respectively to avoid matching errors caused by inconsistent naming.

[0020] To illustrate, patient medical record data and infection detection indicators are updated in real time along with the treatment process to ensure that infection assessment is consistent with the patient's current status. All patient-related data is de-identified, strictly adheres to medical data security regulations, and is used only for prescription review scenarios.

[0021] S2. Based on the bacterial resistance monitoring data, construct a heterogeneous feature matrix that integrates spatial and temporal dimensions. Based on the preset long short-term memory network and the heterogeneous feature matrix, perform time-series prediction of the resistance trends of different pathogens to various antibiotics to obtain the resistance rate prediction results. Update the preset antibiotic sensitivity map based on the resistance rate prediction results. Specifically, the heterogeneous feature matrix is ​​a matrix structure that integrates multi-dimensional and multi-type features. In this embodiment, it includes both spatial dimension (disease area distribution) and temporal dimension (time series information), and the data types cover various heterogeneous forms such as categorical features and numerical features. Long Short-Term Memory (LSTM) networks are suitable for predicting time-dynamic sequence data such as drug resistance rates. The antibiotic susceptibility map represents a drug resistance reference map with pathogen-antimicrobial drug as the core dimension, marking the sensitivity, intermediate, and resistance status of various pathogens to different drugs and their corresponding resistance rates. It serves as a core reference for prescription review regarding drug resistance.

[0022] This paper illustrates how bacterial resistance monitoring data undergoes feature extraction, incorporating both spatial disease distribution information and temporal sequence information to construct a heterogeneous feature matrix that integrates spatiotemporal dimensions. This transforms the statistical problem of resistance into a time-series prediction problem. The heterogeneous feature matrix is ​​then input into a pre-trained Long Short-Term Memory (LSTM) network, which, for each type of antimicrobial drug corresponding to each pathogen, outputs a predicted resistance rate within a fixed future time window. Based on these resistance rate predictions, the existing antibiotic susceptibility map is updated in real-time, replacing the traditional periodic update model.

[0023] To illustrate, bacterial resistance monitoring data supports real-time incremental synchronization; the heterogeneous feature matrix is ​​updated synchronously with each new drug susceptibility result. The Long Short-Term Memory (LSTM) network is pre-trained and validated using 3-5 years of historical drug resistance monitoring data from the hospital. After deployment, it can be incrementally trained periodically based on new data to continuously optimize prediction accuracy.

[0024] Preferably, based on the bacterial resistance monitoring data, a heterogeneous feature matrix integrating spatial and temporal dimensions is constructed, including: Using the distribution of departments, wards, and beds as the spatial dimension of the matrix, and the sampling calendar time, patient hospitalization duration, and antibiotic frequency as the temporal dimension of the matrix, the pathogen species code, specimen source characteristics, antimicrobial drug ATC code, inhibition zone continuity value, sampling timestamp, and ward spatial coordinates in the bacterial resistance monitoring data are extracted to generate the original feature tensor. Based on the pathogen species code and antimicrobial drug ATC code in the original feature tensor, determine the pathogen sample size and antimicrobial drug sample size; When the number of pathogen samples or the number of antimicrobial drug samples is less than a preset sample quantity threshold, the proportion of pathogens or the proportion of antimicrobial drugs is corrected based on the Bayesian smoothing algorithm to obtain the corrected feature tensor, and the corrected feature tensor is used as a heterogeneous feature matrix.

[0025] Specifically, the spatial dimension refers to the physical spatial hierarchy of pathogen sampling and patient distribution, divided into three levels: department, ward, and bed, reflecting the spatial distribution and clustering characteristics of drug-resistant bacteria. The temporal dimension includes the sampling calendar time, patient hospitalization duration, and antibiotic frequency, used to capture the temporal evolution of drug resistance. The original feature tensor can simultaneously carry multi-dimensional information such as pathogens, antimicrobial drugs, space, and time. The continuous value of the inhibition zone is the diameter of the inhibition zone measured by the disk diffusion method for drug susceptibility testing, containing richer drug susceptibility information compared to qualitative results of drug resistance or susceptibility. The ward spatial coordinates represent the coordinate values ​​obtained by spatially encoding the ward and bed, used to quantify the spatial location of the samples and support spatial dimension feature analysis. The preset sample size threshold can be set to 20.

[0026] Schematic representation: the spatial dimension is hierarchically distributed in three levels: department, ward, and bed, with corresponding spatial coordinates assigned to each ward; the temporal dimension incorporates three types of features: sampling calendar time, patient hospitalization duration, and antibiotic frequency; the attribute dimension extracts pathogen type code, specimen source characteristics, antimicrobial drug ATC code, inhibition zone continuity value, and sampling timestamp. All features are organized in a unified dimensional order to generate a three-dimensional original feature tensor, achieving structured integration of multi-source heterogeneous features.

[0027] Then, based on the pathogen coding dimension of the original feature tensor, the effective sample size for each type of pathogen is calculated; based on the antimicrobial drug ATC coding dimension, the effective sample size for each type of drug is calculated. The sample sizes of both types are compared with preset sample size thresholds to identify pathogen or drug categories with insufficient sample sizes. For pathogens with sample sizes less than the threshold, a Bayesian smoothing algorithm is used to correct their resistance ratio under different antimicrobial drugs; similarly, the resistance ratio of antimicrobial drugs with insufficient sample sizes is corrected. The correction process incorporates the overall resistance distribution of the same type of pathogen or drug across the entire hospital as prior information, and combines it with the current small sample observations to calculate the corrected resistance statistics, ultimately generating a corrected feature tensor as the heterogeneous feature matrix used for subsequent predictions.

[0028] In a preferred embodiment of the present invention, the threshold for pathogen samples can be adjusted according to the size of the ward and the number of samples submitted. For example, the threshold for pathogen samples in a general internal medicine ward can be set to 30 samples, while for a small specialized ward it can be lowered to 20 samples, balancing the correction strength with data accuracy. Simultaneously, numerical characteristics such as inhibition zone diameter and length of hospital stay are normalized, while classification characteristics such as pathogen coding and specimen origin are uniquely coded to ensure compatibility across different types of characteristics. Hospital-wide review can adopt a departmental spatial dimension, while refined management of a single ward can be refined to the bed level, adapting to different management granularity requirements.

[0029] In a preferred embodiment of the present invention, taking the construction of a feature matrix for a respiratory ward in a hospital as an example, the respiratory ward is divided into 3 general wards and 1 intensive care ward, and corresponding spatial coordinate codes are assigned. The time dimension includes the monthly sampling time from January to December 2025, the average length of hospital stay for patients, and the average monthly antibiotic usage frequency in each ward. The attribute dimension extracts the codes for 8 common pathogens, two types of specimen source features, 10 types of antimicrobial drug ATC codes, and the corresponding inhibition zone diameter and sampling timestamp, integrating them to generate an 8×10×12 three-dimensional original feature tensor. There are 18 valid samples of Stenotrophomonas maltophilia throughout the year, which is lower than the preset threshold of 20 cases; there are 16 samples corresponding to tigecycline, which is also lower than the threshold. The overall drug resistance distribution of Stenotrophomonas maltophilia in the whole hospital over the past 2 years is introduced as a priori, and combined with the 18 observed samples from the respiratory ward, the corrected drug resistance ratio of each type of drug is calculated; similarly, the drug resistance statistics of tigecycline are corrected. Finally, the corrected feature tensor is obtained as a heterogeneous feature matrix. Verification showed that the corrected resistance rate of Stenotrophomonas maltophilia to levofloxacin was more consistent with the actual resistance level in the disease area, avoiding the overestimation of results due to small sample size.

[0030] Current techniques for statistically analyzing drug susceptibility data typically calculate resistance rates directly over a period of time, neglecting statistical biases from small samples and failing to incorporate spatial distribution and multi-dimensional attribute information. This is especially problematic when dealing with niche pathogens or small disease areas with limited sample sizes; a few resistant samples can cause a sudden spike in resistance rates, resulting in highly volatile and unreliable statistical results. Furthermore, calculating only the average over time fails to identify clustered changes in resistance within disease areas, wasting much valuable information from the data.

[0031] This embodiment integrates multi-dimensional information such as spatial distribution, time series, specimen attributes, and drug attributes into a single matrix, enabling subsequent prediction models to capture more patterns in drug resistance evolution, rather than relying solely on a single time series. For example, previously only the monthly hospital-wide drug resistance rate was considered; now, differences in drug resistance across different wards, specimen types, and drug intensities can be observed, leading to more accurate predictions. Simultaneously, Bayesian smoothing with small sample correction addresses the distortion caused by insufficient samples from niche pathogens or small wards. Previously, with limited samples, statistical results fluctuated wildly and were unusable; now, smoothed using prior information, the results are more stable and reflect reality, preventing data fluctuations from skewing predictions and ensuring the reliability of prescription review.

[0032] Preferably, based on a preset long short-term memory network and the heterogeneous feature matrix, time-series predictions are made on the resistance trends of different pathogens to various antibiotics to obtain resistance rate prediction results, and a preset antibiotic susceptibility map is updated based on the resistance rate prediction results, including: The heterogeneous feature matrix is ​​input into a preset long short-term memory network. Through a multi-head attention mechanism, weights are assigned to the features corresponding to recent drug-sensitive samples in the same disease area. Based on the weights, time-series prediction is performed to obtain the drug resistance rate prediction result. The drug resistance rate prediction result includes the predicted drug resistance rate of each pathogen to various antibiotics. The predicted drug resistance rate is compared with the preset single drug warning threshold to obtain the first comparison result, and the antimicrobial drug marker warning label in the preset antibiotic susceptibility map is updated according to the first comparison result. The predicted drug resistance rate is compared with the preset drug use restriction threshold to obtain a second comparison result. The drug use priority of antimicrobial drugs in the preset antibiotic susceptibility map is updated according to the second comparison result to obtain the updated antibiotic susceptibility map.

[0033] Specifically, the single-drug warning threshold is a preset resistance rate threshold. When the predicted resistance rate reaches this value, a warning should be marked on the sensitivity map to prompt clinicians to use the drug with caution. The drug restriction threshold is a higher-level resistance rate threshold. When the predicted resistance rate exceeds this value, the drug priority should be adjusted to guide clinicians to switch to more sensitive drugs. The antimicrobial drug priority indicates the ranking of similar antimicrobial drugs according to their resistance rate, prioritizing drugs with low resistance rates and good sensitivity to guide clinicians in rational drug selection.

[0034] Intuitively, a heterogeneous feature matrix is ​​input into a pre-trained Long Short-Term Memory (LSTM) network. During the model's processing of temporal features, an embedded multi-head attention mechanism automatically assigns higher weights to recent drug-susceptibility samples from the same disease area, while reducing the weight ratio of distant samples to samples from other disease areas. This makes the prediction results more closely match the recent drug resistance trends of the target disease area. The model outputs the predicted drug resistance rate for each type of pathogen and each type of antimicrobial drug within a fixed future time window, forming a complete set of drug resistance rate prediction results.

[0035] Then, the predicted resistance rate of each pathogen-antimicrobial drug group is compared with the preset single-drug warning threshold. If the threshold is reached or exceeded, a warning mark is added to the corresponding position on the sensitivity map, indicating an increased risk of resistance. If the threshold is not reached, the normal marking is maintained, completing the first-level map update. The predicted resistance rate is then compared with the preset drug use restriction threshold. If the threshold is reached or exceeded, the drug's priority in the treatment of the corresponding pathogen is lowered, and it is adjusted to an alternative drug, while the priority of similar drugs with better sensitivity is increased. If the threshold is not reached, the original ranking is maintained, and the dynamic update of the sensitivity map is finally completed.

[0036] For illustration, routine reviews can have a one-week window, while ward-based drug resistance control can have a one-month window, adapting to different scenario needs. The single-drug warning threshold for key controlled drugs such as carbapenems can be set at 20%, and the drug use restriction threshold at 30%; for ordinary non-restricted drugs, the single-drug warning threshold can be set at 30%, and the drug use restriction threshold at 50%, matching the management requirements of different drugs.

[0037] In a preferred embodiment of the present invention, five strains of imipenem-resistant Acinetobacter baumannii were detected consecutively in the Intensive Care Unit of a hospital in early October 2025. A heterogeneous feature matrix containing the latest drug susceptibility data was input into a Long Short-Term Memory (LSTM) network. The model, through a multi-head attention mechanism, assigned a high weight of 0.6 to samples from the intensive care unit within the last 7 days, a weight of 0.3 to samples within the last 30 days, and a weight of 0.1 to earlier samples. The predicted resistance rate of Acinetobacter baumannii to imipenem for the next two weeks was output as 42.1%. Comparing this to the single-drug warning threshold (preset 20%), the predicted resistance rate far exceeded the threshold. The system added a red warning marker to the Acinetobacter baumannii-imipenem entry corresponding to the intensive care unit in the susceptibility map. Comparing this to the drug restriction threshold (preset 30%), the predicted resistance rate had already exceeded the threshold. The system downgraded the priority of imipenem in the treatment of Acinetobacter baumannii from 2nd to 5th place, while also adjusting the priority of other drugs. After the update, when clinicians prescribe relevant medications in the intensive care unit, the system will automatically refer to the updated atlas, highlight imipenem prescriptions, and guide them to prioritize highly sensitive drugs.

[0038] Existing antibiotic susceptibility profiles are static, reflecting only average levels over a period of time. They cannot predict rising resistance rates or differentiate markers for different disease areas, leaving clinical drug selection entirely based on experience. Moreover, most profiles only provide qualitative resistance or susceptibility markers, lacking tiered warnings and priority adjustments, thus offering very limited practical guidance for clinical drug selection.

[0039] In this embodiment, a long short-term memory network combined with multi-head attention focuses on recent data from the same ward, resulting in predictions that more closely reflect the actual trends in that ward. Simply put, the model prioritizes drug susceptibility results from the past week or two within the ward, rather than averaging data from several months ago with data from other wards. This leads to greater accuracy and specificity in predictions, enabling early detection of rising drug resistance in a particular ward. Dual-threshold grading not only indicates risk but also directly adjusts medication priorities. Previously, the drug resistance profile only told doctors about drug resistance; now it tells them that a drug has a high resistance rate and to prioritize other options, providing more specific guidance and truly guiding doctors to select sensitive drugs, reducing the pressure of resistance screening. Simultaneously, the profile is dynamically and automatically updated, eliminating the need for quarterly manual statistics. The profile is updated synchronously as soon as the drug resistance trend changes, fundamentally solving the problem of outdated reference data.

[0040] S3. Based on the patient's medical record data and the infection test indicators, perform an infection assessment to obtain the patient's infection status vector; Specifically, the patient infection status vector is a feature vector obtained by vectorizing the patient's infection-related clinical characteristics, which can quantitatively reflect the patient's infection severity, infection type, progression trend, and other statuses.

[0041] To illustrate, by integrating diagnostic and treatment information and infection test indicators from patient medical records, infection-related features are extracted from both textual and numerical information dimensions. Through feature fusion, a standardized patient infection status vector is generated, thereby achieving a quantitative representation of the individual patient's infection status.

[0042] Preferably, infection assessment is performed based on the patient's medical record data and the infection test indicators to obtain a patient infection status vector, including: Entity extraction is performed on the unstructured medical history text in the patient's medical record data to obtain the semantic features of the case. The time-series values, the slope of index changes, and the fluctuation range are extracted from the infection test indicators as time-series test features. The patient's infection status vector is obtained by fusing features based on the semantic features of the case and the temporal test features.

[0043] Specifically, unstructured medical record text refers to unformatted medical records and admission notes written by doctors in electronic medical records. This includes a large amount of semantic information describing the patient's symptoms, signs, and changes in condition. The slope of indicator changes reflects the upward or downward trend of the indicator and is an important basis for judging the progression or improvement of infection. The fluctuation range reflects the stability of the condition.

[0044] This paper illustrates how unstructured texts such as patient medical records, admission notes, and physical examination records are preprocessed. A medical entity recognition model is then used to extract infection-related entities, including infection symptoms, signs, infection sites, and suspected pathogen types. The extracted entities are standardized and vectorized to obtain semantic features of the case, thus transforming unstructured text into structured features. The medical record text undergoes word segmentation, stop word removal, and medical terminology standardization; for example, "fever" is uniformly mapped to "excessive fever," improving the accuracy of entity extraction. Patient past medical history and immune status can be incorporated into the semantic features to enhance the comprehensiveness of the assessment.

[0045] Then, all test values ​​within the target observation period are extracted from the infection testing indicators to form a time-series numerical sequence. Based on the numerical sequence, the slope of change of adjacent test values ​​is calculated to reflect the trend of indicator change. At the same time, the maximum value, minimum value and fluctuation range of the indicator within the observation period are calculated to reflect the stability of the indicator. The three types of features are integrated to obtain the time-series testing features.

[0046] This study employs a combination of feature concatenation and weighted fusion to integrate case semantic features with temporal test features. First, the two types of features are aligned and standardized. Then, they are weighted and fused according to preset weight ratios to generate a unified patient infection status vector. Each dimension corresponds to the dimensions of infection severity, infection type, progression trend, and suspected pathogen type. Specifically, the weight of case semantic features can be appropriately increased for respiratory infections, and the weight of temporal test features can be appropriately increased for bloodstream infections, adapting to the assessment characteristics of different infection types.

[0047] Current prescription review processes generally lack individualized infection status assessments. At most, they check for an infection diagnosis, without delving into the severity and progression of the infection. Applying a single standard to all patients easily leads to overly strict or lenient reviews. For example, the rationale for using the same type of antibiotic differs significantly between mild and severe infections, but current technology cannot differentiate between them.

[0048] This embodiment simultaneously mines textual semantics and test sequence information, utilizing doctors' written medical records. This essentially combines doctors' clinical judgment with objective test data, resulting in an infection status assessment that more closely reflects clinical reality. For example, some patients may not have significantly elevated test indicators, but their symptoms and signs already suggest a severe infection; this can be captured through semantic features, avoiding missed diagnoses. Quantifying the infection status using vectors allows for direct matching calculations with prescription and drug resistance data, achieving personalized and precise review. Previously, reviews were one-size-fits-all; now, the matching scale can be adjusted according to each patient's infection status, better aligning with the actual logic of clinical diagnosis and treatment, significantly improving the rationality of the review.

[0049] S4. Perform multi-dimensional prescription matching analysis based on the antibiotic prescription to be reviewed, the updated antibiotic sensitivity map, the patient infection status vector, and the preset medical constraint rules to obtain a comprehensive review score; Specifically, medical constraint rules refer to a set of medication rules based on guidelines for the clinical application of antimicrobial drugs, drug instructions, and hospital management systems. These rules include indications and contraindications, dosage restrictions, contraindications for combined drug use, and hierarchical management requirements. The comprehensive review score represents a quantitative score calculated through a weighted average of multiple indicators. It is used to comprehensively measure the rationality of the prescription; a higher comprehensive review score indicates a stronger overall rationality of the prescription.

[0050] To illustrate, based on the medication information of the antibiotic prescription to be reviewed, the updated antibiotic sensitivity map, the patient infection status vector, and the preset medical constraint rules are matched and analyzed from multiple dimensions such as drug resistance risk, infection compatibility, medication safety and management compliance. After calculating the scores of each item, the comprehensive review score is obtained by integrating them through weighted rules, so as to realize the quantitative evaluation of the rationality of the prescription.

[0051] Preferably, a multi-dimensional prescription matching analysis is performed based on the antibiotic prescription to be reviewed, the updated antibiotic susceptibility profile, the patient's infection status vector, and preset medical constraint rules to obtain a comprehensive review score, including: Extract the drug's antibacterial spectrum, antibiotic grade, dosage, type of combined medication, and pathogen identification from the antibiotic prescription to be reviewed to generate the basic characteristics of the prescription. Based on the basic characteristics of the prescription and the updated antibiotic susceptibility profile, a resistance risk score is calculated. Based on the basic features of the prescription and the patient's infection status vector, the infection indicator matching score is calculated; The basic characteristics of the prescription are verified for safety according to preset medical constraint rules to obtain a medication safety score; The antibiotic classification score is determined based on the antibiotic classification level in the basic characteristics of the prescription. Based on the corresponding weighting coefficients, the infection indication matching score, the medication safety score, and the antibiotic classification score are positively weighted, and the drug resistance risk score is negatively weighted to obtain the comprehensive review score. Specifically, when performing security verification on the basic features of the prescription according to preset medical constraint rules, if a hard medical rule in the medical constraint rules is triggered, the medication safety score will be directly set to 0.

[0052] Specifically, the antimicrobial spectrum of a drug refers to the range of pathogens that an antimicrobial drug can cover, and it is the core basis for assessing the match between the drug and the infecting pathogen. Pathogen submission identification indicates whether a patient has submitted a pathogen specimen for testing before medication, and it is an important assessment indicator for the proper use of antimicrobial drugs. The drug resistance risk score is a quantitative score measuring the risk of resistance to the target pathogen by the prescribed drug; a higher score indicates a greater risk of resistance and a lower degree of prescription rationality. Violation of strict medical rules can directly lead to serious medication safety problems, such as drug allergies, serious contraindications, and overdosing; prescriptions that trigger such rules are directly deemed unqualified.

[0053] To illustrate, the antibacterial spectrum, antibiotic classification, type of combined medication, and pathogen submission status are extracted from antibiotic prescriptions awaiting review. This information is then structured and encoded to generate a set of basic prescription features.

[0054] Then, the antibiotics in the prescription are matched with the updated antibiotic susceptibility map based on the pathogen information to obtain the corresponding predicted resistance rate. The higher the resistance rate, the higher the resistance risk score. The prescription's antibacterial spectrum and indication information are matched with the patient's infection status vector, and scored from three dimensions: infection site matching degree, pathogen coverage, and infection severity suitability, to obtain an infection indication matching score. The higher the matching degree, the higher the score. Medical constraint rules are invoked to perform safety checks on the prescription's usage, dosage, combined medication, contraindications, allergy history, etc. If a medical hard rule is triggered, the medication safety score is directly set to 0; otherwise, the score is calculated based on the degree of rule compliance. According to the antibiotic classification, combined with the prescribing physician's prescription authority and the patient's indication compliance score, the antibiotic classification score is high if the classification management requirements are met, and low if the use exceeds authority.

[0055] Finally, corresponding weighting coefficients were assigned to each sub-item. The infection indication matching score, medication safety score, and antibiotic classification score were positive scores, with positive weighting coefficients; the higher the score, the greater the positive contribution. The drug resistance risk score was a negative score, with a negative weighting coefficient; the higher the risk, the greater the negative impact. The scores of each sub-item were multiplied by their corresponding coefficients and then summed to obtain the final comprehensive review score. All sub-item scores were uniformly mapped to a range of 0-100 points, and the comprehensive review score was also mapped to 0-100 points; the higher the score, the better the reasonableness.

[0056] In a preferred embodiment of the present invention, a doctor prescribes cefepime for intravenous infusion to a patient with Klebsiella pneumoniae pneumonia. The system performs multi-dimensional scoring: the antibacterial spectrum covers Gram-negative bacilli, restricted use level, dosage and frequency are in accordance with routine, monotherapy is being administered, and sputum samples have been collected and sent for testing. Then, the sub-scores are calculated and matched against an updated antibiotic resistance map. The predicted resistance rate of Klebsiella pneumoniae to cefepime in this ward is 22%, corresponding to a resistance risk score of 28. The patient has a moderate pulmonary bacterial infection, the antibacterial spectrum covers the target pathogen, the indication matches, and the infection indication matching score is 88. The dosage and frequency comply with the instructions, there is no history of allergies or contraindications to combined medication, no strict rules are triggered, and the medication safety score is 92. The doctor has restricted prescription authority, the indication meets the requirements, and the antibiotic classification score is 90. The weighting coefficients are: resistance risk -0.2, infection matching 0.35, medication safety 0.25, and classification management 0.2. The calculated comprehensive review score is 88×0.35+92×0.25+90×0.2-28×0.2=66.2 points.

[0057] Current prescription review processes mostly rely on isolated rule checks, such as grading and dosage. These checkpoints lack correlation, making it impossible to form a comprehensive assessment. This easily leads to situations where one item meets the criteria but the overall prescription is unreasonable, or a minor issue exists but the overall prescription is acceptable, resulting in a very one-sided conclusion. Furthermore, the correlation between drug resistance risk and the rationality of medication use is not quantified, failing to reflect the impact of drug resistance trends on the review process.

[0058] This embodiment incorporates drug resistance risk, infection matching, medication safety, and tiered management into a multi-dimensional scoring system, resulting in a more comprehensive and objective assessment that avoids the bias of a single rule. The positive and negative weighted calculation method aligns with clinical understanding: higher drug resistance risk correlates with lower rationality, while higher infection matching correlates with higher rationality, more accurately reflecting the actual rationality of the prescription. The mechanism of setting strict medical rules to zero safeguards the bottom line of medication safety. Simply put, serious problems are directly blocked, while minor issues are reflected through scoring. This approach neither overlooks safety risks nor rejects prescriptions due to minor issues, balancing safety and efficiency.

[0059] S5. Based on the comprehensive review score, review the prescription and generate a prescription review record.

[0060] To illustrate, the corresponding review processing operations are performed based on the comprehensive review score, and the scoring results and processing operations of the review process are fully recorded to form a traceable prescription review record, which meets the quality control and traceability requirements of antimicrobial drug management.

[0061] Preferably, prescription review is performed based on the comprehensive review score, and a prescription review record is generated, including: The comprehensive review score is compared with a preset first interception threshold and a second interception threshold. If the overall review score is less than the first interception threshold, a prescription interception operation is performed and a serious alarm message is generated; If the overall review score is greater than the second interception threshold, the prescription review is approved. If the overall review score is not less than the first interception threshold and not greater than the second interception threshold, a manual prescription review operation will be performed. Based on the comprehensive review score, the corresponding prescription operation, and the operation time, a prescription review record is generated.

[0062] Specifically, the first interception threshold is the minimum critical value of the comprehensive review score. A score below this value indicates a serious problem with the prescription's reasonableness and requires direct interception. The second interception threshold is the passing threshold of the comprehensive review score. A score above this value indicates a good degree of reasonableness and allows for direct approval. Prescription interception means directly rejecting the prescription and refusing to dispense the medication, while simultaneously sending an alert to the prescribing physician. Manual review means neither approving nor intercepting the prescription, but submitting it to a dedicated pharmacist for manual review; this is used to handle borderline prescriptions.

[0063] Indicatively, the comprehensive review score is compared with two preset interception thresholds to determine the score range. These two thresholds divide the score into three intervals: below the first interception threshold is considered seriously unacceptable; between the two thresholds is considered pending review; and above the second interception threshold is considered acceptable. When the comprehensive review score is less than the first interception threshold, the prescription is deemed seriously unreasonable, the system automatically intercepts it, the prescription cannot be submitted to the pharmacy, and a serious alarm is generated and pushed to the doctor's end, clearly explaining the reason for the unreasonableness and suggestions for rectification. When the comprehensive review score is greater than the second interception threshold, the prescription is deemed reasonable, the system automatically approves it, and the prescription enters the normal dispensing process. When the comprehensive review score is between the two thresholds, the prescription is deemed to have a reasonableness dispute, the system automatically submits it to the dedicated pharmacist's end, and the pharmacist makes the final judgment based on clinical practice. Regardless of the operation performed, the system automatically integrates the comprehensive score, processing operation, operation time, alarm information, or review opinions to generate a standardized prescription review record, which is stored in the antimicrobial drug management database, supporting query, statistics, and quality control analysis.

[0064] In a preferred embodiment of the present invention, a hospital sets a first interception threshold of 40 points and a second interception threshold of 70 points. A prescription for special-use meropenem has a comprehensive review score of 32 points, which is below the first interception threshold. The system directly intercepts the prescription and sends an alert to the doctor: the predicted resistance rate of Pseudomonas aeruginosa to meropenem in this ward is 45%, indicating a high risk. Since the patient has no indication for special-use meropenem, it is recommended to switch to another medication. Simultaneously, an audit record containing detailed scoring, alert content, and interception time is generated. A prescription for non-restricted amoxicillin capsules has a comprehensive review score of 85 points, which is above the second interception threshold. The system automatically approves the prescription, and it is transferred to the pharmacy normally, with an approval record generated simultaneously. A prescription for restricted levofloxacin has a comprehensive review score of 58 points, which is in the pending review range. The system automatically submits the prescription to the anti-infective pharmacist's workbench, noting: the infection indication matching is average, the drug resistance risk is moderate, and manual review is recommended. After reviewing the patient's condition and drug resistance data, the pharmacist determined that the medication met the indications, approved it, and entered their opinion, ultimately generating a complete review record that included the pharmacist's review opinion.

[0065] This implementation adopts a three-tiered review system, with dual thresholds dividing prescriptions into three processing levels. Severe cases are directly blocked to maintain the minimum standard, compliant prescriptions are directly approved to improve efficiency, and borderline cases are handled by manual review to ensure flexibility. This approach ensures medication safety and drug resistance control requirements while allowing reasonable flexibility for clinical practice, preventing excessive review from affecting treatment and making it more acceptable to clinicians. Tiered processing also improves review efficiency; most reasonable prescriptions are automatically approved, with only a small number of borderline prescriptions requiring manual processing. This reduces pharmacists' workload, allowing them to focus on reviewing complex prescriptions and improving overall management efficiency.

[0066] By implementing this embodiment, a heterogeneous feature matrix integrating spatial and temporal dimensions is constructed. Combined with a long short-term memory network, this matrix performs time-series prediction of antibiotic resistance trends in different pathogens, outputting resistance rate prediction results and dynamically updating the antibiotic sensitivity map. This upgrades traditional static statistics to dynamic prediction. It not only predicts resistance evolution trends but also ensures that the sensitivity standards used for prescription review are synchronized with the dynamic changes in ward resistance, fundamentally avoiding irrational drug use due to mismatch between drug selection and actual resistance conditions caused by outdated review criteria. This significantly improves the timeliness of review references. Simultaneously, patient medical record data and infection test indicators are acquired for infection assessment. By obtaining patient infection status vectors and combining population-level drug resistance trend data with individual patient infection characteristics, the limitations of single drug sensitivity data are overcome. This allows prescription review to take into account both the overall drug resistance prevalence in the ward and the individual patient infection status, effectively improving the accuracy of prescription review. Based on antibiotic prescriptions to be reviewed, dynamically updated antibiotic sensitivity maps, patient infection status vectors, and preset medical constraint rules, multi-dimensional prescription matching analysis is conducted and a comprehensive review score is output. Multiple dimensions such as drug resistance suitability, patient infection suitability, and medical rule compliance are incorporated into a unified evaluation system, which can comprehensively and promptly review the rationality of prescriptions and improve the efficiency and accuracy of drug management.

[0067] See Figure 2 This is a schematic diagram of the structure of an antibiotic prescription review device based on drug resistance trend analysis according to an embodiment of the present invention, comprising: The data acquisition module is used to acquire bacterial resistance monitoring data, patient medical record data, infection test indicators, and antibiotic prescriptions pending review. The drug resistance trend analysis module is used to construct a heterogeneous feature matrix that integrates spatial and temporal dimensions based on the bacterial drug resistance monitoring data. Based on the preset long short-term memory network and the heterogeneous feature matrix, it performs time-series prediction of the drug resistance trends of different pathogens against various antibiotics, obtains the drug resistance rate prediction results, and updates the preset antibiotic susceptibility map based on the drug resistance rate prediction results. The infection assessment module is used to assess the infection based on the patient's medical record data and the infection test indicators, and obtain the patient's infection status vector. The comprehensive review score analysis module is used to perform multi-dimensional prescription matching analysis based on the antibiotic prescription to be reviewed, the updated antibiotic sensitivity map, the patient infection status vector, and preset medical constraint rules to obtain a comprehensive review score. The prescription review record generation module is used to review prescriptions based on the comprehensive review score and generate prescription review records.

[0068] Preferably, the drug resistance trend analysis module is used to perform time-series prediction of the drug resistance trends of different pathogens against various antibiotics based on a preset long short-term memory network and the heterogeneous feature matrix, to obtain drug resistance rate prediction results, and to update the preset antibiotic susceptibility map based on the drug resistance rate prediction results, including: The heterogeneous feature matrix is ​​input into a preset long short-term memory network. Through a multi-head attention mechanism, weights are assigned to the features corresponding to recent drug-sensitive samples in the same disease area. Based on the weights, time-series prediction is performed to obtain the drug resistance rate prediction result. The drug resistance rate prediction result includes the predicted drug resistance rate of each pathogen to various antibiotics. The predicted drug resistance rate is compared with the preset single drug warning threshold to obtain the first comparison result, and the antimicrobial drug marker warning label in the preset antibiotic susceptibility map is updated according to the first comparison result. The predicted drug resistance rate is compared with the preset drug use restriction threshold to obtain a second comparison result. The drug use priority of antimicrobial drugs in the preset antibiotic susceptibility map is updated according to the second comparison result to obtain the updated antibiotic susceptibility map.

[0069] This invention provides an antibiotic prescription review device based on drug resistance trend analysis. By constructing a heterogeneous feature matrix integrating spatial and temporal dimensions, and combining it with a long short-term memory network, it performs time-series prediction of various antibiotic resistance trends for different pathogens, outputting the predicted resistance rate results and dynamically updating the antibiotic sensitivity map accordingly. This upgrades traditional static statistics to dynamic prediction. It can not only predict drug resistance evolution trends but also ensure that the sensitivity standards used for prescription review are synchronously matched with the dynamic changes in drug resistance in the ward, fundamentally avoiding irrational drug use problems caused by outdated review criteria and mismatch between drug selection and actual drug resistance, significantly improving the timeliness of review references. It also simultaneously acquires patient medical record data and infection test results. The indicator, obtained through infection assessment to obtain a patient infection status vector, combines population-level drug resistance trend data with individual patient infection characteristics, breaking the limitations of single drug sensitivity data. This allows prescription review to take into account both the overall drug resistance prevalence in the ward and the individual patient infection status, effectively improving the accuracy of prescription review. Based on the antibiotic prescription to be reviewed, dynamically updated antibiotic sensitivity maps, patient infection status vectors, and preset medical constraint rules, multi-dimensional prescription matching analysis is carried out and a comprehensive review score is output. Multiple dimensions such as drug resistance suitability, patient infection suitability, and medical rule compliance are incorporated into a unified evaluation system, which can comprehensively and timely review the rationality of prescriptions and improve the efficiency and accuracy of drug management.

[0070] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0071] Those skilled in the art will understand that, for convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0072] Another embodiment of the present invention provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements an antibiotic prescription review method based on drug resistance trend analysis as described in the above embodiments. The terminal device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The terminal device may include, but is not limited to, a processor and a memory.

[0073] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the terminal device, connecting all parts of the terminal device via various interfaces and lines.

[0074] The memory can be used to store the computer program. The processor implements various functions of the terminal device by running or executing the computer program stored in the memory and calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device or other volatile solid-state storage device.

[0075] Another embodiment of the present invention provides a computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform an antibiotic prescription review method based on drug resistance trend analysis as described in the above embodiment.

[0076] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0077] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for reviewing antibiotic prescriptions based on drug resistance trend analysis, characterized in that, include: Obtain bacterial resistance monitoring data, patient medical record data, infection test indicators, and antibiotic prescriptions pending review; Based on the bacterial resistance monitoring data, a heterogeneous feature matrix integrating spatial and temporal dimensions is constructed. Based on the preset long short-term memory network and the heterogeneous feature matrix, the resistance trends of different pathogens to various antibiotics are predicted in time series to obtain the resistance rate prediction results. The preset antibiotic sensitivity map is updated based on the resistance rate prediction results. Infection assessment is performed based on the patient's medical record data and the infection test indicators to obtain the patient's infection status vector; A multi-dimensional prescription matching analysis is performed based on the antibiotic prescription to be reviewed, the updated antibiotic sensitivity map, the patient's infection status vector, and the preset medical constraint rules to obtain a comprehensive review score. Prescriptions are reviewed based on the comprehensive review score, and prescription review records are generated.

2. The antibiotic prescription review method based on drug resistance trend analysis as described in claim 1, characterized in that, Based on the bacterial resistance monitoring data, a heterogeneous feature matrix integrating spatial and temporal dimensions is constructed, including: Using the distribution of departments, wards, and beds as the spatial dimension of the matrix, and the sampling calendar time, patient hospitalization duration, and antibiotic frequency as the temporal dimension of the matrix, the pathogen species code, specimen source characteristics, antimicrobial drug ATC code, inhibition zone continuity value, sampling timestamp, and ward spatial coordinates in the bacterial resistance monitoring data are extracted to generate the original feature tensor. Based on the pathogen species code and antimicrobial drug ATC code in the original feature tensor, determine the pathogen sample size and antimicrobial drug sample size; When the number of pathogen samples or the number of antimicrobial drug samples is less than a preset sample quantity threshold, the proportion of pathogens or the proportion of antimicrobial drugs is corrected based on the Bayesian smoothing algorithm to obtain the corrected feature tensor, and the corrected feature tensor is used as a heterogeneous feature matrix.

3. The antibiotic prescription review method based on drug resistance trend analysis as described in claim 1, characterized in that, Based on a pre-defined long short-term memory network and the heterogeneous feature matrix, time-series predictions are made of the resistance trends of different pathogens to various antibiotics to obtain resistance rate prediction results. The pre-defined antibiotic susceptibility map is then updated based on these resistance rate prediction results, including: The heterogeneous feature matrix is ​​input into a preset long short-term memory network. Through a multi-head attention mechanism, weights are assigned to the features corresponding to recent drug-sensitive samples in the same disease area. Based on the weights, time-series prediction is performed to obtain the drug resistance rate prediction result. The drug resistance rate prediction result includes the predicted drug resistance rate of each pathogen to various antibiotics. The predicted drug resistance rate is compared with the preset single drug warning threshold to obtain the first comparison result, and the antimicrobial drug marker warning label in the preset antibiotic susceptibility map is updated according to the first comparison result. The predicted drug resistance rate is compared with the preset drug use restriction threshold to obtain a second comparison result. The drug use priority of antimicrobial drugs in the preset antibiotic susceptibility map is updated according to the second comparison result to obtain the updated antibiotic susceptibility map.

4. The antibiotic prescription review method based on drug resistance trend analysis as described in claim 1, characterized in that, Infection assessment is performed based on the patient's medical record data and the infection test indicators to obtain a patient infection status vector, including: Entity extraction is performed on the unstructured medical history text in the patient's medical record data to obtain the semantic features of the case. The time-series values, the slope of index changes, and the fluctuation range are extracted from the infection test indicators as time-series test features. The patient's infection status vector is obtained by fusing features based on the semantic features of the case and the temporal test features.

5. The antibiotic prescription review method based on drug resistance trend analysis as described in claim 1, characterized in that, A multi-dimensional prescription matching analysis is performed based on the antibiotic prescription to be reviewed, the updated antibiotic sensitivity profile, the patient's infection status vector, and preset medical constraint rules to obtain a comprehensive review score, including: Extract the drug's antibacterial spectrum, antibiotic grade, dosage, type of combined medication, and pathogen identification from the antibiotic prescription to be reviewed to generate the basic characteristics of the prescription. Based on the basic characteristics of the prescription and the updated antibiotic susceptibility profile, a resistance risk score is calculated. Based on the basic features of the prescription and the patient's infection status vector, the infection indicator matching score is calculated; The basic characteristics of the prescription are verified for safety according to preset medical constraint rules to obtain a medication safety score; The antibiotic classification score is determined based on the antibiotic classification level in the basic characteristics of the prescription. Based on the corresponding weighting coefficients, the infection indication matching score, the medication safety score, and the antibiotic classification score are positively weighted, and the drug resistance risk score is negatively weighted to obtain the comprehensive review score. Specifically, when performing security verification on the basic features of the prescription according to preset medical constraint rules, if a hard medical rule in the medical constraint rules is triggered, the medication safety score will be directly set to 0.

6. The antibiotic prescription review method based on drug resistance trend analysis as described in claim 1, characterized in that, Prescriptions are reviewed based on the comprehensive review score, and a prescription review record is generated, including: The comprehensive review score is compared with a preset first interception threshold and a second interception threshold. If the overall review score is less than the first interception threshold, a prescription interception operation is performed and a serious alarm message is generated; If the overall review score is greater than the second interception threshold, the prescription review is approved. If the overall review score is not less than the first interception threshold and not greater than the second interception threshold, a manual prescription review operation will be performed. Based on the comprehensive review score, the corresponding prescription operation, and the operation time, a prescription review record is generated.

7. An antibiotic prescription review device based on drug resistance trend analysis, characterized in that, include: The data acquisition module is used to acquire bacterial resistance monitoring data, patient medical record data, infection test indicators, and antibiotic prescriptions pending review. The drug resistance trend analysis module is used to construct a heterogeneous feature matrix that integrates spatial and temporal dimensions based on the bacterial drug resistance monitoring data. Based on the preset long short-term memory network and the heterogeneous feature matrix, it performs time-series prediction of the drug resistance trends of different pathogens against various antibiotics, obtains the drug resistance rate prediction results, and updates the preset antibiotic susceptibility map based on the drug resistance rate prediction results. The infection assessment module is used to assess the infection based on the patient's medical record data and the infection test indicators, and obtain the patient's infection status vector. The comprehensive review score analysis module is used to perform multi-dimensional prescription matching analysis based on the antibiotic prescription to be reviewed, the updated antibiotic sensitivity map, the patient infection status vector, and preset medical constraint rules to obtain a comprehensive review score. The prescription review record generation module is used to review prescriptions based on the comprehensive review score and generate prescription review records.

8. The antibiotic prescription review device based on drug resistance trend analysis as described in claim 7, characterized in that, The drug resistance trend analysis module is used to perform time-series prediction of the drug resistance trends of different pathogens against various antibiotics based on a preset long short-term memory network and the heterogeneous feature matrix, obtain drug resistance rate prediction results, and update the preset antibiotic susceptibility map based on the drug resistance rate prediction results, including: The heterogeneous feature matrix is ​​input into a preset long short-term memory network. Through a multi-head attention mechanism, weights are assigned to the features corresponding to recent drug-sensitive samples in the same disease area. Based on the weights, time-series prediction is performed to obtain the drug resistance rate prediction result. The drug resistance rate prediction result includes the predicted drug resistance rate of each pathogen to various antibiotics. The predicted drug resistance rate is compared with the preset single drug warning threshold to obtain the first comparison result, and the antimicrobial drug marker warning label in the preset antibiotic susceptibility map is updated according to the first comparison result. The predicted drug resistance rate is compared with the preset drug use restriction threshold to obtain a second comparison result. The drug use priority of antimicrobial drugs in the preset antibiotic susceptibility map is updated according to the second comparison result to obtain the updated antibiotic susceptibility map.

9. A terminal device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements an antibiotic prescription review method based on resistance trend analysis as described in any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device on which the computer-readable storage medium is located to perform an antibiotic prescription review method based on drug resistance trend analysis as described in any one of claims 1 to 6.

Images