Intelligent recommendation method and system for inpatient dynamic thrombosis prevention measures

Abstract

This invention provides a method and system for intelligent recommendation of dynamic thrombosis prevention measures for hospitalized patients, relating to the field of medical technology. The method involves acquiring patient physiological monitoring, medication records, and activity status data; performing multi-dimensional correlation analysis based on a pre-set risk assessment model; extracting feature vectors to obtain risk assessment results; matching and associating these results with a knowledge base of prevention measures to generate recommendation results; and adaptively adjusting the model based on actual thrombotic event data. This invention achieves intelligent recommendation of thrombosis prevention measures, improving the accuracy and effectiveness of preventative interventions.

Description

Technical Field

[0001] This invention relates to medical technology, and more particularly to a method and system for intelligent recommendation of dynamic thrombosis prevention measures for hospitalized patients. Background Technology

[0002] Venous thromboembolism (VTE) is a common clinical complication in hospitalized patients, including deep vein thrombosis and pulmonary embolism, which can significantly increase hospital stay, medical costs, and the risk of death. Epidemiological surveys show that VTE is one of the leading causes of unexpected deaths in hospitals, with approximately 70% of hospitalized patients at varying degrees of thrombotic risk. Implementing preventative measures is crucial to reducing this risk.

[0003] Traditional thrombosis prevention in hospitalized patients typically relies on healthcare professionals conducting risk assessments based on clinical guidelines, such as the Caprini or Padua scores, and developing corresponding prevention plans. These preventative measures mainly include physical preventative methods (such as compression stockings and intermittent pneumatic compression pumps) and pharmacological preventative methods (such as low molecular weight heparin and direct oral anticoagulants). However, current thrombosis prevention management has several shortcomings.

[0004] Most existing methods for assessing thrombosis risk rely on static data and cannot respond in real time to changes in a patient's condition. Healthcare professionals typically conduct a one-time risk assessment upon admission, lacking continuous monitoring and analysis of dynamic data such as the patient's physiological status, medication use, and activity level during hospitalization. This results in inaccurate and untimely risk assessments.

[0005] Traditional preventive measures recommendations lack individualization and precision. Existing systems struggle to comprehensively analyze multidimensional patient data and deeply integrate it with a professional knowledge base, resulting in recommended preventive measures that are not entirely applicable to specific patients. They fail to fully consider individual patient differences and contraindications, increasing the risk of inappropriate treatment. Summary of the Invention

[0006] This invention provides a method and system for intelligent recommendation of dynamic thrombosis prevention measures for hospitalized patients, which can solve the problems in the prior art.

[0007] A first aspect of the present invention provides a method for intelligent recommendation of dynamic thrombosis prevention measures for hospitalized patients, comprising: Acquire physiological monitoring data, medication record data, and activity status data of the target patients; Based on a preset risk assessment model, a multi-dimensional correlation analysis is performed on the physiological monitoring data, the medication record data, and the activity status data to extract feature vectors that characterize the risk of thrombosis and obtain risk assessment results. Based on the risk level identifier in the risk assessment results, the target patient is matched and associated with the prevention measures in the prevention measures knowledge base to obtain a set of candidate prevention measures; and based on the set of candidate prevention measures and the contraindications of the target patient, a recommendation result of applicable prevention measures is generated. Based on the recommended results of the applicable preventive measures, preventive interventions are performed on the target patients, and actual thrombotic event data are collected after the intervention. The feature weights in the risk assessment model are then adaptively adjusted using the actual thrombotic event data.

[0008] Based on a pre-defined risk assessment model, a multi-dimensional correlation analysis is performed on the physiological monitoring data, the medication record data, and the activity status data to extract feature vectors characterizing the risk of thrombosis. The resulting risk assessment results include: The physiological monitoring data, medication record data and activity status data are spatiotemporally aligned. The spatiotemporally aligned data are then used to construct a multidimensional temporal feature matrix. The row vectors of the multidimensional temporal feature matrix represent data sampling at different time points, and the column vectors represent the feature dimensions of different data types. Based on the preset risk assessment model, dynamic time warping calculation is performed on the multidimensional time-series feature matrix to extract feature vectors of blood viscosity change trend, vascular resistance change trend and platelet activity change trend. Combined with the anticoagulant drug usage in the medication record data, the comprehensive weight coefficient of the feature vector is calculated. The feature vector and the comprehensive weight coefficient are input into the risk assessment model to generate the risk assessment result.

[0009] Based on a preset risk assessment model, dynamic time warping is performed on the multidimensional time-series feature matrix to extract feature vectors of blood viscosity change trends, vascular resistance change trends, and platelet activity change trends, including: Based on a preset risk assessment model, dynamic time warping calculation is performed on the multidimensional time series feature matrix. The dynamic time warping calculation eliminates the fluctuation interference of time series data by nonlinearly scaling the time axis to obtain warped time series data. Based on the normalized time-series data, the rate of change of blood viscosity, vascular resistance and platelet activity at adjacent time points is calculated respectively, and feature vectors representing the changing trends of the three physiological indicators are constructed according to the rate of change.

[0010] Based on the risk level identifier in the risk assessment results, the target patient is matched and associated with the preventive measures in the preventive measures knowledge base to obtain a set of candidate preventive measures, including: The risk assessment results include risk level identifiers and access to a preventive measures knowledge base, which stores measure plans corresponding to different risk levels. A matching vector is constructed based on the risk level identifier. The matching vector includes feature parameters such as risk level, patient age, and underlying disease status. The treatment plan is then screened based on the feature parameters. The selected measures are then subjected to similarity calculation, which is based on a weighted average of historical treatment effect scores, to generate a set of candidate preventive measures.

[0011] Based on the set of candidate preventive measures and the contraindications of the target patient, the following recommended preventive measures are generated: Obtain the contraindications of the target patient, generate filtering rules based on the contraindications, conduct a safety assessment on each preventive measure in the candidate preventive measure set according to the filtering rules, eliminate preventive measures that conflict with the contraindications, and obtain preliminary screening results; The applicability of the preventive measures in the preliminary screening results is ranked. The applicability ranking is based on a safety coefficient calculated from historical medication feedback data, and the safety coefficient is used as the ranking weight to generate a recommendation result for applicable preventive measures.

[0012] Based on the recommended applicable preventive measures, preventive interventions are performed on the target patients, and actual thrombotic event data are collected after the intervention. The feature weights in the risk assessment model are then adaptively adjusted using the actual thrombotic event data, including: Based on the recommendations of applicable preventive measures, preventive interventions are performed on the target patients, and data on the actual occurrence of thrombotic events in the target patients after the preventive interventions are collected. The actual thrombotic event data is compared with the prediction results of the risk assessment model to calculate the prediction deviation value. Based on the prediction deviation value, the feature weights in the risk assessment model are adaptively adjusted so that the prediction results of the risk assessment model are closer to the actual thrombotic event data.

[0013] A second aspect of the present invention provides an intelligent recommendation system for dynamic thrombosis prevention measures for hospitalized patients, comprising: The first unit is used to acquire physiological monitoring data, medication record data, and activity status data of the target patient; The second unit is used to perform multi-dimensional correlation analysis on the physiological monitoring data, the medication record data and the activity status data based on a preset risk assessment model, extract feature vectors that characterize the risk of thrombosis, and obtain risk assessment results. The third unit is used to match and associate the target patient with the preventive measures in the preventive measures knowledge base based on the risk level identifier in the risk assessment results to obtain a set of candidate preventive measures; and to generate a recommendation result of applicable preventive measures based on the set of candidate preventive measures and the contraindications of the target patient. The fourth unit is used to perform preventive intervention operations on the target patient based on the recommended results of the applicable preventive measures and to collect actual thrombotic event data after the intervention, and to adaptively adjust the feature weights in the risk assessment model using the actual thrombotic event data.

[0014] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0015] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0016] The beneficial effects of this application are as follows: By acquiring physiological monitoring data, medication record data, and activity status data of target patients, comprehensive collection of multi-dimensional data was achieved, overcoming the limitations of traditional thrombosis risk assessment methods with a single data source, and providing a more comprehensive data foundation for risk assessment.

[0017] Based on a pre-defined risk assessment model, correlation analysis is performed on multi-dimensional data and feature vectors are extracted, enabling a precise quantitative assessment of the risk of thrombosis. Compared with traditional fixed scoring standards, this approach is more targeted and accurate.

[0018] By matching and associating risk level identifiers with a knowledge base of preventive measures, and combining this with individual patient contraindications, personalized recommendations for applicable preventive measures can be generated for different patients, avoiding the shortcomings of traditional approaches.

[0019] By employing a closed-loop feedback mechanism, the feature weights of the risk assessment model are adaptively adjusted based on actual thrombotic event data, enabling the system to have self-learning capabilities and continuously optimize assessment accuracy as clinical data accumulates, effectively improving the precision and timeliness of thrombosis prevention. Attached Figure Description

[0020] Figure 1 This is a flowchart illustrating the intelligent recommendation method for dynamic thrombosis prevention measures for hospitalized patients according to an embodiment of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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.

[0022] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0023] Figure 1 This is a flowchart illustrating the intelligent recommendation method for dynamic thrombosis prevention measures for hospitalized patients according to an embodiment of the present invention. Figure 1 As shown, the method includes: Acquire physiological monitoring data, medication record data, and activity status data of the target patients; Based on a preset risk assessment model, a multi-dimensional correlation analysis is performed on the physiological monitoring data, the medication record data, and the activity status data to extract feature vectors that characterize the risk of thrombosis and obtain risk assessment results. Based on the risk level identifier in the risk assessment results, the target patient is matched and associated with the prevention measures in the prevention measures knowledge base to obtain a set of candidate prevention measures; and based on the set of candidate prevention measures and the contraindications of the target patient, a recommendation result of applicable prevention measures is generated. Based on the recommended results of the applicable preventive measures, preventive interventions are performed on the target patients, and actual thrombotic event data are collected after the intervention. The feature weights in the risk assessment model are then adaptively adjusted using the actual thrombotic event data.

[0024] In one optional implementation, a multi-dimensional correlation analysis is performed on the physiological monitoring data, medication record data, and activity status data based on a preset risk assessment model to extract feature vectors characterizing the risk of thrombosis, and the resulting risk assessment results include: The physiological monitoring data, medication record data and activity status data are spatiotemporally aligned. The spatiotemporally aligned data are then used to construct a multidimensional temporal feature matrix. The row vectors of the multidimensional temporal feature matrix represent data sampling at different time points, and the column vectors represent the feature dimensions of different data types. Based on the preset risk assessment model, dynamic time warping calculation is performed on the multidimensional time-series feature matrix to extract feature vectors of blood viscosity change trend, vascular resistance change trend and platelet activity change trend. Combined with the anticoagulant drug usage in the medication record data, the comprehensive weight coefficient of the feature vector is calculated. The feature vector and the comprehensive weight coefficient are input into the risk assessment model to generate the risk assessment result.

[0025] In one specific embodiment, the thrombosis risk assessment process first requires acquiring the patient's physiological monitoring data, medication record data, and activity status data. Physiological monitoring data includes, but is not limited to, blood pressure, heart rate, blood oxygen saturation, and blood viscosity indicators; medication record data includes information such as the type, dosage, and duration of use of anticoagulants; and activity status data includes information such as the patient's exercise volume, sedentary time, and bed rest time.

[0026] Spatiotemporal alignment of acquired data is fundamental to accurate risk assessment. This process unifies data from different time points and sampling frequencies to a common time reference system, resolving inconsistencies in data collection times. For example, blood pressure data is measured every 4 hours, while activity data is recorded continuously; these data need to be aligned to a unified time point. In practice, interpolation algorithms are used to interpolate low-frequency sampling data, aligning it with high-frequency data on the time axis. Simultaneously, missing data points are addressed using methods such as forward padding or average padding to fill in gaps.

[0027] After spatiotemporal alignment, a multidimensional temporal feature matrix is ​​constructed, where each row of the matrix represents a data sample at a given time point, and each column represents a data feature. For example, the matrix contains multiple indicator values ​​at time point t1, such as blood pressure, heart rate, blood oxygen, drug concentration, and activity intensity. In this way, an n×m matrix can represent m types of feature data at n time points, forming a complete temporal representation of the patient's physiological state.

[0028] Based on the constructed multidimensional time-series feature matrix, dynamic time warping calculation is performed. This algorithm can handle time-series data of varying lengths and rates, identifying the inherent patterns of data change. Specifically, the algorithm first calculates the changing trends of each indicator in the feature matrix over time, identifying the changing patterns of blood viscosity, vascular resistance, and platelet activity. For example, by analyzing continuous changes in blood viscosity, it can identify whether the blood exhibits a sustained thickening trend; by analyzing the coordinated changes in blood pressure and heart rate, it can infer changes in vascular resistance; and by analyzing fluctuations in specific biochemical indicators, it can assess the changing trends of platelet activity.

[0029] Feature vector extraction is a core step in risk assessment. For trends in blood viscosity, features such as average value, coefficient of variation, and rate of increase / decrease are extracted; for trends in vascular resistance, features such as blood pressure fluctuation amplitude and vasomotor index are extracted; and for trends in platelet activity, features such as platelet aggregation index and activation level are extracted. These features together constitute the feature vector characterizing the risk of thrombosis.

[0030] By combining anticoagulant usage data from medication records, a comprehensive weighting coefficient for the feature vector is calculated. The type, dosage, and duration of use of anticoagulants directly affect the risk of thrombosis. For example, for patients using warfarin, the weight of blood viscosity characteristics is adjusted based on international normalized ratio (INR) monitoring results; for patients using low molecular weight heparin, the weight of platelet activity characteristics is adjusted based on their anti-Xa activity levels. The effective drug concentration in vivo is calculated using a pharmacodynamic model, and the weighting coefficients of each feature vector are dynamically adjusted accordingly.

[0031] The extracted feature vectors and calculated comprehensive weight coefficients are input into a pre-defined risk assessment model. This model can be a classifier trained based on machine learning algorithms, such as support vector machines, random forests, or deep neural networks. The model receives the feature vectors and weight coefficients as input and outputs the risk level of thrombosis. The risk level can be divided into three levels: low risk, medium risk, and high risk, or a more refined percentage risk score can be used.

[0032] After the risk assessment results are generated, the system can also provide personalized risk intervention suggestions based on the patient's specific situation. For high-risk patients, it is recommended to increase the dosage of anticoagulants or adjust the type of medication; for intermediate-risk patients, it is recommended to increase physical activity or adjust lifestyle habits; for low-risk patients, it is recommended to maintain the existing treatment plan and have regular follow-up examinations.

[0033] In practical applications, taking hospitalized patients as an example, a multidimensional time-series characteristic matrix of the patient is constructed by continuously monitoring their heart rate, activity level, and other data through smart bracelets, combined with daily routine measurements of blood pressure and blood oxygen saturation, as well as medication usage records in the hospital information system. After model analysis, if it is found that the patient's blood viscosity is continuously increasing and their activity level is significantly decreasing, while the anticoagulant dosage is insufficient, the system will determine that the patient has a high risk of thrombosis and promptly issue a warning to medical staff, suggesting adjustments to the treatment plan or strengthening preventative measures.

[0034] The advantage of this risk assessment method is that it can comprehensively consider multiple influencing factors, dynamically assess the risk of thrombosis in patients in real time, provide a scientific basis for clinical decision-making, and effectively reduce the incidence of thrombosis-related complications.

[0035] In one optional implementation, dynamic time warping is performed on the multidimensional time-series feature matrix based on a preset risk assessment model to extract feature vectors of blood viscosity change trends, vascular resistance change trends, and platelet activity change trends, including: Based on a preset risk assessment model, dynamic time warping calculation is performed on the multidimensional time series feature matrix. The dynamic time warping calculation eliminates the fluctuation interference of time series data by nonlinearly scaling the time axis to obtain warped time series data. Based on the normalized time-series data, the rate of change of blood viscosity, vascular resistance and platelet activity at adjacent time points is calculated respectively, and feature vectors representing the changing trends of the three physiological indicators are constructed according to the rate of change.

[0036] The multidimensional time-series feature matrix is ​​processed based on a pre-defined risk assessment model, and fluctuations in the time-series data are eliminated through dynamic time warping. Dynamic time warping is an algorithm used to measure the similarity between two time series. It can handle non-linear scaling of the time axis and effectively eliminate fluctuations in time-series data caused by inconsistent acquisition frequencies or external interference factors.

[0037] In practical applications, the multidimensional time-series feature matrix of collected blood viscosity, vascular resistance, and platelet activity is first represented as T = {x1, x2, ..., x}. n}, where x i The feature vector representing the data collected at the i-th time point contains values ​​for three indicators: blood viscosity, vascular resistance, and platelet activity. To process this time-series data, the reference template sequence is defined as R = {r1, r2, ..., r...}. n}, where R can be a standard pattern in historical data or a standard variation pattern determined by medical experts.

[0038] During the dynamic time warping calculation, an n×m distance matrix D is constructed, where D(i,j) represents x in T. i r in R j The Euclidean distance between T and R is used. Based on this distance matrix, a dynamic programming algorithm is employed to find an optimal path W from D(1,1) to D(n,m) that minimizes the total distance along this path. This path defines the optimal alignment between T and R, thereby achieving non-linear scaling of the time axis and eliminating fluctuations in the time series data.

[0039] For blood viscosity data, assuming data was collected from patients over 10 consecutive days, but due to various reasons, the data collection time points on some days are irregular or contain noise. Through dynamic time warping calculations, these irregular time point data are aligned with a standard template to obtain warped blood viscosity time-series data.

[0040] The normalized time series data is represented as T'={x'1, x'2, ..., x' n}, where p is not equal to the original sequence length n. This processing makes the time series data smoother, facilitating subsequent analysis.

[0041] Based on the normalized time-series data, the rates of change of blood viscosity, vascular resistance, and platelet activity at adjacent time points were calculated. For each time point i (i>1) in the normalized time-series data T', the rates of change of the three physiological indicators were calculated: For blood viscosity, the rate of change ΔV i = (V' i - V' i-1 ) / V' i-1 , where V' i This represents the blood viscosity value at the i-th time point in the normalized time series data.

[0042] For vascular resistance, the rate of change ΔR i = (R' i - R' i-1 ) / R' i-1 , where R' i This represents the vascular resistance value at the i-th time point in the normalized time series data.

[0043] For platelet activity, the rate of change ΔP i = (P' i - P' i-1 ) / P' i-1 , where P' i This represents the platelet activity value at the i-th time point in the normalized time series data.

[0044] In practical applications, such as monitoring a hypertensive patient, blood viscosity, vascular resistance, and platelet activity are measured daily for seven consecutive days. The normalized data shows that blood viscosity gradually increased from 4.5 mPa·s on day 1 to 5.8 mPa·s on day 7, vascular resistance increased from 120 PRU to 150 PRU, and platelet activity increased from 30% to 45%. By calculating the rate of change between adjacent time points, the following sequences were obtained: blood viscosity change rate [0.06, 0.05, 0.07, 0.04, 0.08, 0.05], vascular resistance change rate [0.04, 0.05, 0.03, 0.06, 0.04, 0.03], and platelet activity change rate [0.05, 0.07, 0.08, 0.04, 0.06, 0.05].

[0045] Based on the calculated rate of change sequence, a feature vector F = [F_V, F_R, F_P] is constructed, where F_V, F_R, and F_P represent the characteristic trends of blood viscosity, vascular resistance, and platelet activity, respectively. Each feature representation can include multiple statistics, such as the average rate of change, the maximum rate of change, and the standard deviation of the rate of change, to comprehensively capture the characteristics of the changing trends.

[0046] In a specific embodiment, F_V can be defined as the combination of the mean, maximum, and standard deviation of the rate of change of blood viscosity, i.e., F_V = [mean(ΔV), max(ΔV), std(ΔV)]; similarly, F_R and F_P are defined. For the example of the hypertensive patient mentioned above, F_V = [0.058, 0.08, 0.014], F_R = [0.042, 0.06, 0.011], and F_P = [0.058, 0.08, 0.015]. These feature vectors capture the trends of changes in the patient's blood viscosity, vascular resistance, and platelet activity over time, providing important information for subsequent risk assessment.

[0047] The feature vector F can be further input into a pre-defined risk assessment model, such as a support vector machine, random forest, or deep neural network, to assess the patient's risk of developing thrombotic diseases. For example, for the hypertensive patient mentioned above, based on their feature vector F, the risk assessment model gives a moderate risk assessment result, suggesting that the doctor closely monitor the patient's condition and consider adjusting the treatment plan.

[0048] This method, based on dynamic time warping calculation and rate of change feature extraction, can effectively handle noise and irregular sampling problems in time-series data, accurately capture the changing trends of blood viscosity, vascular resistance and platelet activity, and provide reliable technical support for thrombotic disease risk assessment.

[0049] In one optional implementation, the target patient is matched and associated with preventive measures in a knowledge base based on the risk level identifier in the risk assessment results, resulting in a set of candidate preventive measures, including: The risk assessment results include risk level identifiers and access to a preventive measures knowledge base, which stores measure plans corresponding to different risk levels. A matching vector is constructed based on the risk level identifier. The matching vector includes feature parameters such as risk level, patient age, and underlying disease status. The treatment plan is then screened based on the feature parameters. The selected measures are then subjected to similarity calculation, which is based on a weighted average of historical treatment effect scores, to generate a set of candidate preventive measures.

[0050] The process of matching target patients with preventative measures in a knowledge base based on risk level labels from risk assessment results first requires obtaining risk assessment results containing risk level labels. Risk level labels typically include three levels: high risk, medium risk, and low risk, each representing the likelihood of a patient experiencing an adverse event. For example, for fall risk assessment, a score greater than 45 using the Morse Fall Risk Assessment Scale is labeled as high risk; a score between 25 and 45 is labeled as medium risk; and a score below 25 is labeled as low risk.

[0051] Obtaining a knowledge base of preventative measures is fundamental to achieving precise prevention. This knowledge base stores measures corresponding to different risk levels, constructed based on evidence-based medicine and clinical experience. Each record in the knowledge base includes the risk level, applicable age range, applicable underlying disease type, and specific preventative measures. For example, for elderly patients at high risk of falls, the knowledge base stores preventative measures such as bed rail protection, non-slip mats, and regular checkups; for patients at medium risk, it stores measures such as assistive walking tools and environmental safety assessments; and for patients at low risk, it includes more basic preventative measures such as basic safety education.

[0052] Constructing a matching vector based on risk level identifiers is a crucial step in accurately selecting intervention plans. The matching vector includes feature parameters such as risk level, patient age, and underlying disease status. Risk level is directly adopted from the risk assessment results, such as high, medium, and low. Patient age is divided into four ranges: children (0-14 years), young adults (15-44 years), middle-aged (45-59 years), and elderly (60 years and above). Underlying disease status is labeled according to the patient's existing conditions, such as cardiovascular disease, neurological disease, and metabolic disease. After the matching vector is constructed, it is matched with intervention plans in the preventive measures knowledge base to select intervention plans that match the risk level, are age-appropriate, and take into account the patient's underlying disease status.

[0053] When calculating the similarity of the selected intervention plans, a weighted cosine similarity algorithm was used to extract key features of each plan, including applicable risk level, applicable age range, applicable underlying disease type, and intervention intensity. These features were then vectorized and compared with the actual situation of the target patients to calculate the similarity degree. During the similarity calculation, weighting was applied based on historical treatment effect scores. These scores are derived from previous clinical efficacy data of the intervention plan; a higher score indicates better effectiveness of the plan in similar patient populations. Through weighting, intervention plans with better historical efficacy received higher similarity scores, thus being prioritized for inclusion in the candidate prevention measure set.

[0054] The generation of a candidate prevention measure set is the final output of the matching and association process. Based on the similarity calculation results, the measures are sorted from high to low similarity scores, and the top N measures are selected to form the candidate prevention measure set. Each prevention measure in the set includes detailed information such as specific implementation methods, required resources, and expected effects, providing direct guidance for subsequent intervention implementation. For example, for a 75-year-old patient with mild cognitive impairment and assessed as having a high risk of falls, the generated candidate prevention measure set includes specific measures such as: laying anti-slip mats at the bedside, regular rounds (every 2 hours), setting up a low bed, two-person assisted toileting, and cognitive reminder signs.

[0055] The entire matching and association process achieves a precise transformation from risk assessment to preventive measures. By considering individual patient characteristics and historical treatment outcomes, it ensures the targeted nature and effectiveness of the selected preventive measures. This precise matching method based on risk level and patient characteristics significantly improves the effectiveness of preventive measures, reduces the incidence of adverse events, and optimizes the allocation and use of medical resources.

[0056] In practical application, an 80-year-old patient was admitted for treatment. Risk assessment indicated a "high-risk" fall risk level, along with hypertension and mild cognitive impairment. The constructed matching vector was ["high-risk", "elderly", "cardiovascular disease, cognitive impairment"]. The system selected appropriate measures from a preventative measures knowledge base, including over ten measures such as bedside protection, walking assistance interventions, and environmental adaptation modifications. After similarity calculation and weighting based on historical treatment effect scores, the final candidate preventative measures set included five specific measures: bedside anti-slip mats, fall detection alarms, regular rounds, and dedicated caregivers. This provided a scientific basis for medical staff to develop individualized prevention plans.

[0057] In one optional implementation, generating a recommended set of applicable preventative measures based on the candidate preventative measures set and the contraindications of the target patient includes: Obtain the contraindications of the target patient, generate filtering rules based on the contraindications, conduct a safety assessment on each preventive measure in the candidate preventive measure set according to the filtering rules, eliminate preventive measures that conflict with the contraindications, and obtain preliminary screening results; The applicability of the preventive measures in the preliminary screening results is ranked. The applicability ranking is based on a safety coefficient calculated from historical medication feedback data, and the safety coefficient is used as the ranking weight to generate a recommendation result for applicable preventive measures.

[0058] Obtain contraindications for the target patient by using electronic medical records, patient health records, or medical consultation forms. This information includes drug allergy history, past adverse reactions, special physiological conditions (such as pregnancy or lactation), and chronic diseases (such as liver or kidney dysfunction). For example, a patient may be allergic to penicillin and also have moderate renal insufficiency (creatinine clearance of 45 ml / min).

[0059] Filtering rules are generated based on contraindication constraints. The acquired contraindication information is transformed into a structured set of filtering rules. Each rule includes attributes such as contraindication item, scope of application, and degree of contraindication. The contraindication item describes the specific type of contraindication; the scope of application indicates the category or component of preventive measures affected by the contraindication; the degree of contraindication is identified as "absolute contraindication" or "relative contraindication," corresponding to situations that must be avoided and those that should be used with caution, respectively. For example, for patients with penicillin allergy, the rule can be generated as: "Contraindication item = penicillin allergy, scope of application = penicillin antibiotics and their derivatives, degree of contraindication = absolute contraindication"; for patients with renal insufficiency, the rule can be generated as: "Contraindication item = moderate renal insufficiency, scope of application = drugs excreted by the kidneys, degree of contraindication = relative contraindication, adjustment threshold = dose halved or monitoring."

[0060] The candidate prophylaxis set is safety-assessed based on the generated filtering rules, and a prophylaxis knowledge base is established, containing information such as the active ingredient, pharmacological mechanism of action, excretion route, common adverse reactions, and contraindications for each prophylaxis. A rule matching engine is used to match each candidate prophylaxis with the filtering rules, calculating the degree of conflict. Prophylaxis with absolute contraindications are directly eliminated; those with relative contraindications are marked with risk levels and necessary adjustment suggestions are recorded. For example, if a patient is allergic to penicillin, all prophylactic antibiotics containing penicillin are excluded; for patients with moderate renal insufficiency, dose adjustment factors are calculated based on creatinine clearance for drugs primarily excreted through the kidneys (such as certain antiviral prophylaxis drugs). Preliminary screening results are obtained after the safety assessment, including prophylaxis items that pass the safety screening and corresponding safety adjustment suggestions.

[0061] The applicability of preventive measures in the initial screening results was ranked, and a safety factor calculation model based on historical medication feedback was established. Historical medication records and adverse reaction reports of patients with similar contraindications were extracted from the medical database, and the safety factor of each preventive measure under specific contraindication conditions was calculated. The safety factor calculation considered the following factors: historical adverse reaction incidence rate, severity weighting, adjusted effectiveness data, and expert consensus score. For example, the safety factor of a certain antiviral prophylactic drug in patients with renal insufficiency can be expressed as follows: total historical use 200 times, 5 adverse reactions, including 3 mild adverse reactions (weight 0.3) and 2 moderate adverse reactions (weight 0.7), and an expert score of 4.2 (out of 5), then the overall safety factor is 0.86.

[0062] The safety factor is used as a ranking weighting factor, and combined with the effectiveness index of preventive measures to generate the final ranking score. A multi-factor weighted algorithm can be used for ranking, with the weight allocation as follows: safety factor 60%, effectiveness index 30%, and ease of use 10%. Based on this, a recommended list of preventive measures is generated, arranged in descending order of the comprehensive score. For preventive measures that are relatively contraindicated but can be safely used after dosage adjustment, specific adjustment plans are indicated in the recommendation results, such as "dosage reduction of 50% required" or "monitoring of renal function required."

[0063] The recommendations for applicable preventative measures are formulated, including recommendation levels, safe usage instructions, adjustment suggestions, and key monitoring points. Recommendation levels are divided into three categories: "preferred," "optional," and "use with caution," each corresponding to different ranges of safety levels. For example, for a diabetic patient with moderate renal insufficiency, dose-adjusted neuraminidase inhibitors would be the preferred preventative measure during the flu season, rather than other antiviral drugs that place a heavier metabolic burden on the kidneys.

[0064] By employing the above technical solutions, and based on the contraindications of the target patients, filtering rules and safety factor ranking methods can be used to generate safe, effective, and personalized preventive measures recommendations, thereby improving the accuracy of preventive measures and patient safety.

[0065] In one optional implementation, performing preventive intervention on the target patient based on the recommended results of the applicable preventive measures and collecting actual thrombotic event data after the intervention, and adaptively adjusting the feature weights in the risk assessment model using the actual thrombotic event data, includes: Based on the recommendations of applicable preventive measures, preventive interventions are performed on the target patients, and data on the actual occurrence of thrombotic events in the target patients after the preventive interventions are collected. The actual thrombotic event data is compared with the prediction results of the risk assessment model to calculate the prediction deviation value. Based on the prediction deviation value, the feature weights in the risk assessment model are adaptively adjusted so that the prediction results of the risk assessment model are closer to the actual thrombotic event data.

[0066] Based on the recommended applicable preventive measures, preventive interventions are performed on the target patients, and data on actual thrombotic events after the intervention are collected. This data is then used to adaptively adjust the feature weights in the risk assessment model. In this way, the risk assessment model can be continuously optimized, making its predictions more closely reflect reality and improving its predictive accuracy.

[0067] Preventive interventions are implemented for target patients based on the recommendations of applicable preventive measures. These interventions include pharmacological interventions (such as the use of anticoagulants like low molecular weight heparin and warfarin), physical interventions (such as the use of intermittent pneumatic compression devices and compression stockings), and lifestyle interventions (such as early mobilization and adequate hydration). The specific intervention protocols vary depending on the patient's risk level. For example, low-risk patients only require early mobilization and adequate hydration; medium-risk patients need compression stockings or intermittent pneumatic compression devices; and high-risk patients require both physical and pharmacological preventive measures.

[0068] After implementing preventative interventions, data on actual thrombotic events in the target patients should be collected. This data includes whether a thrombotic event occurred, and if so, the type (e.g., deep vein thrombosis, pulmonary embolism), severity, and time of occurrence. Additionally, basic patient information, clinical manifestations, laboratory test results, and imaging findings should be recorded for subsequent analysis. Data collection can be conducted through various channels, including hospital information systems, electronic medical record systems, and follow-up systems, to ensure data completeness and accuracy.

[0069] The prediction bias is calculated by comparing actual thrombotic event data with the prediction results of the risk assessment model. The prediction bias can be calculated in various ways, such as using mean squared error, mean absolute error, or logarithmic loss. In this embodiment, mean squared error is used as the method for calculating the prediction bias. Specifically, for each patient, the risk assessment model provides a predicted thrombotic risk value (e.g., risk percentage), while the actual data records whether the patient actually experienced a thrombotic event (which can be represented as 0 or 1). The mean squared error is obtained by summing the squares of the differences between the predicted and actual values ​​for all patients and then dividing by the number of patients.

[0070] The feature weights in the risk assessment model are adaptively adjusted based on the prediction bias. The goal of this adjustment is to make the model's predictions more closely reflect actual thrombotic events, i.e., to reduce the prediction bias. There are various methods for adjusting feature weights, such as gradient descent and Newton's method. In this embodiment, gradient descent is used to adjust the feature weights.

[0071] Specifically, for each feature weight in the risk assessment model, the partial derivative of the prediction deviation with respect to that weight, i.e., the gradient, is calculated. Then, the feature weights are updated in the opposite direction of the gradient with a certain step size. The choice of step size requires a trade-off between convergence speed and stability; a fixed step size or an adaptive step size strategy can be used. The formula for updating the feature weights is: New weight = Old weight - Step size × Gradient. Through multiple iterations, the feature weights will gradually adjust to the position that minimizes the prediction deviation.

[0072] When adjusting feature weights, it's crucial to prevent overfitting. Overfitting occurs when a model fits the training data too precisely, leading to a decrease in its predictive ability for new data. To prevent overfitting, regularization methods such as L1 and L2 regularization can be employed. In this embodiment, L2 regularization is used, adding a penalty term based on the sum of squared feature weights when calculating the prediction bias, thus keeping the feature weights within a relatively small range.

[0073] After adjustments, the updated risk assessment model should be validated. Validation can use an independent validation dataset or methods such as cross-validation. Validation metrics include accuracy, precision, recall, F1 score, and AUC. If the validation results show improved model performance, the adjustment is accepted; otherwise, hyperparameters such as the learning rate and regularization coefficient need to be adjusted, or different adjustment methods should be tried.

[0074] Through the above steps, the feature weights in the risk assessment model are adaptively adjusted, making the model's predictions closer to actual thrombotic event data and improving the model's prediction accuracy. This adaptive adjustment is an ongoing process; as more data accumulates, the model can be continuously optimized to better serve thrombotic risk assessment and prevention intervention decisions.

[0075] The method further includes: When acquiring physiological monitoring data, medication record data, and activity status data of target patients, relevant information is collected in real time from the electronic medical record system, nursing record system, and medical order execution system through the data interface of the hospital information system. Physiological monitoring data includes blood pressure, heart rate, body temperature, blood oxygen saturation, and coagulation function indicators. Each indicator records the measurement timestamp, measurement value, and unit identifier. Medication record data includes the order time, start time, end time, dosage, and route of administration for anticoagulant medications, and the order time, start time, end time, and type of prophylactic device for mechanical devices. Activity status data includes the patient's bed rest time, number of times they got out of bed, and duration of activity. The data acquisition module is set to a polling cycle of 5 minutes. When the data interface response timeout exceeds 3 seconds, a retry mechanism is initiated, with a retry interval of 1 second and a maximum of 3 retries. In the data cleaning stage, a forward-filling strategy is used for missing values. Outliers are marked by comparing them with the standard deviation of the patient's historical data; values ​​exceeding 3 times the standard deviation are marked as abnormal and trigger a manual review process.

[0076] When performing multi-dimensional correlation analysis on the physiological monitoring data, medication record data, and activity status data based on the preset risk assessment model, the risk assessment model adopts a hierarchical assessment structure, including a venous thromboembolism risk assessment layer, a bleeding risk assessment layer, and a mechanical contraindication assessment layer. The venous thromboembolism risk assessment layer calculates a risk score based on the patient's age, surgical type identifier, tumor history identifier, previous thrombotic history identifier, and activity status data. One point is added for every 10 years of age, three points are added for a history of malignant tumors, five points are added for a previous thrombotic history, and four points are added for bed rest exceeding 72 hours. The scores from each category are accumulated and compared with a preset threshold to obtain the risk level identifier.

[0077] The bleeding risk assessment layer calculates a bleeding risk score based on coagulation function indicators, platelet count, and bleeding history markers. A prothrombin time international normalized ratio (ITR) greater than 1.5 adds 3 points, a platelet count below 50 × 10^9 / L adds 4 points, and a history of active bleeding adds 5 points. The mechanical contraindication assessment layer determines mechanical contraindication based on lower extremity skin integrity markers, deep vein thrombosis diagnosis markers, and severe arterial insufficiency markers. In the multi-dimensional correlation analysis, a time-series decay mechanism is introduced to assign differentiated weights to the physiological monitoring data at different times. The time difference between each physiological monitoring data record and the current assessment time is calculated. Data with a time difference within 0 to 24 hours has a weight of 1.0; data with a time difference within 24 to 48 hours has a weight that decreases from 1.0 to 0.6; data with a time difference within 48 to 72 hours has a weight that decreases from 0.6 to 0.3; and data with a time difference exceeding 72 hours has a weight of 0.1.

[0078] The weighted index value is obtained by multiplying the measured value of each physiological monitoring indicator by its corresponding time weight and then averaging the results. When extracting the feature vector characterizing the risk of thrombosis, the risk scores, risk level indicators, and contraindication status indicators output from each assessment layer are combined in a fixed order to form the feature vector. The feature vector has 8 dimensions, namely: venous thromboembolism risk score, venous thromboembolism risk level indicator, bleeding risk score, bleeding risk level indicator, mechanical contraindication status indicator, latest assessment timestamp, assessment validity indicator, and risk change trend indicator. The risk change trend indicator is determined by comparing the risk score difference between the current assessment result and the previous assessment result; a difference greater than 2 points indicates an upward trend, and a difference less than -2 points indicates a downward trend.

[0079] When matching and associating the target patient with preventive measures in the preventive measures knowledge base based on the risk level identifier in the risk assessment results, the preventive measures knowledge base includes a measure plan table, an indication rule table, and a contraindication rule table. The indication rule table records the correspondence between the measure plan identifier and the risk level identifier. When the risk level identifier for venous thromboembolism is low, it is associated with the basic preventive measures plan; when the risk level identifier is intermediate or high, it is associated with the basic preventive measures plan, the pharmacological preventive measures plan, and the mechanical preventive measures plan.

[0080] The contraindication rule table records the correspondence between intervention protocol identifiers and contraindications. Drug prophylaxis protocols are associated with bleeding risk level identifiers indicating high-risk contraindications, while mechanical prophylaxis protocols are associated with mechanical contraindication status identifiers indicating the presence of a contraindication. The matching and association process queries the indication rule table based on the venous thromboembolism risk level identifier to obtain a preliminary set of candidate intervention protocols. For each intervention protocol identifier, the contraindication rule table is queried to determine if a contraindication is met. Intervention protocols that meet the contraindication are removed from the candidate set.

[0081] When generating recommended applicable preventive measures based on the candidate preventive measure set and the contraindications of the target patient, a dynamic prioritization rule is established. This rule is adjusted in real time based on the degree of conflict between the risk change trend in the risk assessment results and the contraindications. When the risk change trend is upward, the basic priority of the pharmacological preventive measure is set to 90, the basic priority of the mechanical preventive measure is set to 80, and the basic preventive measure is set to 70. When the risk change trend is stable or downward, the basic priority of the pharmacological preventive measure is set to 80, the basic priority of the mechanical preventive measure is set to 70, and the basic preventive measure is set to 60.

[0082] The degree of conflict of contraindications was determined by statistically analyzing the number and severity levels of contraindications associated with each measure. A conflict level of 0 was calculated when there were 0 contraindications, 10 when there were 1 contraindication with a mild severity level, and 30 when there were 1 contraindication with a moderate or severe severity level. The final priority value of each measure was calculated by subtracting the conflict level value from the base priority value and then adding the indication matching value. All measures in the candidate preventive measure set were sorted from highest to lowest final priority value, and measures with a final priority value greater than 60 were selected to form the recommended applicable preventive measures.

[0083] When performing preventive interventions on the target patients based on the recommended applicable preventive measures and collecting data on actual thrombotic events after the intervention, the preventive intervention is triggered by sending a medical order execution request to the medical order execution system. When collecting data on actual thrombotic events after the intervention, thrombosis-related clinical events are monitored during the implementation of preventive measures and within 7 days after the implementation, including confirmed deep vein thrombosis events, confirmed pulmonary embolism events, and records of suspected thrombotic symptoms. When adaptively adjusting the feature weights in the risk assessment model using the actual thrombotic event data, the deviation between the actual and expected proportions of patients experiencing thrombotic events under each risk level is statistically analyzed.

[0084] The expected occurrence rate is determined based on a baseline model trained on historical data. The expected occurrence rate is 0.5% for low-risk levels, 3% for medium-risk levels, and 10% for high-risk levels. When the actual occurrence rate at a certain risk level exceeds 1.5 times the expected occurrence rate, the feature weights corresponding to that risk level are adjusted upwards by 10% of their current values. Feature weight adjustments are performed weekly. After adjustment, model performance metrics are validated on the test dataset. If any metric decreases by more than 5%, the model is rolled back to the version before adjustment.

[0085] A second aspect of the present invention provides an intelligent recommendation system for dynamic thrombosis prevention measures for hospitalized patients, comprising: The first unit is used to acquire physiological monitoring data, medication record data, and activity status data of the target patient; The second unit is used to perform multi-dimensional correlation analysis on the physiological monitoring data, the medication record data and the activity status data based on a preset risk assessment model, extract feature vectors that characterize the risk of thrombosis, and obtain risk assessment results. The third unit is used to match and associate the target patient with the preventive measures in the preventive measures knowledge base based on the risk level identifier in the risk assessment results to obtain a set of candidate preventive measures; and to generate a recommendation result of applicable preventive measures based on the set of candidate preventive measures and the contraindications of the target patient. The fourth unit is used to perform preventive intervention operations on the target patient based on the recommended results of the applicable preventive measures and to collect actual thrombotic event data after the intervention, and to adaptively adjust the feature weights in the risk assessment model using the actual thrombotic event data.

[0086] A third aspect of the present invention provides an electronic device, comprising: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0087] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0088] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent recommendation method for inpatient dynamic thromboprophylaxis measures, characterized in that, include: Acquire physiological monitoring data, medication record data, and activity status data of the target patients; Based on a preset risk assessment model, a multi-dimensional correlation analysis is performed on the physiological monitoring data, the medication record data, and the activity status data to extract feature vectors that characterize the risk of thrombosis and obtain risk assessment results. Based on the risk level identifier in the risk assessment results, the target patient is matched and associated with the prevention measures in the prevention measures knowledge base to obtain a set of candidate prevention measures; and based on the set of candidate prevention measures and the contraindications of the target patient, a recommendation result of applicable prevention measures is generated. Based on the recommended results of the applicable preventive measures, preventive interventions are performed on the target patients, and actual thrombotic event data are collected after the intervention. The feature weights in the risk assessment model are then adaptively adjusted using the actual thrombotic event data.

2. The method according to claim 1, characterized in that, Based on a pre-defined risk assessment model, a multi-dimensional correlation analysis is performed on the physiological monitoring data, the medication record data, and the activity status data to extract feature vectors characterizing the risk of thrombosis. The resulting risk assessment results include: The physiological monitoring data, medication record data and activity status data are spatiotemporally aligned. The spatiotemporally aligned data are then used to construct a multidimensional temporal feature matrix. The row vectors of the multidimensional temporal feature matrix represent data sampling at different time points, and the column vectors represent the feature dimensions of different data types. Based on the preset risk assessment model, dynamic time warping calculation is performed on the multidimensional time-series feature matrix to extract feature vectors of blood viscosity change trend, vascular resistance change trend and platelet activity change trend. Combined with the anticoagulant drug usage in the medication record data, the comprehensive weight coefficient of the feature vector is calculated. The feature vector and the comprehensive weight coefficient are input into the risk assessment model to generate the risk assessment result.

3. The method according to claim 2, characterized in that, Based on a preset risk assessment model, dynamic time warping is performed on the multidimensional time-series feature matrix to extract feature vectors of blood viscosity change trends, vascular resistance change trends, and platelet activity change trends, including: Based on a preset risk assessment model, dynamic time warping calculation is performed on the multidimensional time series feature matrix. The dynamic time warping calculation eliminates the fluctuation interference of time series data by nonlinearly scaling the time axis to obtain warped time series data. Based on the normalized time-series data, the rate of change of blood viscosity, vascular resistance and platelet activity at adjacent time points is calculated respectively, and feature vectors representing the changing trends of the three physiological indicators are constructed according to the rate of change.

4. The method according to claim 1, characterized in that, Based on the risk level identifier in the risk assessment results, the target patient is matched and associated with the preventive measures in the preventive measures knowledge base to obtain a set of candidate preventive measures, including: The risk assessment results include risk level identifiers and access to a preventive measures knowledge base, which stores measure plans corresponding to different risk levels. A matching vector is constructed based on the risk level identifier. The matching vector includes feature parameters such as risk level, patient age, and underlying disease status. The treatment plan is then screened based on the feature parameters. The selected measures are then subjected to similarity calculation, which is based on a weighted average of historical treatment effect scores, to generate a set of candidate preventive measures.

5. The method according to claim 1, characterized in that, Based on the set of candidate preventive measures and the contraindications of the target patient, the following recommended preventive measures are generated: Obtain the contraindications of the target patient, generate filtering rules based on the contraindications, conduct a safety assessment on each preventive measure in the candidate preventive measure set according to the filtering rules, eliminate preventive measures that conflict with the contraindications, and obtain preliminary screening results; The applicability of the preventive measures in the preliminary screening results is ranked. The applicability ranking is based on a safety coefficient calculated from historical medication feedback data, and the safety coefficient is used as the ranking weight to generate a recommendation result for applicable preventive measures.

6. The method according to claim 1, characterized in that, Based on the recommended applicable preventive measures, preventive interventions are performed on the target patients, and actual thrombotic event data are collected after the intervention. The feature weights in the risk assessment model are then adaptively adjusted using the actual thrombotic event data, including: Based on the recommendations of applicable preventive measures, preventive interventions are performed on the target patients, and data on the actual occurrence of thrombotic events in the target patients after the preventive interventions are collected. The actual thrombotic event data is compared with the prediction results of the risk assessment model to calculate the prediction deviation value. Based on the prediction deviation value, the feature weights in the risk assessment model are adaptively adjusted so that the prediction results of the risk assessment model are closer to the actual thrombotic event data.

7. A dynamic thrombosis prevention measure intelligent recommendation system for hospitalized patients, used to implement the method of any one of claims 1-6, characterized in that, include: The first unit is used to acquire physiological monitoring data, medication record data, and activity status data of the target patient; The second unit is used to perform multi-dimensional correlation analysis on the physiological monitoring data, the medication record data and the activity status data based on a preset risk assessment model, extract feature vectors that characterize the risk of thrombosis, and obtain risk assessment results. The third unit is used to match and associate the target patient with the preventive measures in the preventive measures knowledge base based on the risk level identifier in the risk assessment results to obtain a set of candidate preventive measures; and to generate a recommendation result of applicable preventive measures based on the set of candidate preventive measures and the contraindications of the target patient. The fourth unit is used to perform preventive intervention operations on the target patient based on the recommended results of the applicable preventive measures and to collect actual thrombotic event data after the intervention, and to adaptively adjust the feature weights in the risk assessment model using the actual thrombotic event data.

8. An electronic device, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to invoke instructions stored in the memory to execute the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer program instructions stored thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 6.

Images