Adverse drug reaction prediction and evaluation method and system fusing multi-modal data

Abstract

This invention discloses a method and system for predicting and assessing adverse drug reactions by integrating multimodal data, belonging to the field of drug safety assessment technology. It constructs a multimodal data acquisition module to obtain clinical medication data, drug molecular characteristic data, patient physiological indicator data, and environmental impact data, forming a multidimensional feature vector of drug-patient interaction. It then constructs a risk demand vector for adverse drug reactions and calculates the risk factor fit between the drug-patient combination and the corresponding drug. Finally, it calculates the intensity of the comprehensive risk effect and the risk value of adverse reactions. Finally, it aggregates all medication risks for the patient at the current time point, combines environmental factors to construct an individualized dynamic risk threshold, and achieves risk level determination and real-time early warning. This enables continuous quantitative assessment and dynamic prediction of adverse drug reactions, improving the accuracy and individualization of risk identification.

Description

Technical Field

[0001] This invention relates to the field of drug safety assessment technology, specifically to a method and system for predicting and assessing adverse drug reactions by integrating multimodal data. Background Technology

[0002] Adverse drug reaction monitoring and prediction technologies are important research directions in clinical pharmacy, pharmacovigilance, and precision medicine. With the widespread application of high-throughput molecular detection technologies, wearable physiological monitoring devices, and hospital information systems, drug safety assessment has evolved from traditional post-hoc statistical analysis to data-driven prospective prediction. Existing research typically focuses on electronic medical record data mining, drug molecular structure similarity analysis, drug-target interaction network construction, and population epidemiological statistical models, which have improved the ability to identify potential drug risks to some extent. However, significant heterogeneity exists among different data sources in terms of collection mechanisms, time scales, and expression dimensions, making it difficult for a single data modality to comprehensively reflect the risk triggering conditions of drugs in real-world medication scenarios. Furthermore, the prevalence of combined drug use, significant dynamic fluctuations in individual physiological states, and increased differences in living environments contribute to the complex characteristics of adverse drug reactions, including multi-factor coupling, time-series correlation, and individual dependence, objectively driving the development of multimodal fusion modeling in drug safety assessment.

[0003] While existing technologies have attempted to incorporate multi-source data for risk analysis, several key shortcomings remain: First, most methods focus on statistical correlation modeling at the population level, lacking a detailed characterization of the comprehensive medication status of individual patients at specific time points, making it difficult to achieve individualized real-time risk assessment. Second, existing models often treat clinical indicators, molecular characteristics, and environmental factors in a fragmented manner, failing to form a unified multidimensional feature expression structure, making it difficult to compare and integrate the contributions of different risk factors on the same scale. Third, environmental behavioral factors such as smoking, alcohol consumption, and dietary structure are often simply introduced as auxiliary variables without being incorporated into the core computational framework of risk triggering mechanisms, weakening the ability to characterize the actual conditions inducing adverse reactions. Fourth, most predictive models output static risk probabilities, lacking a dynamic threshold adjustment mechanism linked to changes in the patient's real-time physiological state, easily leading to over-warning or under-warning. Therefore, existing technologies still have significant room for improvement in how to construct a unified drug-patient multidimensional feature expression method, how to characterize the adaptation relationship between risk factors and individual states, and how to establish a dynamic risk assessment mechanism that evolves over time. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for predicting and assessing adverse drug reactions by integrating multimodal data, so as to solve the problems mentioned in the background art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0006] A method for predicting and assessing adverse drug reactions by integrating multimodal data is proposed. This method includes the following steps: Step S1: Constructing a multimodal data acquisition module to collect clinical medication data, drug molecular characteristic data, patient physiological indicator data, and environmental impact data; constructing a drug-patient multidimensional feature vector; Step S2: Constructing a drug adverse reaction risk demand vector; calculating the risk factor fit between the drug-patient combination and the drug at the data sampling time point based on the drug-patient multidimensional feature vector of the drug-patient combination; Step S3: Calculating the comprehensive risk intensity of the drug-patient combination and the drug at the data sampling time point based on the drug-patient multidimensional feature vector and risk factor fit; calculating the environmental risk trigger ratio; and calculating the adverse reaction risk value of the drug-patient combination and the drug at the data sampling time point based on the comprehensive risk intensity and the environmental risk trigger ratio; Step S4: Calculating the comprehensive adverse reaction risk value of the patient and all drugs at the data sampling time point based on the adverse reaction risk value; calculating the individualized dynamic risk threshold for the patient at the data sampling time point; and analyzing and providing dynamic early warnings.

[0007] As a preferred embodiment of the drug adverse reaction prediction and assessment method integrating multimodal data described in this invention, a multimodal data acquisition module is constructed. The multimodal data acquisition module is used to collect drug-patient multidimensional data. The multimodal data acquisition module includes a clinical data acquisition unit, a molecular structure analysis unit, a physiological indicator monitoring unit, and an environmental factor acquisition unit. The drug-patient multidimensional data includes clinical drug use data, drug molecular characteristic data, patient physiological indicator data, and environmental impact data.

[0008] The clinical data acquisition unit collects clinical medication data of the target drug through the hospital information system and pharmacovigilance system, including clinical drug dosage, frequency of use, concomitant medications, and the patient's history of adverse reactions. The molecular structure analysis unit, based on molecular docking simulation tools, collects drug molecular characteristic data of the target drug, including molecular structure parameters, target affinity, and binding capacity of key enzymes in metabolic pathways. The physiological indicator monitoring unit, including wearable sensors and laboratory testing equipment, collects patients' physiological indicator data, including liver and kidney function indicators, blood routine data, and gene polymorphism site information. The environmental factor acquisition unit collects patients' environmental impact data through questionnaires and lifestyle monitoring equipment, including smoking and drinking history, dietary structure, and work-rest patterns.

[0009] As a preferred embodiment of the drug adverse reaction prediction and assessment method integrating multimodal data described in this invention, a data sampling time period is constructed, denoted as... ,in, Let A represent the a-th data sampling time point, and A represent the total number of data sampling time points; the data sampling time points are respectively... The clinical medication data, drug molecular characteristic data, patient physiological index data, and environmental impact data collected below are recorded as follows: , , and ;

[0010] Construct a drug-patient combination set, denoted as ,in, This represents the drug-patient combination of the c-th patient and the k-th drug, where C represents the total number of patients and K represents the total number of drugs; the data sampling time points are respectively... Drug-Patient Combination Collected Clinical drug use data, drug molecular characteristics data, patient physiological indicators data, and environmental impact data are denoted as , , and ;

[0011] Constructing data sampling time points Drug-Patient Combination The drug-patient multidimensional feature vector, denoted as .

[0012] As a preferred embodiment of the drug adverse reaction prediction and assessment method integrating multimodal data described in this invention, adverse reaction-related information of all target drugs is obtained from a drug safety database, and corresponding adverse reaction risk requirement parameters are preset for each drug to construct an adverse reaction risk requirement vector. The adverse reaction risk requirement vector of the k-th drug is denoted as... ,in, This represents the threshold of the key risk clinical indicator for the k-th drug. This represents the high-risk molecular characteristic parameter of the k-th drug. This represents the range of sensitive physiological indicators for the k-th drug. This represents the set of contraindications to the k-th drug;

[0013] Based on data sampling time points Drug-Patient Combination Drug-Patient Multidimensional Feature Vector Calculate the data sampling time point Drug-Patient Combination The formula for calculating the risk factor fit with the k-th drug is as follows:

[0014] ;

[0015] in, Indicates the data sampling time point Drug-Patient Combination Risk factor fit with the k-th drug This represents a preset constant. This represents the total number of factors in the set of contraindications for the k-th drug. This indicates the number of prohibited environmental factors included in the environmental impact data.

[0016] As a preferred embodiment of the drug adverse reaction prediction and assessment method that integrates multimodal data as described in this invention, based on data sampling time points... Drug-Patient Combination Drug-Patient Multidimensional Feature Vector and data sampling time points Drug-Patient Combination Risk factor fit with the kth drug Calculate the data sampling time point Drug-Patient Combination The combined risk intensity of the drug and the k-th drug is calculated using the following formula:

[0017] ;

[0018] in, Indicates the data sampling time point Drug-Patient Combination The combined risk intensity with the kth drug This represents a preset constant;

[0019] The environmental risk trigger ratio is calculated using the following formula:

[0020] ;

[0021] in, Indicates the data sampling time point Drug-Patient Combination The proportion of environmental risk triggering for the kth drug, i.e., the data sampling time point Drug-Patient Combination The proportion of environmental factors that fall into the set of contraindicated environmental factors for the k-th drug. This indicates the preset impact factor;

[0022] Based on data sampling time points Drug-Patient Combination Combined risk intensity with the kth drug With data sampling time point Drug-Patient Combination Environmental risk trigger ratio of the kth drug Calculate the data sampling time point Drug-Patient Combination The adverse reaction risk value of the k-th drug is calculated using the following formula:

[0023] ;

[0024] in, Indicates the data sampling time point Drug-Patient Combination The adverse reaction risk value compared to the k-th drug.

[0025] As a preferred embodiment of the drug adverse reaction prediction and assessment method that integrates multimodal data as described in this invention, based on data sampling time points... Drug-Patient Combination Adverse reaction risk value of the kth drug Calculate the data sampling time point The combined adverse reaction risk value for the c-th patient and all drugs is calculated using the following formula:

[0026] ;

[0027] in, Indicates the data sampling time point The combined adverse reaction risk value of the c-th patient and all drugs, where K represents the total number of drugs;

[0028] Calculate data sampling time points The individualized dynamic risk threshold for the c-th patient is calculated using the following formula:

[0029] ;

[0030] in, Indicates the data sampling time point The individualized dynamic risk threshold for the c-th patient. This represents the preset basic risk threshold constant.

[0031] As a preferred embodiment of the drug adverse reaction prediction and assessment method that integrates multimodal data as described in this invention, if the data sampling time point The combined adverse reaction risk value of the cth patient and all drugs Greater than or equal to the data sampling time point Individualized dynamic risk threshold for the cth patient Then determine the data sampling time point. If a high-risk adverse drug reaction is detected, an early warning will be issued;

[0032] The system acquires the comprehensive adverse reaction risk value at each data sampling time point in real time and calculates the individualized dynamic risk threshold in real time to perform dynamic real-time prediction and assessment of adverse drug reactions.

[0033] The adverse drug reaction prediction and assessment system integrates multimodal data. This system includes: a data acquisition and vector construction module, a fit calculation module, an intensity calculation and risk value calculation module, and a threshold calculation and analysis module.

[0034] The data acquisition and vector construction module includes: constructing a multimodal data acquisition module to collect clinical medication data, drug molecular characteristic data, patient physiological indicator data, and environmental impact data; and constructing a drug-patient multidimensional feature vector.

[0035] The fit calculation module: constructs a drug adverse reaction risk demand vector; and calculates the risk factor fit between the drug-patient combination and the drug at the data sampling time point based on the drug-patient multi-dimensional feature vector of the drug-patient combination.

[0036] The intensity calculation and risk value calculation module calculates the comprehensive risk intensity of the drug-patient combination and the drug at the data sampling time point based on the drug-patient multi-dimensional feature vector and risk factor fit; it also calculates the environmental risk trigger ratio and, based on the comprehensive risk intensity and the environmental risk trigger ratio, calculates the adverse reaction risk value of the drug-patient combination and the drug at the data sampling time point.

[0037] The threshold calculation and analysis module calculates the overall adverse reaction risk value of the patient and all drugs at the data sampling time point based on the adverse reaction risk value; calculates the individualized dynamic risk threshold of the patient at the data sampling time point; and analyzes and provides dynamic early warning.

[0038] Furthermore, the intensity calculation and risk value calculation module includes an intensity calculation unit and a risk value calculation unit;

[0039] The intensity calculation unit calculates the combined risk intensity of the drug-patient combination and the kth drug at the data sampling time point based on the drug-patient combination's multi-dimensional feature vector at the data sampling time point and the risk factor fit between the drug-patient combination and the kth drug at the data sampling time point.

[0040] The risk value calculation unit calculates the environmental risk trigger ratio and, based on the combined risk intensity of the drug-patient combination and the kth drug at the data sampling time point and the environmental risk trigger ratio of the drug-patient combination and the kth drug at the data sampling time point, calculates the adverse reaction risk value of the drug-patient combination and the kth drug at the data sampling time point.

[0041] Furthermore, the threshold calculation and analysis module includes a threshold calculation unit and an analysis unit;

[0042] The threshold calculation unit: based on the adverse reaction risk value of the drug-patient combination and the kth drug at the data sampling time point, calculates the comprehensive adverse reaction risk value of the cth patient and all drugs at the data sampling time point; and calculates the individualized dynamic risk threshold of the cth patient at the data sampling time point.

[0043] The analysis unit: if the combined adverse reaction risk value of the c-th patient and all drugs at the data sampling time point is greater than or equal to the individualized dynamic risk threshold of the c-th patient at the data sampling time point, then it determines that there is a high-risk adverse drug reaction at the data sampling time point and issues an early warning; it acquires the combined adverse reaction risk value at each data sampling time point in real time and calculates the individualized dynamic risk threshold in real time to perform dynamic real-time prediction and assessment of adverse drug reactions.

[0044] Compared with existing technologies, the beneficial effects achieved by this invention are as follows: The drug adverse reaction prediction and assessment method and system provided by this invention, which integrates multimodal data, constructs a multimodal data acquisition module and forms a drug-patient multidimensional feature vector. This enables the unified representation of clinical medication information, drug molecular structure characteristics, patient physiological state, and environmental behavioral factors within the same state space. This expands adverse reaction risk assessment from the analysis of a single medical record or medication record to a panoramic state modeling covering pharmacological mechanisms, individual susceptibility, and external triggers, solving the problem of existing technologies having a single risk characterization dimension and difficulty in reflecting individual differences. Furthermore, by constructing a drug adverse reaction risk demand vector and calculating the risk factor fit, it transforms clinical thresholds, molecular toxicity characteristics, physiological sensitivity ranges, and contraindicated environmental factors in drug safety knowledge into calculable structured parameters, allowing the patient's current state to be correlated with drug risk triggering conditions. Quantitative matching is used to establish a risk structure matching relationship under a unified dimension, improving the computability and mechanistic constraints of multi-source data fusion. Furthermore, by constructing a comprehensive risk effect intensity based on risk adaptation and introducing the proportion of environmental risk triggers, the risk is transformed from a static judgment of "whether the conditions are met" to a continuous quantitative judgment of "risk formation intensity." This directly links the magnitude of the risk to the degree of deviation of each dimension's indicators from the safe range and the superimposed effect of environmental triggers, enhancing the risk assessment's ability to reflect the actual toxicity triggering process. Finally, by aggregating the risks of single drugs as a whole and constructing individualized dynamic risk thresholds, a unified assessment of the overall exposure risk of patients in the context of combined drug use is achieved. At the same time, the judgment criteria are adaptively adjusted according to the proportion of environmental contraindications, thereby completing the linkage between risk intensity, individual status, and judgment thresholds. Ultimately, this achieves the technical effect of continuous, dynamic, and individualized prediction and early warning of adverse drug reactions. Attached Figure Description

[0045] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0046] Figure 1 This is a schematic diagram illustrating the steps of the drug adverse reaction prediction and assessment method that integrates multimodal data according to the present invention;

[0047] Figure 2 This is a schematic diagram of the structure of the drug adverse reaction prediction and assessment system that integrates multimodal data according to the present invention. Detailed Implementation

[0048] 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.

[0049] Please see Figure 1 In this first embodiment: a method for predicting and assessing adverse drug reactions by fusing multimodal data is provided, the method comprising the following steps:

[0050] Step S1: Construct a multimodal data acquisition module to collect clinical drug use data, drug molecular characteristic data, patient physiological indicator data, and environmental impact data; construct a drug-patient multidimensional feature vector.

[0051] Specifically, a multimodal data acquisition module is constructed, which is used to collect drug-patient multidimensional data. The multimodal data acquisition module includes a clinical data acquisition unit, a molecular structure analysis unit, a physiological indicator monitoring unit, and an environmental factor acquisition unit. The drug-patient multidimensional data includes clinical drug use data, drug molecular characteristic data, patient physiological indicator data, and environmental impact data.

[0052] The clinical data acquisition unit collects clinical medication data of the target drug through the hospital information system and pharmacovigilance system, including clinical drug dosage, frequency of use, concomitant medications, and the patient's history of adverse reactions. The molecular structure analysis unit, based on molecular docking simulation tools, collects drug molecular characteristic data of the target drug, including molecular structure parameters, target affinity, and binding capacity of key enzymes in metabolic pathways. The physiological indicator monitoring unit, including wearable sensors and laboratory testing equipment, collects patients' physiological indicator data, including liver and kidney function indicators, blood routine data, and gene polymorphism site information. The environmental factor acquisition unit collects patients' environmental impact data through questionnaires and lifestyle monitoring equipment, including smoking and drinking history, dietary structure, and work-rest patterns.

[0053] Furthermore, the data sampling time period is constructed, denoted as . ,in, Let A represent the a-th data sampling time point, and A represent the total number of data sampling time points; the data sampling time points are respectively... The clinical medication data, drug molecular characteristic data, patient physiological index data, and environmental impact data collected below are recorded as follows: , , and ;

[0054] Construct a drug-patient combination set, denoted as ,in, This represents the drug-patient combination of the c-th patient and the k-th drug, where C represents the total number of patients and K represents the total number of drugs; the data sampling time points are respectively... Drug-Patient Combination Collected Clinical drug use data, drug molecular characteristics data, patient physiological indicators data, and environmental impact data are denoted as , , and ;

[0055] Constructing data sampling time points Drug-Patient Combination The drug-patient multidimensional feature vector, denoted as .

[0056] In this invention, by constructing a multimodal data acquisition module and forming a drug-patient multidimensional feature vector, the transformation of drug risk assessment from single-source data to panoramic state modeling of "biology-pharmacology-clinical-environment" is realized. Its role is to expand the real influencing factors of adverse reactions from the traditional "drug record dimension" to: molecular structure layer (inherent drug toxicity and metabolic pathway), physiological state layer (individual susceptibility), and behavioral environment layer (external triggers).

[0057] The core purpose of this step is to establish a state-space basis for adverse reaction risk assessment, addressing the problem that existing technologies rely solely on electronic medical records or pharmacovigilance databases, resulting in a lack of individual variability characterization in risk assessment. By introducing intrinsic toxicological properties of drugs through molecular structure parameters and target affinity, the risk assessment is supported by pharmacological mechanisms rather than purely statistical correlations. The introduction of physiological indicators and environmental behavioral data enables the model to respond to "short-term state changes," providing a physical basis for subsequent dynamic assessment.

[0058] Step S2: Construct the adverse reaction risk demand vector of the drug; based on the drug-patient combination drug-patient multi-dimensional feature vector at the data sampling time point, calculate the risk factor fit between the drug-patient combination and the drug at the data sampling time point.

[0059] Specifically, adverse reaction information for all target drugs is retrieved from the drug safety database, and corresponding adverse reaction risk requirement parameters are preset for each drug to construct an adverse reaction risk requirement vector. The adverse reaction risk requirement vector for the k-th drug is denoted as... ,in, This represents the threshold of the key risk clinical indicator for the k-th drug. This represents the high-risk molecular characteristic parameter of the k-th drug. This represents the range of sensitive physiological indicators for the k-th drug. This represents the set of contraindications to the k-th drug;

[0060] Based on data sampling time points Drug-Patient Combination Drug-Patient Multidimensional Feature Vector Calculate the data sampling time point Drug-Patient Combination The formula for calculating the risk factor fit with the k-th drug is as follows:

[0061] ;

[0062] in, Indicates the data sampling time point Drug-Patient Combination Risk factor fit with the k-th drug This represents a preset constant. This represents the total number of factors in the set of contraindications for the k-th drug. This indicates the number of prohibited environmental factors included in the environmental impact data.

[0063] It should be noted that the core of this formula is to quantify the degree of matching between "drug risk needs" and "patient's actual condition", which is essentially a weighted average of "reverse normalization deviation". The closer it is to 1, the closer the patient's clinical medication, molecular characteristics, physiological indicators, and environmental factors are to the adverse reaction risk triggering conditions of the drug, and the higher the risk fit. The closer it is to 0, the greater the difference between the patient's condition and the drug's risk conditions, and the lower the suitability.

[0064] middle, It is the patient's actual clinical medication data (such as dosage and frequency). It is the key clinical risk threshold of a drug (such as the maximum safe daily dose), in absolute terms. To measure the deviation between actual medication use and the safety threshold, divide by This is for normalization (to eliminate the differences in threshold dimensions between different drugs, so that the results can be compared across drugs).

[0065] middle, It is the characteristic vector of the drug molecule (such as molecular structure parameters and target affinity). These are high-risk molecular characteristic parameters of drugs (such as toxic molecular structure thresholds), expressed using vector norms. This is because molecular features are multidimensional data, and the norm can measure the overall difference between two vectors;

[0066] middle, These are the patient's physiological indicators (such as liver and kidney function, and gene polymorphism). It refers to the range of sensitive physiological indicators of drugs (such as the safe range of liver function), which is consistent with the clinical dimension and measures the deviation of physiological state from the risk range;

[0067] middle, It is the patient's environmental factors (such as smoking and diet). It is a collection of environmental factors that contraindicate drugs (such as the contraindication of a certain antibiotic to alcohol consumption). It is the number of contraindications in the patient's environment, divided by The percentage of prohibited environmental factors is obtained by adding a constant to the total number of prohibited factors.

[0068] Step S3: Based on the drug-patient multidimensional feature vector and risk factor fit, calculate the comprehensive risk effect intensity of the drug-patient combination and the drug at the data sampling time point; calculate the environmental risk trigger ratio, and based on the comprehensive risk effect intensity and the environmental risk trigger ratio, calculate the adverse reaction risk value of the drug-patient combination and the drug at the data sampling time point.

[0069] Specifically, based on the data sampling time point Drug-Patient Combination Drug-Patient Multidimensional Feature Vector and data sampling time points Drug-Patient Combination Risk factor fit with the kth drug Calculate the data sampling time point Drug-Patient Combination The combined risk intensity of the drug and the k-th drug is calculated using the following formula:

[0070] ;

[0071] in, Indicates the data sampling time point Drug-Patient Combination The combined risk intensity with the kth drug This represents a preset constant;

[0072] It should be noted that the core of this formula is to upgrade "risk fit" to "risk intensity," which is essentially "matching degree × relative intensity of actual indicators":

[0073] The risk depends not only on "whether the patient is close to the drug risk condition" (fitness), but also on "the degree of closeness" (e.g., the more the dose is exceeded and the closer the molecular characteristics are to the toxicity threshold, the stronger the risk effect).

[0074] The three items in parentheses represent the relative strength of the actual indicators relative to the risk threshold:

[0075] like =15mg, =10mg, then ≈1.5 indicates that the dosage is 1.5 times the safe threshold, with an intensity greater than 1; if =8mg, =10mg, then the relative strength is The value is approximately 0.8, which is less than 1, indicating that the dose did not reach the risk threshold.

[0076] The product of relative intensity means that a high-risk action intensity (such as...) will only occur when both the fit is high and the actual index intensity is high. =0.7, total relative intensity =3, then =2.1; if =0.3, total relative intensity =3, then =0.9, the risk is significantly reduced.

[0077] The environmental risk trigger ratio is calculated using the following formula:

[0078] ;

[0079] in, Indicates the data sampling time point Drug-Patient Combination The proportion of environmental risk triggering for the kth drug, i.e., the data sampling time point Drug-Patient Combination The proportion of environmental factors that fall into the set of contraindicated environmental factors for the k-th drug. This indicates the preset impact factor;

[0080] It should be noted that the core of this formula is to quantify the amplifying effect of environmental factors on risk; its essence is a base coefficient plus the percentage of environmental taboos multiplied by the weight.

[0081] A base value of 1 indicates that, in the absence of contraindications, the environment has no amplifying effect on risk. =1, does not change the basic risk intensity. );

[0082] This indicates a preset impact factor (e.g., 0.5-2.0, adjusted according to drug type), used to regulate the weight of environmental factors—for environmentally sensitive drugs (e.g., antibiotics, sedatives). Take the larger value (e.g., 2.0) for drugs that are not sensitive to environmental factors (e.g., vitamins). Take the smaller value (e.g., 0.5);

[0083] Percentage of environmental taboos Consistent with the environmental component in the risk fit formula, this measures the proportion of contraindications in the patient's environment; the higher the proportion, the better. The larger the size, the stronger the environmental amplification effect.

[0084] Based on data sampling time points Drug-Patient Combination Combined risk intensity with the kth drug With data sampling time point Drug-Patient Combination Environmental risk trigger ratio of the kth drug Calculate the data sampling time point Drug-Patient Combination The adverse reaction risk value of the k-th drug is calculated using the following formula:

[0085] ;

[0086] in, Indicates the data sampling time point Drug-Patient Combination The adverse reaction risk value compared to the k-th drug.

[0087] In this invention, by constructing a comprehensive risk effect intensity and an environmental risk trigger ratio, the transformation model of risk from a static matching relationship to a risk formation driving force is realized; whether the risk conditions are met is further transformed into the intensity level of risk formation.

[0088] Overall risk intensity By coupling the fit degree with the multidimensional state ratio, the magnitude of the risk is proportional to the degree to which it exceeds the safe range, avoiding information loss caused by binarized judgment; environmental risk trigger ratio. As an independent amplification item, it reflects the medical fact that adverse reactions are often triggered by precipitating factors rather than determined by a single intrinsic factor; risk value The product form is used to amplify the effect of environmental factors when a high-risk state is met, which is consistent with the actual toxicity triggering mechanism.

[0089] Step S4: Based on the adverse reaction risk value, calculate the comprehensive adverse reaction risk value of the patient and all drugs at the data sampling time point; calculate the individualized dynamic risk threshold of the patient at the data sampling time point; analyze and provide dynamic early warning.

[0090] Specifically, based on the data sampling time point Drug-Patient Combination Adverse reaction risk value of the kth drug Calculate the data sampling time point The combined adverse reaction risk value for the c-th patient and all drugs is calculated using the following formula:

[0091] ;

[0092] in, Indicates the data sampling time point The combined adverse reaction risk value of the c-th patient and all drugs, where K represents the total number of drugs;

[0093] Calculate data sampling time points The individualized dynamic risk threshold for the c-th patient is calculated using the following formula:

[0094] ;

[0095] in, Indicates the data sampling time point The individualized dynamic risk threshold for the c-th patient. This represents the preset basic risk threshold constant.

[0096] It should be noted that the core of this formula is to construct an "individualized early warning threshold that changes with the environment," which is essentially "base threshold × environmental adjustment coefficient": It is a preset basic risk threshold (e.g., 3.0, a safety threshold based on a large amount of clinical data statistics). It can be expressed as an environmental adjustment coefficient. The logic is that the more contraindicated environments a patient is in, the smaller the adjustment coefficient and the lower the threshold—because the more dangerous the environment, the lower the tolerance for risk should be, and the easier it is to trigger an early warning.

[0097] Different patients have different environmental factors, use different medications, and therefore the average percentage of environmental contraindications varies. It is a threshold specific to the patient's current state, rather than a uniform, fixed threshold.

[0098] Furthermore, if the data sampling time point The combined adverse reaction risk value of the cth patient and all drugs Greater than or equal to the data sampling time point Individualized dynamic risk threshold for the cth patient Then determine the data sampling time point. If a high-risk adverse drug reaction is detected, an early warning will be issued;

[0099] The system acquires the comprehensive adverse reaction risk value at each data sampling time point in real time and calculates the individualized dynamic risk threshold in real time to perform dynamic real-time prediction and assessment of adverse drug reactions.

[0100] In this invention, by integrating all drug risks of a patient and constructing an individualized dynamic threshold, the risk assessment is transformed from a fixed threshold mode to an individualized adaptive mode.

[0101] This step transforms single-drug risk into overall exposure risk and dynamically adjusts the assessment criteria based on environmental conditions; through The risk of all drugs is averaged to address the issue that the risk of a single drug in a combination therapy context cannot reflect the overall exposure status; dynamic threshold. As the proportion of contraindicated environmental factors changes, the risk assessment criteria are automatically adjusted to the fluctuations in the patient's condition, improving individualized adaptability; real-time comparison of risk values ​​with thresholds and output of early warnings enable the system to have continuous monitoring and real-time intervention capabilities.

[0102] Please see Figure 2 In this second embodiment: a drug adverse reaction prediction and assessment system integrating multimodal data is provided. The system includes: a data acquisition and vector construction module, a fit calculation module, an intensity calculation and risk value calculation module, and a threshold calculation and analysis module.

[0103] The data acquisition and vector construction module includes: constructing a multimodal data acquisition module to collect clinical medication data, drug molecular characteristic data, patient physiological indicator data, and environmental impact data; and constructing a drug-patient multidimensional feature vector.

[0104] The fit calculation module: constructs a drug adverse reaction risk demand vector; and calculates the risk factor fit between the drug-patient combination and the drug at the data sampling time point based on the drug-patient multi-dimensional feature vector of the drug-patient combination.

[0105] The intensity calculation and risk value calculation module calculates the comprehensive risk intensity of the drug-patient combination and the drug at the data sampling time point based on the drug-patient multi-dimensional feature vector and risk factor fit; it also calculates the environmental risk trigger ratio and, based on the comprehensive risk intensity and the environmental risk trigger ratio, calculates the adverse reaction risk value of the drug-patient combination and the drug at the data sampling time point.

[0106] The threshold calculation and analysis module calculates the overall adverse reaction risk value of the patient and all drugs at the data sampling time point based on the adverse reaction risk value; calculates the individualized dynamic risk threshold of the patient at the data sampling time point; and analyzes and provides dynamic early warning.

[0107] Furthermore, the intensity calculation and risk value calculation module includes an intensity calculation unit and a risk value calculation unit;

[0108] The intensity calculation unit calculates the combined risk intensity of the drug-patient combination and the kth drug at the data sampling time point based on the drug-patient combination's multi-dimensional feature vector at the data sampling time point and the risk factor fit between the drug-patient combination and the kth drug at the data sampling time point.

[0109] The risk value calculation unit calculates the environmental risk trigger ratio and, based on the combined risk intensity of the drug-patient combination and the kth drug at the data sampling time point and the environmental risk trigger ratio of the drug-patient combination and the kth drug at the data sampling time point, calculates the adverse reaction risk value of the drug-patient combination and the kth drug at the data sampling time point.

[0110] Furthermore, the threshold calculation and analysis module includes a threshold calculation unit and an analysis unit;

[0111] The threshold calculation unit: based on the adverse reaction risk value of the drug-patient combination and the kth drug at the data sampling time point, calculates the comprehensive adverse reaction risk value of the cth patient and all drugs at the data sampling time point; and calculates the individualized dynamic risk threshold of the cth patient at the data sampling time point.

[0112] The analysis unit: if the combined adverse reaction risk value of the c-th patient and all drugs at the data sampling time point is greater than or equal to the individualized dynamic risk threshold of the c-th patient at the data sampling time point, then it determines that there is a high-risk adverse drug reaction at the data sampling time point and issues an early warning; it acquires the combined adverse reaction risk value at each data sampling time point in real time and calculates the individualized dynamic risk threshold in real time to perform dynamic real-time prediction and assessment of adverse drug reactions.

[0113] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0114] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting and assessing adverse drug reactions by integrating multimodal data, characterized in that, The method includes the following steps: Step S1: Construct a multimodal data acquisition module to collect clinical medication data, drug molecular characteristic data, patient physiological indicator data, and environmental impact data; construct a drug-patient multidimensional feature vector; Step S2: Construct the adverse reaction risk demand vector of the drug; based on the drug-patient combination drug-patient multi-dimensional feature vector at the data sampling time point, calculate the risk factor fit between the drug-patient combination and the drug at the data sampling time point; Step S3: Based on the drug-patient multidimensional feature vector and risk factor fit, calculate the comprehensive risk effect intensity of the drug-patient combination and the drug at the data sampling time point; calculate the environmental risk trigger ratio, and based on the comprehensive risk effect intensity and the environmental risk trigger ratio, calculate the adverse reaction risk value of the drug-patient combination and the drug at the data sampling time point. Step S4: Based on the adverse reaction risk value, calculate the comprehensive adverse reaction risk value of the patient and all drugs at the data sampling time point; calculate the individualized dynamic risk threshold of the patient at the data sampling time point; analyze and provide dynamic early warning.

2. The method for predicting and assessing adverse drug reactions by fusing multimodal data according to claim 1, characterized in that, The specific implementation process of step S1 includes: A multimodal data acquisition module is constructed, which is used to collect drug-patient multidimensional data. The multimodal data acquisition module includes a clinical data acquisition unit, a molecular structure analysis unit, a physiological indicator monitoring unit, and an environmental factor acquisition unit. The drug-patient multidimensional data includes clinical drug use data, drug molecular characteristic data, patient physiological indicator data, and environmental impact data. The clinical data acquisition unit collects clinical medication data of the target drug through the hospital information system and pharmacovigilance system; the molecular structure analysis unit collects drug molecular characteristic data of the target drug based on molecular docking simulation tools; the physiological indicator monitoring unit includes wearable sensors and laboratory testing equipment to collect patients' physiological indicator data; and the environmental factor acquisition unit collects patients' environmental impact data through questionnaires and lifestyle monitoring equipment.

3. The method for predicting and assessing adverse drug reactions by fusing multimodal data according to claim 2, characterized in that, The specific implementation process of step S1 also includes: The data sampling time period is defined as follows: ,in, Let A represent the a-th data sampling time point, and A represent the total number of data sampling time points; the data sampling time points are respectively... The clinical medication data, drug molecular characteristic data, patient physiological index data, and environmental impact data collected below are recorded as follows: , , and ; Construct a drug-patient combination set, denoted as ,in, This represents the drug-patient combination of the c-th patient and the k-th drug, where C represents the total number of patients and K represents the total number of drugs; the data sampling time points are respectively... Drug-Patient Combination Collected Clinical drug use data, drug molecular characteristics data, patient physiological indicators data, and environmental impact data are denoted as , , and ; Constructing data sampling time points Drug-Patient Combination The drug-patient multidimensional feature vector, denoted as .

4. The method for predicting and assessing adverse drug reactions by fusing multimodal data according to claim 3, characterized in that, The specific implementation process of step S2 includes: Adverse reaction information for all target drugs is retrieved from the drug safety database, and corresponding adverse reaction risk requirement parameters are preset for each drug. An adverse reaction risk requirement vector is constructed, and the adverse reaction risk requirement vector for the k-th drug is denoted as... ,in, This represents the threshold of the key risk clinical indicator for the k-th drug. This represents the high-risk molecular characteristic parameter of the k-th drug. This represents the range of sensitive physiological indicators for the k-th drug. This represents the set of contraindications to the k-th drug; Based on data sampling time points Drug-Patient Combination Drug-Patient Multidimensional Feature Vector Calculate the data sampling time point Drug-Patient Combination The formula for calculating the risk factor fit with the k-th drug is as follows: ； in, Indicates the data sampling time point Drug-Patient Combination Risk factor fit with the k-th drug This represents a preset constant. This represents the total number of factors in the set of contraindications for the k-th drug. This indicates the number of prohibited environmental factors included in the environmental impact data.

5. The method for predicting and assessing adverse drug reactions by fusing multimodal data according to claim 4, characterized in that, The specific implementation process of step S3 includes: Based on data sampling time points Drug-Patient Combination Drug-Patient Multidimensional Feature Vector and data sampling time points Drug-Patient Combination Risk factor fit with the kth drug Calculate the data sampling time point Drug-Patient Combination The combined risk intensity of the drug and the k-th drug is calculated using the following formula: ； in, Indicates the data sampling time point Drug-Patient Combination The combined risk intensity with the kth drug This represents a preset constant; The environmental risk trigger ratio is calculated using the following formula: ； in, Indicates the data sampling time point Drug-Patient Combination The proportion of environmental risk triggering for the kth drug, i.e., the data sampling time point Drug-Patient Combination The proportion of environmental factors that fall into the set of contraindicated environmental factors for the k-th drug. This indicates the pre-defined impact factor; Based on data sampling time points Drug-Patient Combination Combined risk intensity with the kth drug With data sampling time point Drug-Patient Combination Environmental risk trigger ratio of the kth drug Calculate the data sampling time point Drug-Patient Combination The adverse reaction risk value of the k-th drug is calculated using the following formula: ； in, Indicates the data sampling time point Drug-Patient Combination The adverse reaction risk value compared to the k-th drug.

6. The method for predicting and assessing adverse drug reactions by fusing multimodal data according to claim 5, characterized in that, The specific implementation process of step S4 includes: Based on data sampling time points Drug-Patient Combination Adverse reaction risk value of the kth drug Calculate the data sampling time point The combined adverse reaction risk value for the c-th patient and all drugs is calculated using the following formula: ； in, Indicates the data sampling time point The combined adverse reaction risk value of the c-th patient and all drugs, where K represents the total number of drugs; Calculate data sampling time points The individualized dynamic risk threshold for the c-th patient is calculated using the following formula: ； in, Indicates the data sampling time point The individualized dynamic risk threshold for the c-th patient. This represents the preset basic risk threshold constant.

7. The method for predicting and assessing adverse drug reactions by fusing multimodal data according to claim 6, characterized in that, The specific implementation process of step S4 also includes: If the data sampling time point The combined adverse reaction risk value of the cth patient and all drugs Greater than or equal to the data sampling time point Individualized dynamic risk threshold for the cth patient Then determine the data sampling time point. If a high-risk adverse drug reaction is detected, an early warning will be issued; The system acquires the comprehensive adverse reaction risk value at each data sampling time point in real time and calculates the individualized dynamic risk threshold in real time to perform dynamic real-time prediction and assessment of adverse drug reactions.

8. A drug adverse reaction prediction and assessment system integrating multimodal data, executing the drug adverse reaction prediction and assessment method integrating multimodal data as described in any one of claims 1-7, characterized in that, The system includes: a data acquisition and vector construction module, an fitness calculation module, an intensity calculation and risk value calculation module, and a threshold calculation and analysis module; The data acquisition and vector construction module includes: constructing a multimodal data acquisition module to collect clinical medication data, drug molecular characteristic data, patient physiological indicator data, and environmental impact data; and constructing a drug-patient multidimensional feature vector. The fit calculation module: constructs a drug adverse reaction risk demand vector; and calculates the risk factor fit between the drug-patient combination and the drug at the data sampling time point based on the drug-patient multi-dimensional feature vector of the drug-patient combination. The intensity calculation and risk value calculation module calculates the comprehensive risk intensity of the drug-patient combination and the drug at the data sampling time point based on the drug-patient multi-dimensional feature vector and risk factor fit; it also calculates the environmental risk trigger ratio and, based on the comprehensive risk intensity and the environmental risk trigger ratio, calculates the adverse reaction risk value of the drug-patient combination and the drug at the data sampling time point. The threshold calculation and analysis module calculates the overall adverse reaction risk value of the patient and all drugs at the data sampling time point based on the adverse reaction risk value; calculates the individualized dynamic risk threshold of the patient at the data sampling time point; and analyzes and provides dynamic early warning.

9. The adverse drug reaction prediction and assessment system integrating multimodal data according to claim 8, characterized in that: The intensity calculation and risk value calculation module includes an intensity calculation unit and a risk value calculation unit; The intensity calculation unit calculates the combined risk intensity of the drug-patient combination and the kth drug at the data sampling time point based on the drug-patient combination's multi-dimensional feature vector at the data sampling time point and the risk factor fit between the drug-patient combination and the kth drug at the data sampling time point. The risk value calculation unit calculates the environmental risk trigger ratio. Based on the combined risk intensity of the drug-patient combination and the kth drug at the data sampling time point and the environmental risk triggering ratio of the drug-patient combination and the kth drug at the data sampling time point, the adverse reaction risk value of the drug-patient combination and the kth drug at the data sampling time point is calculated.

10. The adverse drug reaction prediction and assessment system integrating multimodal data according to claim 9, characterized in that: The threshold calculation and analysis module includes a threshold calculation unit and an analysis unit; The threshold calculation unit calculates the combined adverse reaction risk value of the c-th patient and all drugs at the data sampling time point based on the adverse reaction risk value of the drug-patient combination and the k-th drug at the data sampling time point. Calculate the individualized dynamic risk threshold for the c-th patient at the data sampling time point; The analysis unit: if the combined adverse reaction risk value of the c-th patient and all drugs at the data sampling time point is greater than or equal to the individualized dynamic risk threshold of the c-th patient at the data sampling time point, then it determines that there is a high-risk adverse drug reaction at the data sampling time point and issues an early warning; it acquires the combined adverse reaction risk value at each data sampling time point in real time and calculates the individualized dynamic risk threshold in real time to perform dynamic real-time prediction and assessment of adverse drug reactions.

Images