Active drug adverse reaction detection system utilizing a periodic conversion database and artificial intelligence

Abstract

A method for generating prediction results for adverse drug reactions using artificial intelligence techniques, performed by a computing device, is disclosed. The method may include the steps of: constructing a periodic transformation database using raw data for adverse drug reaction detection, which is scheduled to be updated in predetermined time intervals; determining a plurality of input variables for training an artificial intelligence-based model for adverse drug reaction detection based on the data stored in the periodic transformation database; and training the artificial intelligence-based prediction model using the determined input variables and a training dataset containing adverse reaction information for a specific drug and a specific disease, such that the prediction model outputs adverse reaction information for a specific drug and a specific disease in response to receiving input of the input variables.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and more specifically, to a method and apparatus for active drug adverse reaction detection using artificial intelligence technology.

Background Art

[0002] Drug adverse reactions are defined as harmful and unintended reactions to pharmaceuticals, which can endanger patients' health and vary from mild to very severe. In recent years, the global sales volume of pharmaceuticals has been increasing, and drug adverse reactions also tend to increase. Therefore, opinions suggesting the need for countermeasures have emerged, and pharmaceutical safety evaluations and domestic and foreign regulatory authorities are conducting research on the causative factors of drug adverse reactions and proactive countermeasures. The World Health Organization (WHO) manages "VigiBase" (https: / / who-umc.org / vigibase / ), a global pharmaceutical safety information database that collects drug adverse reaction information occurring worldwide. The US Food and Drug Administration (FDA) operates a database called "FAERS (FDA Adverse Event Reporting System)" as a regulatory agency to ensure the safety and effectiveness of pharmaceuticals and collects drug adverse reaction information reported to the FDA. "SIDER (SideEffectResource, http: / / sideeffects.embl.de / about / )" managed by a research team at the University of Copenhagen in Denmark is a database containing information on marketed drugs and their adverse reactions. In Korea, the Ministry of Food and Drug Safety operates "KAERS (https: / / kaers.drugsafe.or.kr / ). Standard information on drug adverse reactions can be collected from such diverse databases related to drug adverse reactions.

[0003] In recent years, research utilizing artificial intelligence has advanced in the field of adverse drug reactions, with machine learning, deep learning, and natural language processing being the main technologies being used. A study that searched papers published from January 1, 2015 to July 9, 2021, and examined those containing key terms related to artificial intelligence and drug safety in the title or abstract, found that the main application areas among the retrieved papers included identification of adverse drug reactions and drug responses (57.6%), processing of safety reports (21.2%), extraction of drug-drug interactions (7.6%), identification of population groups at high risk of drug toxicity or personalized treatment guidelines (7.6%), adverse drug reaction prediction (3.0%), clinical trial simulation (1.5%), and integration of predictive uncertainty in diagnostic classifiers. Furthermore, it is expected that the use of artificial intelligence in the field of adverse drug reaction and drug response prediction will expand even further in the future.

[0004] To actively detect adverse drug reactions, drug use information and associated reactions (diagnostic test data, disease diagnosis data, nursing records, etc.) are essential. A database converted to a common data model standard using electronic medical records can collect data from multiple healthcare institutions, and by utilizing a regularly updated database, data from healthcare institutions can be used more efficiently.

[0005] In connection with this, the Republic of Korea Published Patent Publication No. 2008-0042256 has been devised. [Overview of the project] [Problems that the invention aims to solve]

[0006] This disclosure was devised in response to the aforementioned background technology and aims to provide a method and apparatus for active drug adverse reaction detection using artificial intelligence technology.

[0007] The technical challenges described herein are not limited to those mentioned above, and any other technical challenges not mentioned above will be clearly understood by a person of ordinary skill as described below. [Means for solving the problem]

[0008] According to one embodiment of the present disclosure, a method is disclosed for generating predictive results for adverse drug reactions using artificial intelligence technology performed by a computing device. The method may include the steps of: constructing a periodic transformation database using raw data for adverse drug reaction detection, which is scheduled to be updated in predetermined time intervals; determining a set of input variables for training an artificial intelligence-based predictive model for adverse drug reaction detection based on the data stored in the periodic transformation database; and training the predictive model using the determined input variables and a training dataset containing adverse reaction information for a specific drug and a specific disease, such that the predictive model outputs adverse reaction information for a specific drug and a specific disease in response to receiving the input variables.

[0009] In one embodiment, the step of constructing the periodic conversion database may include: acquiring raw data for drug adverse reaction detection from an electronic medical record (EMR) - the raw data including at least one of drug administration-related data, disease diagnosis data, diagnostic test data, or patient nursing record data; converting the terminology of the raw data to correspond to the common data standard of the prediction model; and constructing the periodic conversion database, which is updated in predetermined time intervals, using at least one of a database linkage method, a file linkage method, or a change history classification linkage method.

[0010] In one embodiment, the raw data can be further obtained from at least one of FAERS, KAERS, WHO-ART, SIDER, or EU-ADR.

[0011] In one embodiment, the input variable may include at least one of the following: a demographic variable, a drug use-related variable, a diagnostic test-related variable, a nursing record-related variable, or a drug side effect diagnosis-related variable.

[0012] In one embodiment, the demographic variable includes at least one of sex, age, or region; the drug use-related variable includes at least one of drug type, duration of drug use, dosage, or drug use history; the diagnostic test-related variable includes at least one of imaging tests, blood tests, or urine tests; the nursing record-related variable is a variable obtained from nursing record paper, obtained through at least one of optical character recognition (OCR), natural language processing (NLP), or text mining; and the drug side effect diagnosis-related variable may include information related to disease diagnosis due to drug side effects.

[0013] In one embodiment, the input variables can be determined using at least one of the following methods: a first method for determining the importance of variables by using perturbed input data in which at least a portion of the variables have been modified and output data output from the prediction model in response to the perturbed input data; a second method for selecting input variables corresponding to principal components by converting data in a high-dimensional space to a low-dimensional space; or a third method for selecting input variables in a direction that reduces impurity on the decision tree.

[0014] In one embodiment, the input variables can be preprocessed based on the following steps: outlier processing; imputation of missing values; categorization or recategorization; and normalization or standardization.

[0015] In one embodiment, the step of training the predictive model for detecting adverse drug reactions may include: determining the predictive model for detecting adverse drug reactions from among a plurality of candidate models; and training the predictive model to output the presence or absence of adverse drug reactions to a particular drug or a particular disease, and the probability of adverse drug reactions occurring to a particular drug or a particular disease, using the preprocessed input variables as input data.

[0016] In one embodiment, the prediction model may be a model selected from among candidate models consisting of Random Forest, Logistic Regression, Support Vector Machine, Decision Tree, Naive Bayes, k-Nearest Neighbor, XG Boost, Light GBM, Cat Boost, Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Transformer, and Generative Pre-trained Transformer (GPT).

[0017] In one embodiment, the prediction model may include the steps of: obtaining output results from each of the candidate models using a predetermined test dataset; performing a performance evaluation on the output results from each of the candidate models; and determining one of the candidate models as the prediction model based on the results of the performance evaluation.

[0018] In one embodiment, the adverse reaction information for a specific drug and a specific disease includes: first information indicating the presence or absence of an adverse drug reaction to the specific drug; second information indicating the probability of an adverse drug reaction occurring to the specific drug; third information indicating the presence or absence of an adverse drug reaction to the specific disease; and fourth information indicating the probability of an adverse drug reaction occurring to the specific disease.

[0019] In one embodiment, the method may include the step of transmitting the output results of the prediction model for adverse drug reactions to a user terminal.

[0020] In one embodiment, the output results transmitted to the user terminal may include at least one of the following: a first result showing a binary classification of drug adverse reactions as detected or undetected; a second result showing a probability notation for drug adverse reactions; or a third result showing a three-level category notation relating drug, adverse reaction, and risk.

[0021] In one embodiment, the output results from the prediction model can be used to retrain the prediction model based on the accuracy evaluated for the output results.

[0022] In one embodiment, feedback information corresponding to the output result transmitted from a user terminal that receives the output result from the prediction model can be used to retrain the prediction model.

[0023] In one embodiment, a computer program stored in a computer-readable storage medium is disclosed. The computer program causes a computing device to perform an operation at runtime using artificial intelligence technology to generate prediction results for adverse drug reactions, the operation of constructing a periodic conversion database using raw data for adverse drug reaction detection, which is scheduled to be updated in predetermined time intervals; determining a plurality of input variables to be used for training an artificial intelligence-based prediction model for adverse drug reaction detection based on the data stored in the periodic conversion database; and training the prediction model to output adverse drug reaction information for specific drugs and specific diseases in response to receiving input of the input variables, using the determined input variables and a training dataset containing adverse reaction information for specific drugs and specific diseases.

[0024] A computing device according to an embodiment is disclosed. The computing device can include at least one processor and a memory. The at least one processor is configured to: construct a periodic conversion database that is scheduled to be updated in predetermined time period units using raw data for drug adverse reaction detection; determine a plurality of input variables for learning an artificial intelligence-based prediction model for the drug adverse reaction detection based on the data stored in the periodic conversion database; and use the determined input variables and a learning dataset including adverse reaction information for a specific drug and a specific disease to train the prediction model such that the prediction model outputs the adverse reaction information for the specific drug and the specific disease in response to receiving an input of the input variables.

[0025] A method for generating a prediction result for a drug adverse reaction using an artificial intelligence technique executed by a computing device according to an embodiment is disclosed. The method can include: obtaining input data for the drug adverse reaction detection based on the data stored in a periodic conversion database that is scheduled to be updated in predetermined time period units using raw data for drug adverse reaction detection; and obtaining adverse reaction information for a specific drug and a specific disease from the input data using an artificial intelligence-based prediction model.

Advantages of the Invention

[0026] The method and apparatus according to an embodiment of the present disclosure can realize management of drug adverse reactions so as to be generally applicable without distinguishing between inpatient / outpatient categories or specific drugs, and contribute to the construction of a virtuous cycle in the field of drug safety.

Brief Description of the Drawings

[0027] [Figure 1] A block diagram of a computing device according to an embodiment of the present disclosure is schematically shown. [Figure 2] Shows an exemplary structure of an artificial intelligence-based model according to an embodiment of the present disclosure. [Figure 3] Exemplarily shows a method for generating an output result of a model including a prediction for a drug adverse reaction according to an embodiment of the present disclosure. [Figure 4] Shows a flowchart for constructing a periodic conversion database including principles of periodic data loading and change history table loading in an electronic medical record and a periodic conversion database according to an embodiment of the present disclosure. [Figure 5] Shows the concept of file linkage for constructing a periodic conversion database according to an embodiment of the present disclosure. [Figure 6] Exemplarily shows the overall process leading to database construction, development of a drug adverse reaction detection artificial intelligence, and notification of abnormal detection results according to an embodiment of the present disclosure. [Figure 7] Is a schematic diagram of a computing environment according to an embodiment of the present disclosure.

Mode for Carrying Out the Invention

[0028] Various embodiments are described with reference to the drawings. In this specification, various explanations are presented to assist in understanding the present disclosure. Before describing the specific content for implementing the present disclosure, it should be noted that configurations not directly related to the technical gist of the present disclosure are omitted within the scope that does not impair the technical gist of the present invention. Also, the terms or phrases used in this specification and the claims should be interpreted as meanings and concepts consistent with the technical idea of the present invention based on the principle that the inventor can define the concept of appropriate terms for explaining his invention in the best way.

[0029] As used herein, the terms “component,” “module,” “system,” and “unit” refer to computer-related entities, hardware, firmware, software, combinations of software and hardware, or software execution, and can be used interchangeably. For example, a component may be, but is not limited to, a process executed on a processor, a processor, an object, an execution thread, a program, and / or a computer. For example, an application running on a computing device and the entire computing device can be a component. One or more components may reside within a processor and / or an execution thread. A component may be localized within a single computer. A component may be distributed between two or more computers. These components can also be executed from a variety of computer-readable media having diverse data structures stored within them. Components can communicate through local and / or remote processing according to signals having one or more data packets (for example, data from one component interacting with another component in a local system or distributed system, and / or data transmitted to other systems via a network such as the Internet via signals).

[0030] Furthermore, the term "or" is intended to mean inclusive, not exclusive. That is, unless otherwise specified or not clear from the context, "X uses A or B" is intended to mean either of the natural inclusive substitutions. That is, if X uses A; X uses B; or X uses both A and B, "X uses A or B" can apply to any of these cases. Also, the term "and / or" as used herein should be understood to refer to and include any possible combination of one or more of the listed related items.

[0031] Furthermore, the terms “comprise” and / or “comprising” should be understood to mean the presence of the characteristic and / or component in question. However, the terms “comprise” and / or “comprising” should not be understood to exclude the presence or addition of one or more other characteristics, components and / or groups thereof. Moreover, unless otherwise specified, or unless it is clear from the context that a singular form is being referred to, the singular form in this specification and in the claims should generally be interpreted as meaning “one or more.”

[0032] Furthermore, the terms "at least one of A or B" or "at least one of A and B" should be interpreted as meaning "including only A," "including only B," and "a combination of A and B."

[0033] A person of ordinary skill should recognize that the various exemplary logical components, blocks, modules, circuits, means, logic, and algorithms described in relation to the embodiments disclosed herein can be embodied in electronic hardware, computer software, or a combination of both. To clearly illustrate the interoperability of hardware and software, various exemplary components, blocks, means, logic, modules, circuits, and stages have been generally described in terms of their functionality. Whether such functionality is embodied in hardware or software depends on the specific application and design limitations imposed on the overall system. A skilled technician may embodied the functionalities described in various ways for each specific application; however, such a decision to embodied should not be construed as exceeding the scope of this disclosure.

[0034] The descriptions of the embodiments presented are provided so that a person with ordinary skill in the art of this disclosure may utilize or implement the invention. Various modifications to these embodiments will be obvious to a person with ordinary skill in the art of this disclosure. The general principles defined herein can be applied to other embodiments without departing from the scope of this disclosure. Accordingly, the invention is not limited to the embodiments presented herein. The invention should be interpreted in the broadest sense consistent with the principles and novel features presented herein.

[0035] In this disclosure, terms such as "First," "Second," "Third," etc., which are represented as "Nth," are used to distinguish at least one entity. For example, entities represented as "First" and "Second" may be identical or different from each other.

[0036] As used in this disclosure, the term "adverse drug reaction (ADR)" refers to a harmful and unintended reaction that occurs when a drug or other medical device is administered or used, and in which a causal relationship with the drug or device cannot be ruled out. Furthermore, it may include not only adverse reactions that occur at normal doses, but also adverse drug reactions that occur when a drug is used in an overdose intentionally or accidentally, when a drug is abused, withdrawal symptoms, and cases where the expected pharmacological effect does not occur.

[0037] In this context, "drug adverse reaction monitoring" refers to activities aimed at promoting rational drug use and preventing drug-related harm by rapidly and systematically collecting or evaluating various adverse events that occur during drug use, taking appropriate countermeasures, and providing safety information and the results of those measures to healthcare professionals, consumers, etc.

[0038] Examples related to ADR include side effects and adverse drug events (ADEs). Side effects are a concept in contrast to principal actions, which are the effects that occur when a drug is used for a specific purpose. They refer to all unintended effects that occur when a drug is administered at a normal dose. Because this term encompasses all effects other than therapeutic effects, it is a concept that can be widely used in various fields that comprehensively study all effects, regardless of whether they are harmful or not.

[0039] Adverse drug events (ADEs) refer to undesirable and unintended signs, symptoms, or illnesses that occur during the administration or use of a drug, regardless of whether they are causally related to the drug.

[0040] Adverse drug reactions (ADRs) are harmful and unintended reactions that occur during the administration or use of pharmaceuticals or other substances according to their normal usage instructions. In cases where a causal relationship with the drug cannot be ruled out, ADRs are a concept in which management by pharmacists and medical teams is considered important because they are often predictable and preventable.

[0041] Drug metabolism primarily occurs in the liver, where metabolic reactions such as oxidation / reduction, hydrolysis, hydration, conjugation, condensation, and isomerization of drugs are carried out by the cytochrome P450 (CYP450) metabolic enzyme system in the endoplasmic reticulum of hepatocytes. The four main parameters of pharmacokinetics are absorption, distribution, metabolism, and excretion. Dose, frequency of administration, route of administration, tissue distribution, and protein binding of drugs to receptors and other proteins can all affect drug metabolism. Furthermore, pathological factors, including gastrointestinal diseases, cardiopulmonary diseases, respiratory diseases, and renal excretion disorders, can also influence drug metabolism.

[0042] Embodiments in this disclosure may include steps for deriving predictable ADRs based on the type of drug administered for an indication, demographic background, racial / individual genotype / phenotype, medical history, etc., in the order of causality (e.g., certain, likely, possible, unlikely, conditional), classification by severity of response (e.g., mild, moderate, severe), or frequency of occurrence.

[0043] Figure 1 schematically shows a block diagram of a computing device 100 according to one embodiment of the present disclosure.

[0044] A computing device 100 according to one embodiment of the present disclosure may include a processor 110 and memory 130.

[0045] The configuration of the computing device 100 shown in Figure 1 is merely a simplified example. In one embodiment of this disclosure, the computing device 100 may include other configurations for performing the computing environment of the computing device 100, and only some of the disclosed configurations may constitute the computing device 100.

[0046] In this disclosure, the computing device 100 may mean any form of node that constitutes a system for embodying an embodiment of this disclosure. The computing device 100 may mean any form of user terminal or any form of server. The components of the computing device 100 described above are illustrative and may be partially excluded or include additional components. For example, if the computing device 100 described above includes a user terminal, an output unit (not shown) and an input unit (not shown) may be included within the scope of the computing device 100.

[0047] The computing device 100 in this disclosure can perform technical features of embodiments of this disclosure as described later. For example, the computing device 100 can use an artificial intelligence-based predictive model that uses input data corresponding to the patient's medical history, drug allergy history, administered drugs and dosages to generate predictive results that include adverse drug reactions (ADRs) that may occur in the patient corresponding to the input data (for example, a particular drug may trigger excessive immune system activity in a particular individual, inducing a cellular immune response cascade).

[0048] As a further example, the computing device 100 can acquire information on a patient's demographic information, drug use history, medical history, and blood test data (for example, a 65-year-old East Asian male, who has taken serotonin reuptake inhibitors (SSRIs), divalproate, and statins for the past 26 months, and has experienced abdominal pain and constipation as drug-related adverse events) from a database. Based on the acquired information, it can generate ADR prediction results corresponding to the drug or specific disease and provide corresponding materials, including cost-benefit analysis of drug use.

[0049] As a further example, the computing device 100 can more efficiently collect raw data from multiple medical institutions by constructing a periodic conversion database that has been converted to a common data model standard using electronic medical records. The computing device 100 can be configured to perform more up-to-date detection of adverse drug reactions by utilizing a periodic conversion database that is updated regularly.

[0050] In one embodiment of this disclosure, the computing device 100 can obtain the results of a nucleotide sequence analysis (e.g., Next Generation Sequencing) from a server or an external entity. In another embodiment, the computing device 100 can also perform nucleotide sequence analysis on a proteome or gene (e.g., DNA or RNA) obtained from a biological sample derived from a subject. The terms used in this disclosure, nucleotide sequence analysis, can be performed by any form of method capable of analyzing the sequence of bases, and may include, but are not limited to, whole genome sequencing, whole exome sequencing, or whole transcriptome sequencing.

[0051] In this disclosure, the terms "subject" may mean an object or individual from which a biological sample is obtained, including saliva, hair follicles, proteome, genes, and / or combinations thereof.

[0052] The term "sample" as used in this disclosure is not limited to samples obtained from a subject or individual whose gene genome and / or gene phenotype are to be determined, and may include, for example, cells or tissues obtained from biopsies, blood, whole blood, serum, plasma, saliva, cerebrospinal fluid, various secretions, urine, and / or feces. Preferably, the sample may be selected from the group consisting of blood, plasma, serum, saliva, nasal fluid, sputum, ascites, vaginal secretions, and / or urine, and more preferably blood, plasma, or serum. The sample may be pre-treated before use for detection or diagnosis. For example, pre-treatment methods may include homogenization, filtration, distillation, extraction, concentration, inactivation of interfering components, and / or addition of reagents. In this disclosure, biological samples may, but are not limited to, tissues, cells, whole blood, and / or blood.

[0053] In one embodiment, the processor 110 may consist of at least one core and may include processors for data analysis and / or processing, such as a central processing unit (CPU), a general-purpose graphics processing unit (GPGPU), or a tensor processing unit (TPU) of the computing device 100.

[0054] The processor 110 reads a computer program stored in the memory 130 and, according to one embodiment of the present disclosure, can generate a prediction result including whether or not an adverse drug event (ADE) has occurred and / or the probability thereof.

[0055] According to one embodiment of this disclosure, the processor 110 can also perform calculations for training a neural network. In deep learning (DL), the processor 110 can perform calculations for training a neural network, such as processing input data for training, extracting features from input data, calculating errors, and updating the weights of the neural network using backpropagation. At least one of the CPU, GPGPU, and TPU of the processor 110 can process the training of the network function. For example, the CPU and GPGPU can both process the training of the network function and data classification using the network function. In addition, in one embodiment of this disclosure, the processors of multiple computing devices can be used together to process the training of the network function and data classification using the network function. Furthermore, the computer program executed on the computing device according to one embodiment of this disclosure may be a program that can be executed on the CPU, GPGPU, or TPU.

[0056] Furthermore, the processor 110 can typically handle the overall operation of the computing device 100. For example, the processor 110 can provide or process appropriate information or functions to the user by processing data, information, or signals that are input or output by the components included in the computing device 100, or by driving application programs stored in the storage unit.

[0057] According to one embodiment of the present disclosure, the memory 130 can store any form of information generated or determined by the processor 110 and any form of information received by the computing device 100. According to one embodiment of the present disclosure, the memory 130 may be a storage medium for storing computer software that causes the processor 110 to perform the operations according to the embodiments of the present disclosure. Thus, the memory 130 can mean a computer read medium for storing software code necessary to perform embodiments of the present disclosure, data on which the code is executed, and the results of the code execution.

[0058] According to one embodiment of this disclosure, memory 130 can mean any type of storage medium. For example, memory 130 may include at least one type of storage medium from among flash memory type, hard disk type, multimedia card micro type, card type memory (e.g., SD or XD memory), RAM (Random Access Memory), SRAM (Static Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), PROM (Programmable Read-Only Memory), magnetic memory, magnetic disk, and optical disk. The computing device 100 may also operate in relation to web storage that performs the storage function of memory 130 over the internet. The above descriptions of memory are illustrative, and memory 130 used in this disclosure is not limited to the above examples.

[0059] The communication unit (not shown) in this disclosure can be configured in any manner, such as wired or wireless, and can be composed of various communication networks such as Personal Area Networks (PANs) and Wide Area Networks (WANs). Furthermore, the communication unit can operate based on the well-known World Wide Web (WWW) and can also use wireless transmission technologies used for short-range communication, such as infrared (IrDA) or Bluetooth®.

[0060] The computing device 100 in this disclosure may include any form of user terminal and / or any form of server. Therefore, embodiments of this disclosure may be executed by a server and / or user terminal.

[0061] A user terminal can include any form of terminal capable of interacting with a server or other computing device. Examples of user terminals include mobile phones, smartphones, laptop computers, PDAs (personal digital assistants), slate PCs, tablet PCs, and ultrabooks. A server can include any type of computing system or computing device, such as a microprocessor, mainframe computer, digital processor, portable device, and device controller.

[0062] In further embodiments, the aforementioned server may also mean an entity that stores and manages drug administration information and associated responses, electronic medical records (EMRs), diagnostic test data, disease diagnosis data, nursing records, etc. The server can more efficiently collect medical data from multiple healthcare institutions via a database converted to a common data model standard, and may include a storage unit (not shown) for storing relevant information to enable the detection of the latest adverse drug reactions, utilizing a regularly updated database. For example, the storage unit may be contained within the server or under the server's management. As another example, the storage unit may reside outside the server and be embodied in a form that can communicate with the server. In this case, the storage unit may be managed and controlled by another external server different from the server.

[0063] Figure 2 shows an exemplary structure of an artificial intelligence-based model according to one embodiment of the present disclosure.

[0064] Throughout this specification, the terms predictive model, artificial intelligence-based predictive model, artificial intelligence model, artificial intelligence-based model, computational model, neural network, network function, and neural network may be used interchangeably.

[0065] A neural network can consist of a set of interconnected computational units, which can generally be called nodes. These nodes are sometimes also called neurons. A neural network consists of at least one node. The nodes (or neurons) that make up a neural network can be interconnected by one or more links.

[0066] Within a neural network, one or more nodes connected via links can form a relative input-output node relationship. The concepts of input and output nodes are relative; any node that is an output node to another node can be an input node to another node, and vice versa. As mentioned above, the input-output node relationship can be generated around links. One input node can be connected to one or more output nodes via links, and vice versa.

[0067] In a relationship between input and output nodes connected via a single link, the data of the output node may be determined based on the data input to the input node. Here, the link interconnecting the input and output nodes may have weights. The weights may be variable and can be changed by the user or algorithm to enable the neural network to perform desired functions. For example, if one or more input nodes are interconnected to a single output node by their respective links, the output node's value can be determined based on the values ​​input to the input nodes connected to the output node and the weights set for the links corresponding to each input node.

[0068] As mentioned earlier, a neural network consists of one or more nodes interconnected via one or more links, forming input-output node relationships within the network. The number of nodes and links within the neural network, the relationships between nodes and links, and the weights assigned to each link can determine the characteristics of the neural network. For example, if two neural networks exist with the same number of nodes and links but different link weights, the two neural networks may be perceived as different from each other.

[0069] A neural network can consist of a set of one or more nodes. A subset of the nodes that make up a neural network can form a layer. Some of the nodes that make up a neural network can form a layer based on their distance from the first input node. For example, a set of nodes that are n in distance from the first input node can form an n layer. The distance from the first input node can be defined by the minimum number of links that must be traversed to reach that node from the first input node. However, such a definition of a layer is arbitrary for illustrative purposes, and the difference in the number of layers within a neural network can be defined in a different way than described above. For example, a node layer can also be defined by its distance from the final output node.

[0070] In one embodiment of this disclosure, a collection of neurons or nodes can be defined as a layer.

[0071] The initial input node can refer to one or more nodes in a neural network that receive data directly without going through links in relation to other nodes. Alternatively, it can refer to a node in a neural network that does not have other input nodes connected by links in relation to other nodes based on links. Similarly, the final output node can refer to one or more nodes in a neural network that do not have other output nodes in relation to other nodes. Furthermore, hidden nodes can refer to nodes in a neural network that are neither the initial input node nor the final output node.

[0072] A neural network according to one embodiment of this disclosure may have the same number of nodes in the input layer as the number of nodes in the output layer, and may be a neural network in which the number of nodes decreases as the input layer progresses through a hidden layer, and then increases again. Another neural network according to another embodiment of this disclosure may have fewer nodes in the input layer than the number of nodes in the output layer, and may be a neural network in which the number of nodes decreases as the input layer progresses through a hidden layer. Yet another neural network according to yet another embodiment of this disclosure may have more nodes in the input layer than the number of nodes in the output layer, and may be a neural network in which the number of nodes increases as the input layer progresses through a hidden layer. Yet another neural network according to yet another embodiment of this disclosure may be a neural network in which the aforementioned neural networks are combined.

[0073] A deep neural network (DNN) can refer to a neural network that includes multiple hidden layers in addition to the input and output layers. Using deep neural networks, it is possible to grasp the latent structures of data. That is, it is possible to grasp the gene sequence structure, the amino acid sequence structure, the protein sequence structure, the causal relationship between drugs and ADRs and / or the probability of association between symptoms and ADRs, and the latent structures of photographs, text, videos, and audio (for example, what objects are in a photograph, what is the content and emotion of the text, what is the content and emotion of the audio, etc.). Deep neural networks can include convolutional neural networks (CNNs), recurrent neural networks (RNNs), autoencoders, generative adversarial networks (GANs), restricted Boltzmann machines (RBMs), deep belief networks (DBNs), Q networks, U networks, Siam networks, etc. The aforementioned description of deep neural networks is merely illustrative, and this disclosure is not limited thereto.

[0074] As an example, the artificial intelligence-based predictive models described herein may include Random Forest, Logistic Regression, Support Vector Machine, Decision Tree, Naive Bayes, k-Nearest Neighbor, XGBoost, LightGBM, CatBoost, Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM) network, Bidirectional Long Short-Term Memory (BiLSTM) network, Transformers, Generative Pre-trained Transformers (GPT), Bidirectional Encoder Representations from Transformers (BERT), SpanBERT, GRU (Gated Recurrent Unit), or BiGRU (Bidirectional Gated Recurrent Unit).

[0075] The artificial intelligence-based predictive models described herein can be represented by a network structure of any of the aforementioned structures, including an input layer, a hidden layer, and an output layer.

[0076] The neural networks used in the artificial intelligence-based models of this disclosure can be trained using at least one of the following methods: supervised learning, unsupervised learning, semi-supervised learning, or reinforcement learning, federated learning for distributed deep learning, or incremental learning. Training a neural network may be a process in which knowledge is applied to the neural network to enable it to perform a particular action. As an example, a predictive model according to one embodiment of this disclosure can be trained using a semi-supervised learning method that predicts masked symptoms (e.g., adverse drug reactions) after applying a mask to at least some of the symptoms or symptoms representing a patient's genetic genomic information, medications taken, and potential adverse drug events. In this case, at least a portion of the administered drugs and potential adverse drug reactions can be used as labeled data for training the predictive model, and the remaining portion can be used as unlabeled data for training the predictive model.

[0077] Neural networks can be trained to minimize output error. Neural network training is a process in which iterative training data is input into the neural network, the error between the neural network's output and the target for the training data is calculated, and the weights of each node in the neural network are updated by backpropagating the error from the output layer to the input layer in order to reduce the error. In supervised learning, training data with the correct answer labeled is used for each training data point (i.e., labeled training data), while in unsupervised learning, the training data may not have the correct answer labeled for each training data point. For example, in supervised learning for data classification, the training data may be data with a category labeled for each training data point. Labeled training data can be input into a neural network, and the error can be calculated by comparing the neural network's output (category) with the labels of the training data.

[0078] As another example, in unsupervised learning for data classification, the error can be calculated by comparing the input training data with the output of the neural network. The calculated error is backpropagated in the neural network (i.e., from the output layer to the input layer), and the connection weights of each node in each layer of the neural network may be updated according to the backpropagation. The amount of change in the connection weights of each node that are updated can be determined by the learning rate. The calculation of the neural network on the input data and the backpropagation of the error can constitute a learning cycle (epoch). The learning rate can be applied differently depending on the number of iterations of the neural network's learning cycle. For example, a high learning rate can be used in the early stages of learning to increase efficiency by allowing the neural network to quickly achieve a certain level of performance, while a low learning rate can be used in the later stages of learning to improve accuracy.

[0079] In neural network training, training data can generally be a subset of real-world data (i.e., data to be processed using the trained neural network). Therefore, while the error on the training data decreases, there can be training cycles where the error on real-world data increases. Overfitting is a phenomenon where the training data is excessively trained, leading to an increase in the error on real-world data. For example, a neural network trained to recognize a yellow cat as a cat may fail to recognize a cat of a different color as a cat; this is a type of overfitting. Overfitting can act as a cause of increased error in machine learning algorithms. Various optimization methods can be used to prevent such overfitting. To prevent overfitting, methods such as increasing the amount of training data, regularization, dropout (deactivating some of the network nodes during the training process), and the use of a batch normalization layer can be applied.

[0080] A computer-readable medium storing a data structure according to one embodiment of the present disclosure is disclosed. The aforementioned data structure may be stored in the storage unit of the present disclosure, executed by a processor, and transmitted and received by a communication unit.

[0081] A data structure can refer to the organization, management, and storage of data, enabling efficient access to and modification of that data. A data structure can also refer to the organization of data to solve a specific problem (e.g., data retrieval, data storage, data modification in the shortest possible time). A data structure can also be defined as the physical or logical relationships between data elements designed to support specific data processing functions. Logical relationships between data elements can include linking relationships between user-defined data elements. Physical relationships between data elements can include actual relationships between data elements physically stored in a computer-readable storage medium (e.g., a permanent storage device). Specifically, a data structure can include a collection of data, relationships between data, and functions or instructions that can be applied to data. A well-designed data structure allows computing devices to perform operations while minimizing the use of their resources. Specifically, computing devices can improve the efficiency of operations such as arithmetic, reading, inserting, deleting, comparing, exchanging, and retrieving through well-designed data structures.

[0082] Data structures can be classified into linear and nonlinear data structures depending on their form. A linear data structure may be one in which only one data item is linked after another. Linear data structures can include lists, stacks, queues, and deques. A list can represent a set of data items that have an internal order. Lists can include linked lists. A linked list can be a data structure in which data is linked in a linear fashion, with each item having a pointer. Pointers in a linked list can contain linking information to preceding and succeeding data. Linked lists can be represented as single linked lists, double linked lists, or circular linked lists, depending on their form. A stack may be a data list structure in which data can be accessed in a restricted manner. A stack can be a linear data structure in which data can only be processed (e.g., inserted or deleted) from one end of the data structure. Data stored in a stack can be a last-in, first-out (LIFO) data structure. A queue is a data structure that allows restricted access to data, and unlike a stack, it can be a first-in, first-out (FIFO) data structure. A deck can be a data structure that allows data to be processed from both ends of the data structure.

[0083] A nonlinear data structure can be a structure in which multiple data are concatenated after a single data. A nonlinear data structure can include a graph data structure. A graph data structure can be defined as vertices and edges, where an edge can contain a line connecting two distinct vertices, and can include a graph data structure tree. A tree data structure can be a data structure in which, among the multiple vertices contained in the tree, there is only one path connecting two distinct vertices. In other words, it can be a graph data structure that does not form a loop.

[0084] Throughout this specification, the terms predictive model, artificial intelligence-based model, computational model, neural network, network function, and neural network can be used interchangeably. Hereafter, the term neural network will be used consistently. A data structure may include a neural network, and such a data structure may be stored in a computer-readable medium. A data structure including a neural network may also include preprocessed data for processing by the neural network, data input to the neural network, neural network weights, neural network hyperparameters, data obtained from the neural network, activation functions associated with each node or layer of the neural network, loss functions for learning the neural network, etc. A data structure including a neural network may include any of the components in the disclosed configuration. In other words, a data structure containing a neural network may consist of all or any combination thereof of preprocessed data for processing by the neural network, data input to the neural network, neural network weights, neural network hyperparameters, data obtained from the neural network, activation functions associated with each node or layer of the neural network, loss functions for learning the neural network, etc. Beyond the configuration described above, a data structure containing a neural network may also include any other information that determines the properties of the neural network. Furthermore, the data structure may include all forms of data used or generated during the computational process of the neural network, and is not limited to those described above. Computer-readable media may include computer-readable recording media and / or computer-readable transmission media. A neural network may consist of a collection of interconnected computational units, which can generally be referred to as nodes.These nodes are sometimes referred to as neurons. A neural network consists of at least one or more nodes.

[0085] A data structure may include data that is input to a neural network. A data structure containing data that is input to a neural network may be stored in a computer-readable medium. Data that is input to a neural network may include training data that is input during the training process of the neural network and / or input data that is input to the neural network after training is complete. Data that is input to a neural network may include preprocessed data and / or data that is subject to preprocessing. Preprocessing may include data processing processes to prepare data for input into a neural network. Therefore, a data structure may include data that is subject to preprocessing and data that is generated by preprocessing. The data structures described above are merely examples, and this disclosure is not limited thereto.

[0086] The data structure may include the weights of a neural network (in this specification, weights and parameters may be used interchangeably). Furthermore, the data structure containing the weights of a neural network may be stored in a computer-readable medium. A neural network may contain multiple weights. The weights may be variable and can be varied by a user or algorithm to enable the neural network to perform a desired function. For example, if one or more input nodes are interconnected to a single output node by their respective links, the output node can determine the data value output from the output node based on the values ​​input to the input nodes connected to the output node and the weights set on the links corresponding to each input node. The data structures described above are merely examples, and this disclosure is not limited thereto.

[0087] As an unrestricted example, weights may include weights that change during the learning process of a neural network and / or weights that the neural network has completed learning with. Weights that change during the learning process of a neural network may include weights at the start of a learning cycle and / or weights that change during a learning cycle. Weights that the neural network has completed learning with may include weights that are at the end of a learning cycle. Thus, a data structure containing the weights of a neural network may include a data structure containing weights that change during the learning process of a neural network and / or weights that the neural network has completed learning with. Therefore, the aforementioned weights and / or each combination of weights shall be included in the data structure containing the weights of a neural network. The aforementioned data structures are illustrative and the disclosure is not limited thereto.

[0088] A data structure containing neural network weights may be stored in a computer-readable storage medium (e.g., memory, hard disk) after undergoing a serialization process. Serialization may be a process of storing the data structure in the same or other computing device and later reconstructing it into a usable form. The computing device can serialize the data structure and send and receive the data over a network. The serialized data structure containing neural network weights may be reconstructed in the same or other computing device by deserialization. The data structure containing neural network weights is not limited to serialization. Furthermore, the data structure containing neural network weights may include data structures that enhance computational efficiency while minimizing the use of computing device resources (e.g., B-Tree, R-Tree, Trie, m-way search tree, AVL tree, Red-Black Tree in nonlinear data structures). The foregoing is merely illustrative, and this disclosure is not limited thereto.

[0089] The data structure may include the hyperparameters of the neural network. Furthermore, the data structure containing the neural network's hyperparameters may be stored in a computer-readable medium. The hyperparameters may be variables that can be changed by the user. Examples of hyperparameters include the learning rate, cost function, number of learning cycle iterations, weight initialization (e.g., setting the range of weight values ​​to be initialized), and the number of Hidden Units (e.g., the number of hidden layers, the number of nodes in the hidden layers). The data structures described above are merely examples, and this disclosure is not limited thereto.

[0090] A transformer can be considered as a network function for a prediction model according to one embodiment of this disclosure. For example, the prediction model may operate as a transformer-based model. Such a prediction model may operate using, for example, a recurrent neural network to which an attention algorithm is applied or a transformer to which an attention algorithm is applied.

[0091] In one embodiment, the transformer may consist of an encoder that encodes the embedded data and a decoder that decodes the encoded data. The transformer may have a structure that receives a series of data, goes through the encoder and decoder stages, and outputs a series of data of different types. In one embodiment, the series of data may be processed into a form that the transformer can process. The process of processing the series of data into a form that the transformer can process may include an embedding process. Expressions such as data tokens, embedding vectors, embedding tokens, etc., may refer to data that has been embedded in a form that the transformer can process.

[0092] To encode and decode a series of data, a transformer can utilize an attention algorithm to process the encoder and decoder within the transformer. An attention algorithm can be defined as an algorithm that calculates an attention value by determining the similarity of a given query to one or more keys, reflecting this similarity to the value corresponding to each key, and then weighting the values ​​that reflect the similarity.

[0093] Attention algorithms can be classified into various types depending on how the query, key, and value are set. For example, if the query, key, and value are all set to be the same and attention is sought, this can be called a self-attention algorithm. If the dimension of the embedding vector is reduced to process a series of input data in parallel, and a separate attention head is found for each divided embedding vector to obtain attention, this can be called a multi-head attention algorithm.

[0094] In one embodiment, the transformer may consist of modules that perform multiple multi-head self-attention algorithms or multi-head encoder-decoder algorithms. In one embodiment, the transformer may also include additional components that are not attention algorithms, such as an embedding layer, a normalization layer, or a softmask (softmax) layer. Methods for configuring the transformer using attention algorithms may include the method disclosed in Vaswani et al., Attention Is All You Need, 2017 NIPS, which is incorporated herein by reference.

[0095] Transformers can be applied to diverse data domains such as embedded natural language, embedded sequence information, segmented image data, and audio waveforms, and can transform a series of input data into a series of output data. To transform data with diverse data domains into a series of data that can be input to the transformer, the transformer can embed the data. The transformer can process additional data that represents the relative positional or phase relationships between a series of input data. Alternatively, a series of input data may be embedded by further reflecting vectors that represent the relative positional or phase relationships between the input data. As an example, the relative positional relationships between a series of input data may include, but are not limited to, word order in a natural language sentence, the relative positional relationships of segmented images, or the time order of segmented audio waveforms. The process of adding information that represents the relative positional or phase relationships between a series of input data may be called positional encoding.

[0096] One example of a method for embedding data and transforming it using a transformer is disclosed in Dosovitskiy, et al., AN IMAGE IS WORTH 16X16 WORDS: TRANSFORMERS FOR IMAGE RECOGNITION AT SCALE, which is incorporated herein by reference.

[0097] The computing apparatus and computing devices described in this disclosure can be used interchangeably.

[0098] Figure 3 illustrates an exemplary method for generating output results of a model that includes prediction of adverse drug reactions according to one embodiment of the present disclosure.

[0099] Embodiments of this disclosure may include embodiments for training a predictive model and embodiments for actually obtaining inference results (e.g., predictions including possible adverse drug reactions and / or their probabilities) using the trained predictive model.

[0100] In this disclosure, the operation of the predictive model may be described as either a process of training the predictive model or a process of using the predictive model to make inferences about adverse drug reactions (e.g., a process of generating predictive results). This is done for the sake of explanation, and even if either the training process or the inference process is described as a process, it should be interpreted as intended to encompass the other of the training or inference processes.

[0101] In one embodiment, the steps shown in Figure 3 can be performed by a computing device 100. In a further embodiment, the steps in Figure 3 can also be embodied by multiple entities, such as a configuration in which some of the steps shown in Figure 3 are performed on a user terminal and others on a server.

[0102] In one embodiment of the present disclosure, the computing device 100 can construct a periodic conversion database that is scheduled to be updated in predetermined time intervals using raw data for drug adverse reaction detection (S310).

[0103] In one embodiment, raw data can refer to data acquired before it is compiled into a database. For example, raw data can be used to refer to data acquired from various channels, such as electronic medical records.

[0104] In one embodiment, raw data can also be obtained from at least one of FAERS, KAERS, WHO-ART, SIDER, or EU-ADR.

[0105] In one embodiment, raw data can be obtained from an electronic medical record (EMR). The raw data may include at least one of the following: drug use-related data, disease diagnosis data, diagnostic test data, or nursing record data. For example, diagnostic test data may mean data obtained through any form of test or diagnosis. As an example, diagnostic test data may include genetic test records. Specific examples of diagnostic test data, drug use-related data, disease diagnosis data, and nursing record data will be described later.

[0106] In one embodiment, the computing device 100 may be provided with a list of predictable adverse drug reactions (ADRs) from an external rule-based or artificial intelligence-based predictive model different from the predictive model described herein.

[0107] In one embodiment, the provided foreseeable adverse drug reactions (ADRs) may be sorted based on the raw data according to specific criteria (e.g., causality and / or severity of reaction such as Mild, Moderate, Severe), and the computing device 100 may process the provided ADRs into output data that further includes the "probability of occurrence" of the ADRs, etc.

[0108] In one embodiment, the computing device 100 can convert terms contained in raw data to correspond to common data standards in a database or predictive model. For example, dystonia, athetosis, choreoathetosis, and chorea can be defined as similar or identical ADRs in the same category, and the conversion can be provided with a clear distinction from non-drug-related causes that distinguish these abnormal movements suspected to be drug side effects. For example, the computing device 100 can classify cerebral palsy, focal intracranial diseases, Huntington's disease, hepatic encephalopathy, hepatic cirrhosis, etc., which are different from the aforementioned amorphous movements, as diseases or ADRs distinct from dystonia, athetosis, choreoathetosis, and chorea. As a further example, if the cause of the abnormal movement in the nursing record data is not specifically described, the computing device 100 may not use the data as drug adverse reaction prediction learning data, or it may classify and convert the term as an ADR with low causality with the drug.

[0109] In one embodiment of the present disclosure, a periodic conversion database can mean a data storage configured to store and update patient demographic information, the drugs they are taking and their indications, adverse drug events (ADRs) that have occurred, and their medical history at predetermined or optimal time intervals determined by a predictive model. A periodic conversion database can also mean a data storage constructed by collecting all data on newly entered or additionally entered ADRs and patient information. For example, a user of the periodic conversion database can search for the type of drug that causes a particular ADR—e.g., seizure—and if they search for the drug bupropion that a patient is taking, they may also be provided with other possible ADRs or the dosage in clinical cases where patients exhibit seizure symptoms after medication, as well as patient information, including a statistical summary. As an example, an artificial intelligence-based model according to one embodiment of the present disclosure may include a model for determining predetermined time periods to be converted or updated in the periodic conversion database. An artificial intelligence-based model that inputs the amount of data stored in the database, the type of data, and / or the content of the data can be operated to output periodic conversion points in the periodic conversion database.

[0110] In one embodiment, the computing device 100 can construct a periodic conversion database using at least one of the following methods: a database linkage method, a file linkage method, or a change history classification linkage method.

[0111] In one embodiment, the database linking method can mean, for example, a data automation method that accurately or efficiently links information such as the age at onset of a disease, the number of months it lasts, the age at which symptoms improve, and the age at which it is cured, based on information about the patient's age, the time of onset of the disease, and the duration of the disease at the time the data was entered or acquired.

[0112] In one embodiment, the file linking method can mean, for example, a method that links electronic medical record (EMR) documents in a file format that is most suitable for driving a prediction model and simultaneously driving an external model different from the prediction model, without losing the information contained therein, even if the extensions of the electronic medical record (EMR) documents are different.

[0113] In one embodiment, the change history classification linking method can mean a method for organically distinguishing and linking change histories related to ADR based on information such as newly occurring symptoms, symptoms that have disappeared, the severity of symptoms, drugs added or removed, smoking status, alcohol consumption status, and newly discovered medical history for the same patient. In one embodiment of the present disclosure, the update cycle of the periodic conversion database, which is updated using at least one of the following methods: a database linkage method, a file linkage method, or a change history classification linkage method, can be a predetermined time period. For example, the update cycle of the periodic conversion database can be changed from once a month to once a week as needed by the user, or the update cycle can be shortened or extended depending on whether resources are sufficient or insufficient.

[0114] In one embodiment, the computing device 100 can obtain one or more of the following from public databases (e.g., FAERS, KAERS, WHO-ART, SIDER, EU-ADR, etc.): demographic data, drug use-related data, disease diagnosis data, diagnostic test data, and nursing record data. The public database is not limited to the disclosed databases.

[0115] In further embodiments, the diagnostic test-related data may include, for example, gene sequences obtained from patient-derived subjects and experimentally sequenced.

[0116] In one embodiment, the computing device 100 can use any form of big data analysis technique to construct the periodic conversion database. The big data analysis technique may include identifying patterns and relationships hidden in large amounts of data and extracting or generating predictive information. For example, the big data analysis technique may include text mining to extract specific information present in unstructured documents, opinion mining to crawl information related to a specific subject through websites and the like to extract opinions on that subject, and / or web mining to extract desired information from weblogs and / or search terms.

[0117] In one embodiment of the present disclosure, the computing device 100 can determine a plurality of input variables for training an artificial intelligence-based model for drug adverse reaction detection based on data stored in a periodic conversion database (S320).

[0118] In one embodiment, the input variables can mean variables or features used in training or inference for an artificial intelligence model. For example, the input variables may include at least one of the following: demographic variables, drug use-related variables, diagnostic test-related variables, nursing record-related variables, or drug side effect diagnosis-related variables.

[0119] In one embodiment, a demographic variable is a variable used to quantitatively or qualitatively represent population-related developments. The demographic variable may include, for example, at least one of age or region.

[0120] In one embodiment, the drug use variable can represent a quantitative or qualitative indicator that can be defined in relation to the user's drug use. For example, the drug use-related variable may include at least one of the following: type of drug, duration of drug use, dosage of drug use, or drug use history.

[0121] In one embodiment, diagnostic test-related variables can represent quantitative or qualitative indicators that can be derived from various forms of diagnostic or test results. For example, diagnostic test-related variables may include at least one of imaging tests, blood tests, or urine tests.

[0122] In one embodiment, nursing record-related variables may refer to quantitative or qualitative variables obtained from nursing record sheets generated during the treatment and / or nursing process. For example, nursing record-related variables may be generated based on information contained in medical or nursing record charts within a hospital. For example, nursing record-related variables can be obtained through at least one of the following methods: optical character recognition (OCR), natural language processing (NLP), or text mining. In a further embodiment, the drug side effect diagnosis-related variables may include information related to disease diagnosis due to drug side effects.

[0123] In one embodiment, imaging data that may be included in diagnostic test-related variables can be obtained from ultrasound imaging, X-ray (simple radiography), CT (computerized tomography), MRI (magnetic resonance imaging), fMRI (functional Magnetic Resonance Imaging), PET (positron emission tomography), etc.

[0124] In one embodiment, information related to disease diagnosis due to drug side effects, which may be included in drug side effect diagnosis-related variables, may mean data collected when symptoms arising from unintended or unexpected drug administration are diagnosed as a disease. In a further embodiment, the predictive model according to this disclosure may suggest coadministration to prevent unintended symptoms that may result from administered drugs.

[0125] A more detailed explanation of the input variable selection method will be provided later in Figure 6.

[0126] In one embodiment, parameter information related to the preprocessing method can be obtained from data output from an external, separate prediction model that is different from the prediction model described in this disclosure.

[0127] In further examples, the preprocessing for training or inference of the predictive model described herein (e.g., preprocessing of blood test data variables) can be performed using parameters generated by an external predictive model operating on patient demographic data and disease information as input data. For example, if a 36-year-old Caucasian woman is a triple-negative (HER-2, ER, PR-negative) breast cancer (TNBC) patient, the predictive accuracy for the probability of developing certain ADRs (e.g., mucosal inflammation, erythrodysesthesia, elevated serum ALT (Alanine aminotransferase) levels) can be improved based on parameter information generated by the external predictive model.

[0128] In a further embodiment, the ADR prediction model can be driven by updating in real time not only databases of administered drugs and specific indications for specific individuals, and databases of adverse drug reactions that have occurred (e.g., FAERS, KAERS, WHO-ART, SIDER, EU-ADR), but also reports of new research papers. In a further embodiment, the ADR prediction model can be driven by updating in real time clinical response and progress data entered into international clinical trial databases (e.g., ClinicalTrials.gov), and can also be driven by updating in real time results reports of research literature on genes and drug responses in human or non-human individuals.

[0129] In one embodiment, the data that forms the driving basis for an artificial intelligence or rule-based external prediction model that provides parameter information to be received as input by the ADR prediction model (e.g., combined data such as demographic information, genetic information, administered drugs, and known / unknown side effects) is illustrative and not limited to the foregoing disclosure.

[0130] In one embodiment of the present disclosure, the computing device 100 can perform preprocessing by applying a mask to at least one of the following for training and / or inference of a predictive model: data relating to adverse drug reactions (ADRs), data relating to the duration of medication, and some demographic information (e.g., sex, age of onset, race).

[0131] In one embodiment, the computing device 100 can determine the categories of regions to be masked in the training dataset. For example, the regions to be masked may include data from one or more categories. The computing device 100 can cause the predictive model to perform masking-based training based on the determined categories to be masked. In a further embodiment, such masking-based preprocessing and training of the predictive model can contribute to more accurately outputting predicted values ​​of multiple adverse drug reactions (ADRs) in descending order of probability, for example, when two patients with similar medical histories differ only in gender, or when they are the same gender but different in age.

[0132] In one embodiment, the computing device 100 can determine input variables for a prediction model from among multiple variables based on a first method for determining variable importance, by using perturbation input data in which at least some of the variables have been changed and output data output from the prediction model in response to the perturbation input data. For example, the computing device 100 can generate perturbation input data in which at least some of the multiple variables included in the input data have been changed. The computing device 100 can determine which variables to select from among the multiple variables by inputting multiple input data, including input data in which the variables have not been changed and perturbation input data in which the variables have been changed, into an artificial intelligence-based model and comparing the outputs of the model. For example, the computing device 100 can determine the importance of the changed variable by comparing the outputs (first output and second output) of the model for first input data in which the first variable has not been changed and first perturbation input data in which the first variable has been changed. Based on such importance, the input variables used in the artificial intelligence-based prediction model can be determined. In one embodiment, the computing device 100 can also determine input variables by inputting multiple input data, each of which includes input data where the variables have not been changed and perturbation input data where the variables have been changed, to multiple artificial intelligence-based models.

[0133] In one embodiment, the computing device 100 can determine input variables for a prediction model from among multiple variables based on a second method for selecting input variables corresponding to principal components by converting data in a high-dimensional space into a low-dimensional space.

[0134] In one embodiment, the computing device 100 can determine input variables for a prediction model from among multiple variables based on a third method of selecting input variables in a manner that reduces impurity on a decision tree.

[0135] In one embodiment, the computing device 100 can determine input variables based on Boruta SHAP, Principal Component Analysis (PCA), Feature importance, Permutation importance, and / or Gini importance. Such determination of input variables can be included in the preprocessing steps.

[0136] In one embodiment, the preprocessing process may be performed based on at least one of the following: outlier handling techniques, missing value substitution techniques, categorization or recategorization techniques, and normalization or standardization techniques. Through these processes, the input variables can be determined and / or preprocessed. The preprocessed input variables can then be input into a predictive model for training and / or inference.

[0137] In one embodiment of the present disclosure, the computing device 100 can train an artificial intelligence-based predictive model using a training dataset that includes determined input variables and adverse reaction information for specific drugs and specific diseases, so that the predictive model outputs adverse reaction information for specific drugs and specific diseases in response to input variables (S330).

[0138] In one embodiment, the method for training the predictive model for detecting adverse drug reactions may include a step of determining the predictive model for detecting adverse drug reactions from among a plurality of candidate models, and the predictive model may be trained to output whether or not there is an adverse drug reaction to a particular drug or a particular disease, or the probability of an adverse drug reaction occurring to a particular drug or a particular disease, using the preprocessed input variables as input data.

[0139] In one embodiment, the plurality of candidate models may be one or more models selected from among candidate models consisting of Random Forest, Logistic Regression, Support Vector Machine, Decision Tree, Naive Bayes, k-Nearest Neighbor, XGBoost, LightGBM, CatBoost, Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Bidirectional Long Short-Term Memory (BiLSTM), Transformer, and Generative Pre-trained Transformer (GPT).

[0140] In one embodiment, the prediction model can obtain output results from each of the candidate models using a predetermined test dataset, and determine one of the candidate models as the prediction model based on a performance evaluation of the output results from each of the candidate models.

[0141] In a further embodiment, the performance evaluation of the output results of the candidate model may be based on AUC (Area Under the Curve), accuracy, F1-score, precision, recall, sensitivity, and / or specificity.

[0142] In one embodiment, the input variables used for training and / or inference of the predictive model can be preprocessed by performing outlier processing, inputting missing values, categorization / recategorization and / or normalization / standardization.

[0143] In further examples, large volumes of data, including medications taken, possible adverse drug reactions, and patient demographic information, can be retrieved and updated in real time from public data databases (e.g., FAERS, KAERS, WHO-ART, SIDER, EU-ADR).

[0144] In a further embodiment, the predictive model according to this disclosure can output significant predictive results even for ADRs with a low probability of occurrence, considering the underlying health issues of the target patient and whether the patient is an infant or child, despite the enormous amount of data accumulated for administered drugs and indications.

[0145] In one embodiment, the predictive model can improve its learning and / or inference efficiency and / or accuracy based on the drug adverse reaction prediction results of the predictive model, which are referenced at each of the multiple epochs. Specifically, such improvements in efficiency and / or accuracy are achieved by receiving feedback through supervised learning, semi-supervised learning, and / or unsupervised learning on ground truth data for drug adverse reactions with a high causal relationship confirmed clinically.

[0146] As a specific example, in a 47-year-old male with ulcerative colitis diagnosed within six months, if the predictive model based on this disclosure outputs fatigue at 34.4%, joint pain at 30.2%, headache at 28.9%, and bronchitis at 19.3%, then a healthcare professional can decide to prescribe 5-ASA and Vedolizumab to treat the patient's ulcerative colitis, and can also suggest combination drugs to prevent these ADRs based on the predicted probability of occurrence and potential fatalities.

[0147] In a further embodiment, data on adverse drug reactions that occurred or did not occur based on monitoring of the treatment course can be used to update the loss function or hyperparameters of the prediction model that outputs the ADR prediction results. For example, in the case of the 47-year-old male patient with ulcerative colitis, if the bronchitis presented as the actual prediction result paper was moderate and occurred with certain causality, it can be made a mandatory process to feed this data back into the prediction model.

[0148] Specifically, in one embodiment, the computing device 100 can predict, based on information contained in the periodic conversion database, the possible ADRs that may occur in patient X with cardiovascular disease due to drug Z administered for indication Y are: i. inflammatory bowel disease (IBD) with a probability of occurrence of 3.8% as a moderate ADR, ii. diarrhea with a probability of occurrence of 13.2% as a mild ADR, and iii. dyspepsia with a probability of occurrence of 10.9% as a minor ADR.

[0149] Further explanations regarding the training and driving of the predictive model will be provided later in Figures 4 to 7.

[0150] Figure 4 shows a flowchart illustrating the construction of the periodic conversion database 420, including the periodic data loading and change history table loading principles of the electronic medical record 410 and the periodic conversion database 420, according to one embodiment of the present disclosure.

[0151] In one embodiment, Figure 4 illustrates a feature in which the computing device 100 periodically loads data and change history from the OCS (order communication system) and / or EMR (electronic medical record) 411 into the periodic conversion database 420. In one embodiment, Figure 4 shows that the computing device 100 can store data using the ODS (Operational Data Store) 421, ITF (Interface) 422, and CDM (Common Data Model) 423.

[0152] In one embodiment, ODS421 can mean a central database that provides a snapshot of the latest data from multiple transactional systems for operational reporting purposes, as operational data storage. In one embodiment, ODS421 can collect or receive information from newly added or modified OCS and / or EMR411 at predetermined time intervals. In one embodiment, ODS421 can be used as a database designed to integrate data from multiple sources.

[0153] Upon receiving newly added or modified OCS and / or EMR411 information, ODS421 can transmit this information via the ITF422 schema to be stored as a CDM423 schema. The Common Data Model (CDM)423 can be defined as a data model with a single structure and standards applicable to medical data with different structures held by various healthcare institutions. The Common Data Model (CDM)423 allows for the execution of the same analysis code individually at each data-holding institution, thus enabling distributed collaborative research that integrates distributed data.

[0154] In one embodiment of the present disclosure, the periodic conversion database 420 can store information such as patient demographic information, drugs administered to the patient and their indications, adverse drug events (ADRs) that have occurred, and medical history at predetermined or optimal time intervals determined by the prediction model. The periodic conversion database 420 can mean a data storage that collects and constructs data for newly entered or additionally entered ADRs and patient information.

[0155] Figure 5 illustrates the concept of file linkage for constructing a periodic conversion database according to one embodiment of the present disclosure.

[0156] In one embodiment, the computing device 100 can construct a periodic conversion database 420 using at least one of the following methods: a database linkage method 510, a file linkage method 520, and / or a change history classification linkage method 530.

[0157] In one embodiment, in relation to database linkage 510, the computing device 100 periodically loads and receives data from interface tables 511 containing data from the medical institution's electronic medical record (EMR). The loaded data can be transmitted to the ODS DB interface table 513 via an ODS DB loader (operational data store database loader) 512. The computing device 100 can link the loaded data to a CDM table on the CDM database by applying a Common Data Model (CDM) conversion 514 to the ODS DB interface table 513. Linkage between the EMR-related database and the Common Data Model database can be achieved using the method described above.

[0158] The database linkage method 510 can be used in the same or similar way as the file linkage method 520 and / or the change history category linkage method 530, which will be described later.

[0159] In one embodiment, the file linkage 520 method can mean an automated data management method that can more accurately and / or more efficiently link information such as the age of onset of a disease, the number of months it lasts, the age at which symptoms improve, and the age at which it is cured, based on information such as the patient's age at the time of data entry, the time of disease onset, and the duration of the disease. Specifically, in relation to file linkage 520, the computing device 100 can upload data to the ODS DB through periodic loading of data on the medical institution's EMR, CSV (comma-separated values) file, or Parquet file 521. Data loaded in the ODS DB interface table 523 can be uploaded through the ODS DB loader 522. In one embodiment, files related to EMR loaded into the ODS DB interface table 523 can be stored in the form of a CDM table 525 in the CDM database after CDM conversion 524.

[0160] In one embodiment, the file linking method 520 can mean, for example, a method that links electronic medical record (EMR) documents 521 in a file format that is most suitable for driving the prediction model and simultaneously driving an external model different from the prediction model, without losing the information contained therein, even if the extensions of the electronic medical record (EMR) documents 521 are different from each other.

[0161] In one embodiment, the change history classification linkage 530 method can mean a method for organically classifying and linking change histories related to ADR based on information such as newly occurring symptoms, disappeared symptoms, the severity of symptoms, added or removed drugs, smoking status, alcohol consumption status, and newly discovered medical history for the same patient. In one embodiment, in relation to the change history classification linkage 530, the computing device 100 can periodically load data from interface tables 531 and change history tables 531 on the electronic medical record (EMR) of a medical institution. The computing device 100 can upload to a temporary logical area DB of the data warehouse via the ODS DB loader 532. The computing device 100 can load the data as interface tables and change history tables 533 of the ODS DB via the ODS DB loader 532. The loaded data can be stored in the CDM database 535 after CDM conversion 534. In one embodiment, the change classification linkage 530 method uses a change history table that stores the changed data, targets the changed data, and links it, thereby enabling efficient use of computing resources in data linkage.

[0162] In one embodiment of the present disclosure, the update cycle of the periodic conversion database 420, which is updated using at least one of the database linkage method 510, the file linkage method 520, or the change history classification linkage method 530, can correspond to a predetermined time period. For example, the update cycle of the periodic conversion database 420 can be changed from once a month to once a week and / or daily, depending on the user's needs. For example, the update cycle of the periodic conversion database 420 can be determined based on the quantitative capacity of the computing resources of the computing device 100. For example, depending on whether the computing resources are sufficient or insufficient, the computing device 100 can shorten or extend the update cycle.

[0163] In one embodiment, the computing device 100 can obtain one or more of the following from public databases (e.g., FAERS, KAERS, WHO-ART, SIDER, EU-ADR, etc.): demographic data, drug use-related data, disease diagnosis data, diagnostic test data, and nursing record data. The public databases are not limited to those disclosed herein.

[0164] Figure 6 illustrates the overall process, from database construction and development of artificial intelligence for drug adverse reaction detection to notification of anomaly detection results, according to one embodiment of the present disclosure.

[0165] An "adverse drug reaction (ADR)" refers to a harmful and unintended reaction that occurs when a drug or other substance is administered or used, and where a causal relationship with the drug or substance cannot be ruled out. This includes not only adverse reactions that occur at normal doses, but also adverse drug reactions and withdrawal symptoms that occur when a drug is used in excessive amounts intentionally or negligently, or when a drug is abused, as well as cases where the expected pharmacological effect is not observed.

[0166] In contrast, "drug adverse reaction monitoring" involves the rapid and systematic collection and evaluation of various adverse events that occur during drug use, the implementation of countermeasures, and the provision of safety information and the results of those measures to healthcare professionals, consumers, etc. It refers to activities aimed at establishing rational drug use and preventing drug-related harm from occurring in advance.

[0167] Concepts related to ADR include side effects and adverse drug events (ADEs). Side effects are the opposite of the principal action, which is the primary action of a drug used for a specific purpose. They refer to all unintended effects that occur when a drug is administered according to the usual dosage. Because it is a term that encompasses all effects other than therapeutic effects, it is a concept that can be actively used in a variety of fields that comprehensively study all effects, regardless of whether they are harmful or not.

[0168] Drug-induced adverse events (ADEs) refer to undesirable and unintended signs, symptoms, or illnesses that occur during the administration or use of a drug, regardless of whether they are causally related to the drug.

[0169] Adverse drug reactions (ADRs) are harmful and unintended reactions that occur during the normal administration or use of pharmaceuticals, etc., when a causal relationship with the pharmaceutical product cannot be ruled out, and when they are predictable and preventable. Therefore, management by pharmacists and medical teams is considered crucial.

[0170] Drug metabolism mostly takes place in the liver, where metabolic processes occur through oxidation, reduction, or hydrolysis reactions of drugs by the hepatic cytochrome P450 (CYP450) metabolic enzyme system in the endoplasmic reticulum of liver cells.

[0171] Embodiments in this disclosure may include steps for deriving predictable ADRs in order of causality (certain / probable / possible / unlikely / conditional) based on the type of administered drug, race / individual genotype / phenotype, medical history, etc., or classification by severity of reaction (mild / moderate / severe), and then by frequency of occurrence.

[0172] In one embodiment, the computing device 100 can implement an active drug adverse reaction detection system through the processes of constructing a periodic conversion database 610, developing an artificial intelligence for drug adverse reaction detection 620, and notifying abnormality detection results 630.

[0173] In one embodiment, the process for constructing the periodic conversion database 610 may include the unification of terminology and / or term mapping 616 for raw data obtained from the electronic medical record 612 and the adverse drug reaction standards 614. Term mapping 616 can mean the process of unifying the data format (e.g., format) for raw data that has diverse forms. Term mapping 616 can also mean the process of unifying the data representation for raw data that has diverse expressions but the same or similar meanings.

[0174] For example, the term mapping process 616 may operate based on an artificial intelligence-based language model that determines whether multiple input data are synonymous or related to each other. In such an example, the artificial intelligence-based language model can extract features from two or more input data, compare these features in a vector space, and then determine, based on the comparison, whether the two or more input data are synonymous or related to each other. Terms that are synonymous or related in this way can then be mapped and managed to each other.

[0175] In one embodiment, the periodic conversion database 618 can be constructed, for example, according to the OMOP-CDM standard. In one embodiment, the periodic conversion database 618 can be pre-configured to be updated at predetermined time intervals (e.g., daily, weekly, monthly, and / or yearly).

[0176] In one embodiment, the artificial intelligence development 620 for detecting adverse drug reactions may include a process for modeling and / or validating a predictive model.

[0177] In one embodiment, the computing device 100 can perform the processes of data preprocessing 622, artificial intelligence model training 624, artificial intelligence model verification 626, and artificial intelligence model selection 628.

[0178] In one embodiment, data preprocessing 622 may include a process of determining input variables to be used to train an artificial intelligence model from among a number of variables whose terminology is standardized on a periodic transformation database. As illustrated in Figure 6, data preprocessing 622 may include feature selection, outlier detection, missing value substitution, categorization or recategorization, and normalization or standardization.

[0179] In one embodiment, the process related to artificial intelligence model learning 624 may include a process for selecting a specific candidate algorithm from among many candidate algorithms, a process for selecting a specific learning method from among many learning methods, and a process related to tuning or optimizing hyperparameters. For example, the learning method may include Federated learning and / or Incremental learning.

[0180] In one embodiment, the process associated with artificial intelligence model validation 626 may include a validation methodology using an N:1 test dataset and / or a validation methodology through evaluation from an external server. In one embodiment, artificial intelligence model validation 626 may also be performed using metrics related to the performance of the artificial intelligence model, such as AUC, accuracy, and F-1 score.

[0181] In one embodiment, the computing device 100 can select a specific artificial intelligence model from among multiple artificial intelligence models (e.g., algorithms) through artificial intelligence model verification 626.

[0182] In one embodiment, the predictive model of the computing device 100 can process data input from the input layer to the hidden layer. The results of residual connections between multiple hidden layers can be transmitted to the output layer.

[0183] In one embodiment of the present disclosure, the predictive model may include an encoder and a decoder. For example, the encoder and decoder can be pre-trained such that a first input dataset containing input variables or combinations of input variables, such as demographic data, is input to the encoder, and the output of the decoder, which receives the output of the encoder as input, corresponds to the first input dataset. Through such pre-training, the encoder may be configured to accurately or efficiently generate features or vectors for the input variables.

[0184] In further embodiments of the present disclosure, the predictive model may include an encoder and a decoder. For example, the predictive model may be pre-trained such that a first input dataset containing input variables or combinations of input variables, such as demographic data, is input to the encoder, a second input dataset containing data on administered drugs and adverse drug reactions is input to the decoder along with the output of the encoder, and the decoder outputs a predictive result containing administered drug information and possible clinicopathological reactions or probabilities of occurrence related to ADRs.

[0185] In one embodiment, the predictive model can categorize each piece of information in the form of embedding, and perform the necessary tasks for generating output results according to the assigned categories. For example, categorization of family history information, medical history information, and genetic information may be performed in a separate embedding layer within the predictive model to reflect the probability of drug adverse reactions.

[0186] The effects of the embodiments described herein are that, by efficiently and accurately predicting adverse reactions to specific drugs, it is possible to support patients and medical professionals in designing countermeasures in advance to prevent the occurrence of adverse drug reactions (ADRs) or minimize their impact on life development, for example, in patients with diverse information. Furthermore, the computing device 100 may have a structure similar to a specific deep learning network that can be arbitrarily conceived. However, the predictive model in this disclosure is not limited to artificial intelligence models, but may also be a rule-based driven model.

[0187] In one embodiment, in an embodiment relating to the process of anomaly detection result notification 630, the computing device 100 can transmit the output results of a prediction model for adverse drug reactions to a user terminal (transceiver). For example, the output results of the prediction model transmitted to the user terminal may include at least one of the following output results: a first result showing a binary classification of adverse drug reactions, distinguished as detected or undetected; a second result showing a probability notation for adverse drug reactions; or a third result showing a three-stage category notation relating drug, adverse reaction, and risk. For example, the three-stage category notation linking "drug-ADR-risk" output by the prediction model can output a three-stage category notation as a severe abnormal case of overdose of 10 mg or more of wellbutrin (bupropion) - seizure - 71.0%.

[0188] In one embodiment, the output results from the predictive model can be used to retrain the predictive model based on the accuracy evaluated for those output results. As a specific example, for a 29-year-old male patient with ankylosing myelitis who was administered 40 mg of adalimumab once weekly, the output result provided is a 35.2% probability of developing benign, malignant, and unidentified neoplasms, including polyps, as an ADR. If such neoplasms did not actually develop in the patient, the predictive model according to this disclosure may be provided with such feedback data, namely, that unidentified neoplasms did not develop in a 29-year-old male patient with ankylosing myelitis with a specific medical history, as retraining data.

[0189] In one embodiment, feedback information corresponding to the output results transmitted from a user terminal that has received the output results from the prediction model can be used to retrain the prediction model. As a specific example, if follow-up data is collected for a patient for whom accurate demographic information and medical history were unavailable, the follow-up data can be provided to the user terminal that has received the ADR prediction output results from the prediction model, and this feedback information can be used to obtain more accurate prediction results from the prediction model or to retrain the prediction model.

[0190] When managing adverse drug reactions relies on reporting channels such as drug adverse reaction reporting systems, it requires voluntary reporting by patients and healthcare institutions. This means that if the reporter is unaware of the reaction, it may not be reported, and if timely measures are not taken against the adverse drug reaction, it could cause fatal additional harm to the patient.

[0191] An active drug adverse reaction detection system according to one embodiment of the disclosed information is a system that can detect drug adverse reactions in a timely manner, not only preventing harm that could be suffered due to the failure to recognize drug adverse reactions, but also enabling simple and systematic management of drug adverse reactions. While this system is particularly efficient in detecting adverse reactions to specific drugs in hospitalized patients, it can be used universally regardless of whether the patient is hospitalized or outpatient, or whether the drug is specific. Therefore, this system can be useful for institutions lacking personnel to manage drug adverse reactions, as well as for individual patients taking medication.

[0192] Pharmaceutical companies conduct post-marketing surveillance (PMS) for long-term safety monitoring of drugs. However, this requires significant time, personnel, and expense, and has major limitations such as the difficulty of recruiting research participants and collecting data, thus necessitating the active cooperation of healthcare professionals and patients. One embodiment of the method described in this disclosure can reduce the time, personnel, and costs associated with PMS, and can also contribute to building a virtuous cycle in the field of drug safety.

[0193] Figure 7 is a schematic diagram of a computing environment according to one embodiment of the present disclosure.

[0194] In this disclosure, components, modules, or units include routines, procedures, programs, components, data structures, etc., that perform a specific task or embody a specific abstract data type. Furthermore, a person of ordinary skill would fully recognize that the methods presented in this disclosure can be implemented in other computer system configurations, including single-processor or multi-processor computing devices, minicomputers, mainframe computers, as well as personal computers, handheld computing devices, microprocessor-based or programmable consumer electronics, and others (each of which may operate in connection with one or more related devices).

[0195] The embodiments described herein can also be implemented in a distributed computing environment in which a task is performed by remote processing units connected via a communication network. In a distributed computing environment, program modules may reside in both local and remote memory storage devices.

[0196] Computing devices typically include various computer-readable media. Any media accessible by a computer can be computer-readable, and such computer-readable media include volatile and non-volatile media, transient and non-transitory media, and portable and non-portable media. In non-restrictive examples, computer-readable media may include computer-readable storage media and computer-readable transmission media.

[0197] Computer-readable storage media include volatile and non-volatile media, transient and non-transitory media, portable and non-portable media, embodied by any method or technique for storing information such as computer-readable instructions, data structures, program modules, and other data.

[0198] Computer-readable storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory and other memory technologies, CD-ROM, DVD (digital video disk) and other optical disc storage devices, magnetic cassettes, magnetic tapes, magnetic disk storage devices or other magnetic storage devices, or any other media that can be accessed by a computer and used to store desired information.

[0199] Computer-readable transmission media typically include all information transmission media that embody computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism.

[0200] The term "modulated data signal" means a signal in which one or more of its characteristics are set or modified in order to encode information within the signal. Non-restrictive examples include computer-readable transmission media such as wired media like wired networks or direct-wired connections, and wireless media such as acoustic, RF, infrared, and other wireless media.

[0201] Any combination of the aforementioned media shall also be included in the scope of computer-readable transmission media.

[0202] An exemplary environment 2000 embodying various aspects of the present invention, including a computer 2002, is shown, the computer 2002 including a processing unit 2004, system memory 2006, and a system bus 2008. The computer 200 in this specification can be used interchangeably with computing devices. The system bus 2008 connects system components, including (but not limited to) system memory 2006, to the processing unit 2004. The processing unit 2004 may be any processor from a variety of commercially available processors. Dual-processor and other multi-processor architectures can also be used as the processing unit 2004.

[0203] The system bus 2008 may be any of the various types of interconnectable bus structures, in addition to a local bus using any of the memory bus, peripheral bus, and various commercially available bus architectures. The system memory 2006 includes read-only memory (ROM) 2010 and random access memory (RAM) 2012. The basic input / output system (BIOS) is stored in non-volatile memory 2010 such as ROM, EPROM, or EEPROM, and this BIOS includes basic routines that support the transmission of information between components within the computer 2002, such as during startup. RAM 2012 may further include high-speed RAM such as static RAM for casing data.

[0204] Computer 2002 also includes internal hard disk drives (HDDs) 2014 (e.g., EIDE, SATA), magnetic floppy disk drives (FDDs) 2016 (e.g., for reading from or writing to portable diskettes 2018), SSDs, and optical disk drives 2020 (e.g., for reading CD-ROM disks (2022), or for reading from or writing to other high-capacity optical media such as DVDs). The hard disk drives 2014, magnetic disk drives 2016, and optical disk drives 2020 may be connected to the system bus 2008 by hard disk drive interfaces 2024, magnetic disk drive interfaces 2026, and optical drive interfaces 2028, respectively. Interfaces 2024 for external drive implementation include, for example, at least one or both of the USB (Universal Serial Bus) and IEEE 1394 interface technologies.

[0205] These drives and associated computer-readable media provide data, data structures, computer-executable instructions, and other non-volatile storage. In the case of Computer 2002, the drives and media correspond to storing any data in a suitable digital format. While the above description of computer-readable storage media includes HDDs, portable magnetic disks, and portable optical media such as CDs or DVDs, a person of ordinary knowledge will understand that drives (zip drives), magnetic cassettes, flash memory cards, cartridges, and other types of computer-readable storage media can also be used in the exemplary operating environment, and that any such media may contain computer-executable instructions for carrying out the methods of the present invention.

[0206] Numerous program modules, including the operating system 2030, one or more application programs 2032, other program modules 2034, and program data 2036, may be stored in the drive and RAM 2012. All or part of the operating system, applications, modules, and / or data may also be cased in RAM 2012. It will be well understood that the present invention can be embodied in a variety of commercially available operating systems or combinations of operating systems.

[0207] The user can input commands and information to the computer 2002 through one or more wired / wireless input devices, such as a keyboard 2038 and a pointing device such as a mouse 2040. Other input devices (not shown) may include a microphone, IR remote control, joystick, gamepad, stylus pen, touchscreen, and so on. These and other input devices are often connected to the processing unit 2004 through an input device interface 2042 connected to the system bus 2008, but may also be connected through other interfaces such as parallel ports, IEEE1394 serial ports, game ports, USB ports, IR interfaces, and so on.

[0208] Monitor 2044 or other types of display devices are also connected to the system bus 2008 through interfaces such as the video adapter 2046. In addition to Monitor 2044, the computer typically includes other peripheral output devices such as speakers, printers, and so on (not shown).

[0209] Computer 2002 can operate in a networked environment using logical connections to one or more remote computers, such as remote computer(s) 2048, via wired and / or wireless communication. Remote computer(s) 2048 may be a workstation, server computer, router, personal computer, portable computer, microprocessor-based entertainment device, peer device, or other ordinary network node, and may include many or all of the components generally described for computer 2002, but for simplification, only the memory storage device 2050 is shown. The logical connections shown include wired / wireless connections to a local area network (LAN) 2052 and / or a larger network, such as a wide area network (WAN) 2054. Such LAN and WAN networking environments are common in offices and businesses and facilitate enterprise-wide computer networks such as intranets, all of which may connect to global computer networks, such as the Internet.

[0210] When used in a LAN networking environment, computer 2002 connects to the local network 2052 via a wired and / or wireless network interface or adapter 2056. Adapter 2056 can facilitate wired or wireless communication to LAN2052, which may also include a wireless access point provided therein for communication with the wireless adapter 2056. When used in a WAN networking environment, computer 2002 may include a modem 2058 or have other means of determining communication through WAN2054, such as connecting to a communication server on WAN2054 or communicating via the Internet. The modem 2058, which may be internal or external and wired or wireless, connects to the system bus 2008 via a serial port interface 2042. In a networked environment, program modules or parts thereof described with respect to computer 2002 may be stored in remote memory / storage device 2050. The network connections shown are illustrative, and it will be understood that other means of establishing communication links between computers may be used.

[0211] Computer 1602 operates to communicate with any wireless device or individual that is deployed and operates wirelessly, such as a printer, scanner, desktop and / or portable computer, PDA (portable data assistant), communication satellite, any equipment or location associated with a wirelessly discoverable tag, and a telephone. This includes at least Wi-Fi and Bluetooth® wireless technologies. Thus, the communication may be a predefined structure, like a conventional network, or it may simply be ad hoc communication between at least two devices.

[0212] It should be understood that the specific order or hierarchical structure of the steps in the presented process is an example of an exemplary approach. It should be understood that the specific order or hierarchical structure of the steps in the process may be rearranged within the scope of this disclosure based on design priorities. The method claims in this disclosure provide a variety of step elements in sample order, but are not intended to limit themselves to the specific order or hierarchical structure presented.

[0213] As described above, the relevant details are described in the best mode for carrying out the present invention. [Industrial applicability]

[0214] It can be used in devices and systems for managing patients who show adverse reactions to drugs.

Claims

1. A method for generating predictive results for adverse drug reactions using artificial intelligence technology performed by a computing device, The process involves constructing a periodic conversion database using raw data for drug adverse reaction detection, which is scheduled to be updated at predetermined time intervals. The steps include determining a plurality of input variables for training an artificial intelligence-based predictive model for detecting adverse drug reactions, based on the data stored in the periodic conversion database, The process includes: training the predictive model using the determined input variables and a training dataset containing adverse reaction information for specific drugs and specific diseases, such that the predictive model outputs adverse reaction information for specific drugs and specific diseases in response to receiving input of the input variables; The step of constructing the aforementioned periodic conversion database is: The steps include: converting the terminology of the raw data to correspond to the common data standard of the prediction model; The process includes the step of constructing the periodic conversion database, which is updated in predetermined time intervals, using at least one of the following methods: a database linkage method, a file linkage method, or a change history classification linkage method. The input variables are determined using a method for determining variable importance, which involves using perturbed input data in which at least some of the variables have been modified, and output data output from the prediction model in response to the perturbed input data. method.

2. The step of constructing the periodic conversion database further includes the step of obtaining raw data for drug adverse reaction detection from the electronic medical record (EMR), The aforementioned raw data includes at least one of the following: drug use-related data, disease diagnosis data, diagnostic test data, and patient nursing record data. The method according to claim 1.

3. The aforementioned raw data is further obtained from at least one of FAERS, KAERS, WHO-ART, SIDER, and EU-ADR. The method according to claim 1.

4. The aforementioned input variables include at least one of the following: demographic variables, drug use-related variables, diagnostic test-related variables, nursing record-related variables, and drug side effect diagnosis-related variables. The method according to claim 1.

5. The aforementioned demographic variables include at least one of the following: sex, age, and region. The aforementioned drug use-related variables include at least one of the following: type of drug, duration of drug use, dosage of drug use, and drug use history. The aforementioned diagnostic test-related variables include at least one of the following: imaging tests, blood tests, and urine tests. The aforementioned nursing record-related variables are obtained from nursing record sheets through at least one of the following methods: optical character recognition (OCR), natural language processing (NLP), and text mining. The aforementioned drug side effect diagnosis-related variables include information related to disease diagnosis due to drug side effects, The method according to claim 4.

6. The aforementioned input variable is, The stage of handling outliers, The stage of performing missing value substitution, The stage of categorizing or recategorizing, The normalization or standardization stage, and the preprocessing based on, The method according to claim 1.

7. The step of training the predictive model for detecting adverse drug reactions is: The step of determining a predictive model for detecting adverse drug reactions from among several candidate models, The process includes: training the predictive model using the preprocessed input variables as input data so that the predictive model outputs the presence or absence of adverse drug reactions to a specific drug or a specific disease, or the probability of adverse drug reactions occurring to a specific drug or a specific disease; The method according to claim 6.

8. The aforementioned prediction model is selected from among candidate models consisting of Random Forest, Logistic Regression, Support Vector Machine, Decision Tree, Naive Bayes, K-Nearest Neighbor, XGBoost, LightGBM, CatBoost, Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), Transformer, and Generative Pre-trained Transformer (GPT). The method according to claim 1.

9. The steps include obtaining output results from each of the candidate models using a predetermined test dataset, The step involves performing a performance evaluation on the output results from each of the aforementioned candidate models, Based on the results of the performance evaluation, one of the candidate models is selected as the prediction model. The method according to claim 8.

10. The adverse reaction information for the aforementioned specific drugs and specific diseases is, The first piece of information indicates whether or not there is an adverse drug reaction to a specific drug, Second information showing the probability of adverse drug reactions occurring for a specific drug, Third-party information indicating the presence or absence of adverse drug reactions for a specific disease, This includes a fourth piece of information indicating the probability of adverse drug reactions occurring for a specific disease, The method according to claim 1.

11. The further step includes transmitting the output results of the prediction model for the adverse drug reaction to a user terminal. The method according to claim 1.

12. The output result transmitted to the user terminal is: The first result shows a binary classification of adverse drug reactions, categorized as detected or undetected, and The second result shows the probability notation for adverse drug reactions, The output includes at least one of the following: a third result showing a three-level categorization that associates drugs, adverse reactions, and risk levels. The method according to claim 11.

13. The output results from the prediction model are used to retrain the prediction model based on the accuracy evaluated for those output results. The method according to claim 1.

14. The feedback information corresponding to the output results transmitted from the user terminal that receives the output results from the prediction model is used to retrain the prediction model. The method according to claim 1.

15. A computer program stored on a computer-readable storage medium, The computer program, when executed by a computing device, causes the computing device to use artificial intelligence technology to generate prediction results for adverse drug reactions. The aforementioned operation is, The process involves constructing a periodic conversion database that is scheduled to be updated at predetermined time intervals using raw data for drug adverse reaction detection, and Based on the data stored in the periodic conversion database, the operation of determining multiple input variables for training the artificial intelligence-based predictive model for detecting adverse drug reactions, The operation includes training the predictive model to output adverse reaction information for a specific drug and a specific disease in response to input of the input variables, using the determined input variables and a training dataset containing adverse reaction information for a specific drug and a specific disease, The operation to construct the aforementioned periodic conversion database is as follows: The operation involves converting the terminology of the raw data to correspond to the common data standard of the prediction model, The operation includes constructing the periodic conversion database, which is updated in predetermined time intervals, using at least one of the following methods: a database linkage method, a file linkage method, or a change history classification linkage method. The input variables are determined using a method for determining variable importance, which involves using perturbed input data in which at least some of the variables have been modified, and output data output from the prediction model in response to the perturbed input data. A computer program stored on a computer-readable storage medium.

16. A computing device, It includes at least one processor and memory, The at least one processor is The process involves constructing a periodic conversion database that is scheduled to be updated at predetermined time intervals using raw data for detecting adverse drug reactions, and Based on the data stored in the periodic conversion database, the operation of determining multiple input variables for training the artificial intelligence-based predictive model for detecting adverse drug reactions, Using the determined input variables and a training dataset containing adverse reaction information for specific drugs and specific diseases, the predictive model is trained to output adverse reaction information for specific drugs and specific diseases in response to receiving input of the input variables. The operation to construct the aforementioned periodic conversion database is as follows: The operation involves converting the terminology of the raw data to correspond to the common data standard of the prediction model, The operation includes constructing the periodic conversion database, which is updated in predetermined time intervals, using at least one of the following methods: a database linkage method, a file linkage method, or a change history classification linkage method. The input variables are determined using a method for determining variable importance, which involves using perturbed input data in which at least some of the variables have been modified, and output data output from the prediction model in response to the perturbed input data. Computing device.

17. A method for generating predictive results for adverse drug reactions using artificial intelligence technology performed by a computing device, The steps include: obtaining input data for drug adverse reaction detection using raw data for drug adverse reaction detection, based on data stored in a periodic conversion database that is scheduled to be updated at predetermined time intervals; The step includes obtaining information on adverse reactions to specific drugs and specific diseases from the input data using an artificial intelligence-based predictive model, The aforementioned periodic conversion database is The operation involves converting the terminology of the raw data to correspond to the common data standard of the prediction model, The operation of constructing the periodic conversion database, which is updated in predetermined time intervals, using at least one of the following methods: a database linkage method, a file linkage method, or a change history classification linkage method, is generated based on the above. The multiple input variables for training the prediction model are determined using a method for determining variable importance, which involves using perturbed input data in which at least some of the variables are modified, and output data output from the prediction model in response to the perturbed input data. method.

Images