KR102611011B1 - Apparatus and method for analyzing electronic health record data - Google Patents

Abstract

Disclosed is an electronic health record data analysis device and method. An electronic health record data analysis device includes: a memory storing a program for analyzing electronic health record data; and a control unit that analyzes electronic health record data by executing the program, wherein the control unit generates an embedding vector by selectively aggregating function information for each timestamp of the electronic health record data formed based on timestamps. , Characterized in performing analysis on electronic health record data by inputting the generated embedding vector into a pre-trained artificial intelligence model.

Description

Electronic health record data analysis device and method {APPARATUS AND METHOD FOR ANALYZING ELECTRONIC HEALTH RECORD DATA}

Embodiments disclosed herein relate to electronic health record data analysis devices and methods, and more specifically, to electronic health record (EHR) data that effectively processes missing data included in the data and performs analysis of the EHR data. Pertains to a health record data analysis device and method.

Electronic Health Record (EHR) data is information generated by treating patients at medical institutions and refers to the patient's health information that is systematically collected in digital form and stored electronically. Electronic health record data includes, for example, vital signs, past medical history, medications and allergies, laboratory data (lab test function data), immunization dates and imaging reports, personal demographic information such as age and gender, etc. It can contain all information related to the patient.

In modern times, artificial intelligence (AI) models that obtain information by performing analysis based on EHR data and making predictions about various diseases are becoming active.

However, there may be missing data in these EHR data.

Although analysis of EHR data can be performed by replacing missing data with other values, performing analysis of EHR data by replacing missing data with other values varies depending on the imputation method. The performance of AI models may vary depending on alternative methods.

Therefore, research is needed on a system that can handle missing data included in EHR data without deteriorating the performance of AI models that perform analysis on EHR data.

Meanwhile, the above-described background technology is technical information that the inventor possessed for deriving the present invention or acquired in the process of deriving the present invention, and cannot necessarily be said to be known technology disclosed to the general public before filing the application for the present invention. .

Korean Patent No. 10-2304370 (announced on September 24, 2021)

Embodiments disclosed herein are electronic health systems that can prevent the performance of artificial intelligence (AI) models from deteriorating by effectively processing missing data included in EHR data and performing analysis on EHR data. The purpose is to provide a recorded data analysis device and method.

Other objects and advantages of the present invention can be understood from the following description and will be more clearly understood through an example. In addition, it will be readily apparent that the objects and advantages of the present invention can be realized by means and combinations thereof as indicated in the claims.

As a technical means for achieving the above-described technical problem, an electronic health record data analysis device includes: a memory storing a program for analyzing electronic health record data; and a control unit that analyzes electronic health record data by executing the program, wherein the control unit generates an embedding vector by selectively aggregating function information for each timestamp of the electronic health record data formed based on timestamps. , Characterized in performing analysis on electronic health record data by inputting the generated embedding vector into a pre-trained artificial intelligence model.

According to another embodiment, the electronic health record data analysis method performed by the electronic health record data analysis device generates an embedding vector by selectively aggregating function information for each timestamp of electronic health record data formed based on timestamps. step; and performing analysis on electronic health record data by inputting the generated embedding vector into a pre-trained artificial intelligence model.

According to another embodiment, the recording medium is a computer-readable recording medium on which a program for performing an electronic health record data analysis method is recorded. The electronic health record data analysis method includes generating an embedding vector by selectively aggregating function information for each timestamp of electronic health record data formed based on timestamps; and performing analysis on electronic health record data by inputting the generated embedding vector into a pre-trained artificial intelligence model.

According to another embodiment, the computer program is a computer program stored in a recording medium that is performed by an electronic health record data analysis device and performs an electronic health record data analysis method. The electronic health record data analysis method includes generating an embedding vector by selectively aggregating function information for each timestamp of electronic health record data formed based on timestamps; and performing analysis on electronic health record data by inputting the generated embedding vector into a pre-trained artificial intelligence model.

According to one of the means for solving the above-mentioned problem, the missing data included in the EHR data is effectively processed using an imputation free algorithm and the analysis of the EHR data is performed, thereby preventing inappropriate bias during data analysis. ) has the effect of preventing the performance of the AI model from deteriorating.

The effects that can be obtained from the disclosed embodiments are not limited to the effects mentioned above, and other effects not mentioned are clear to those skilled in the art to which the disclosed embodiments belong from the description below. It will be understandable.

Hereinafter, the attached drawings illustrate preferred embodiments disclosed in the present specification, and serve to further understand the technical idea disclosed in the present specification along with specific details for carrying out the invention, and thus the drawings disclosed in the present specification The contents should not be construed as limited to the matters described in such drawings.
1 is a functional block diagram of an electronic health record data analysis device according to an embodiment.
2 and 3 are exemplary diagrams for explaining an electronic health record data analysis device according to an embodiment.
Figure 4 is a flowchart of an electronic health record data analysis method according to one embodiment.

Below, various embodiments will be described in detail with reference to the attached drawings. The embodiments described below may be modified and implemented in various different forms. In order to more clearly explain the characteristics of the embodiments, detailed descriptions of matters widely known to those skilled in the art to which the following embodiments belong have been omitted. In addition, in the drawings, parts that are not related to the description of the embodiments are omitted, and similar parts are given similar reference numerals throughout the specification.

Throughout the specification, when a configuration is said to be “connected” to another configuration, this includes not only cases where it is “directly connected,” but also cases where it is “connected with another configuration in between.” In addition, when a configuration “includes” a configuration, this means that other configurations may be further included rather than excluding other configurations, unless specifically stated to the contrary.

Hereinafter, embodiments will be described in detail with reference to the attached drawings.

However, before explaining this, we first define the meaning of the terms used below.

A deep neural network (DNN) is an artificial neural network (ANN) consisting of several hidden layers between an input layer and an output layer. Deep neural networks, like general artificial neural networks, can model complex non-linear relationships. For example, in a deep neural network structure for an object identification model, each object can be expressed as a hierarchical composition of basic image elements. At this time, additional layers can gradually integrate the characteristics of the lower layers. This characteristic of deep neural networks allows complex data to be modeled with fewer nodes compared to similarly performed artificial neural networks. According to one embodiment, prevalence or mortality rates for various diseases can be predicted by analyzing electronic health record data using a deep neural network. According to one embodiment, the deep neural network may be the known long short-term memory (LSTM) and bidirectional recurrent imputation for time series (BRITS). At this time, LSTM can be designed by referring to the paper (Hochreiter S, Schmidhuber J. Long short-term memory. Neural Computation 1997; 9(8):1735-1780.), and BRITS can be designed by referring to the paper (Cao W, Wang D , Li J, Zhou H, Li L, Li Y. BRITS: Bidirectional Recurrent Imputation for Time Series. Advances in Neural Information Processing Systems 2018.).

Terms that require explanation other than those defined above will be explained separately below.

FIG. 1 is a functional block diagram of an electronic health record data analysis device according to an embodiment, and FIGS. 2 and 3 are exemplary diagrams for explaining an electronic health record data analysis device according to an embodiment.

The electronic health record data analysis device 100 according to this embodiment may be implemented as an electronic terminal with an application that can interact with the user installed, as a server, or as a server-client system, and may be implemented as a server-client system. If available, it may include an electronic terminal with an online service application installed for interaction with the user.

At this time, the electronic terminal can be implemented as a computer, portable terminal, television, wearable device, etc. that can connect to a remote server through a network or connect to other devices and servers. Here, the computer includes, for example, a laptop, desktop, laptop, etc. equipped with a web browser, and the portable terminal is, for example, a wireless communication device that guarantees portability and mobility. , PCS (Personal Communication System), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), GSM (Global System for Mobile communications), IMT (International Mobile Telecommunication)-2000, CDMA (Code) All types of handhelds such as Division Multiple Access)-2000, W-CDMA (W-Code Division Multiple Access), Wibro (Wireless Broadband Internet), Smart Phone, Mobile WiMAX (Mobile Worldwide Interoperability for Microwave Access), etc. (Handheld)-based wireless communication device may be included. Additionally, television may include IPTV (Internet Protocol Television), Internet TV (Internet Television), terrestrial TV, cable TV, etc. Furthermore, a wearable device is a type of information processing device that can be worn directly on the human body, such as a watch, glasses, accessories, clothing, or shoes, and can connect to a remote server or connect to another device via a network directly or through another information processing device. can be connected with

In addition, the server may be implemented as a computer capable of communicating via a network with an electronic terminal installed with an application or web browser for interaction with the user of the electronic health record data analysis device 100, or as a cloud computing server. Additionally, the server may include a storage device capable of storing data or may store data through a third party server.

As described above, the electronic health record data analysis device 100 may be implemented in the form of any one of an electronic terminal, a server, or a server-client system. When implemented as a server, the electronic health record data analysis device 100 The constituting components may be performed on a plurality of physically separated servers or may be performed on one server.

Referring to FIG. 1, the electronic health record data analysis device 100 according to this embodiment includes an input/output unit 110, a communication unit 120, a memory 130, and a control unit 140.

Referring to FIG. 2, the electronic health record data analysis device 100 inputs electronic health record data into a preset algorithm (A) and inputs the output value into a pre-trained artificial intelligence model (B) to analyze electronic health. Analysis can be performed on historical data.

At this time, the preset algorithm (A) may be the Masked Attention-based timestamp-wise Data Embedding (MADE) algorithm. According to one embodiment, the electronic health record data analysis device 100 may form electronic health record data based on timestamps using the MADE algorithm, and may analyze the electronic health record data formed based on the timestamps for each timestamp. Optionally, feature information can be aggregated to generate an embedding vector. Meanwhile, the electronic health record data analysis device 100 may generate an embedding vector by converting multivariate time series data into a token-wise vector format. Here, the functional information may be the patient's health record information included in the electronic health record data. For example, it may be the patient's health record information such as pulse, blood pressure, hemoglobin level, etc. Meanwhile, the generated embedding vector may be a fixed-length dimension vector.

Thereafter, the electronic health record data analysis device 100 may perform analysis on the electronic health record data by inputting the generated embedding vector into the pre-trained artificial intelligence model (B), and the analyzed content may be stored in the electronic health record. This may be information that predicts prevalence or mortality rates for various diseases included in the data. As described above, by generating an embedding vector without replacing missing values of electronic health record data, it is possible to prevent inappropriate bias from occurring during subsequent analysis of electronic health record data.

Meanwhile, the MADE (Masked Attention-based timestamp-wise Data Embedding) algorithm is described in more detail as follows.

■ MADE (Masked Attention-based timestamp-wise Data Embedding) algorithm

MADE is based on the transformer neural network architecture, and the core mechanism of the algorithm is the attention mechanism of the transformer. Basically, MADE generates an embedding vector by selectively aggregating functional information for each timestamp of electronic health record data, which is formed as multivariate time series data that represents the patient's condition serially based on time, and stores the generated embedding vector in advance. It is provided as input to the learned artificial intelligence model.

Since the Transformer Layer uses input data in the format of a token-specific time series vector, the multivariate time series data must be converted into the token-specific time series vector format. For example, input features (here, is the feature number represents, is the timestamp number ) is applied to the MADE algorithm, it second fully connected layer and pass the d-dimensional vector It is converted to . Then the MADE algorithm silver A single timestamp indicated by The entire feature token vector of is used as input data. Meanwhile, the Transformer layer uses the Scaled Dot-Product Algorithm or the QKV Attention Algorithm to determine the token vector Calculate . At this time, the calculation equation is It can be. Here, the matrix , and can refer to the query, key and value respectively, and the numbers for the dimensions are the key matrix of It can be displayed as . Meanwhile, the procession , and , the same input token vector It can be calculated from In the MADE algorithm, the QKV attention matrix is the token vector and the mask matrix is calculated using , and this mask matrix is It is possible to show an index of the validity of feature observations. Throughout the forwarding pass of the MADE algorithm, the input token vector time that information state token vector As it is aggregated, it can be created as an embedding vector. At this time, the generated embedding vector, that is, the state token vector can be extracted as a result from the MADE algorithm and provided as an input vector (Embedded Token) to a deep neural network (DNN) algorithm, a pre-trained artificial intelligence model. Meanwhile, as described above, timestamp data (input feature value) All state token vectors using only one unit of Since calculating , no location encoding tokens or other information tokens (e.g., type embedding tokens, etc.) are needed. Meanwhile, the MADE algorithm described above is optimized by the loss function of LSTM and BRITS, which are pre-trained algorithms described later, so that missing feature values can be ignored and the measured information can be aggregated with the optimal weight for the given task. .

The input/output unit 110 may include an input unit for receiving input from a user and an output unit for displaying information such as a task performance result or the status of the electronic health record data analysis device 100 . For example, the input/output unit 110 may include an operation panel that receives user input and a display panel that displays a screen.

Specifically, the input unit may include devices that can receive various types of input, such as a keyboard, physical button, touch screen, camera, or microphone. Additionally, the output unit may include a display panel or a speaker. However, the input/output unit 110 is not limited to this and may include a configuration that supports various inputs and outputs.

The communication unit 120 may perform wired or wireless communication with other devices and/or networks. To this end, the communication unit 120 may include a communication module that supports at least one of various wired and wireless communication methods. For example, the communication module may be implemented in the form of a chipset.

Meanwhile, wireless communications supported by the communication unit 120 include, for example, Wi-Fi (Wireless Fidelity), Wi-Fi Direct, Bluetooth, Bluetooth Low Energy (BLE), and UWB (Ultra Wide Band). , NFC (Near Field Communication), LTE, LTE-Advanced, etc. may be wireless mobile communications. Additionally, wired communication supported by the communication unit 120 may be, for example, USB or HDMI (High Definition Multimedia Interface).

The memory 130 can install and store various types of data, such as files, applications, and programs, and may be configured to include at least one of various types of memory, such as RAM, HDD, and SSD. The control unit 140, which will be described later, may access and use data stored in the memory 130, or may store new data in the memory 130. Additionally, the control unit 140 may execute a program installed in the memory 130. Meanwhile, a program for performing an electronic health record data analysis method according to an embodiment may be installed in the memory 130. Additionally, various electronic health record data about the patient may be stored in advance in the memory 130.

The control unit 140 includes at least one processor such as CPU, GPU, Arduino, etc., and can control the overall operation of the electronic health record data analysis device 100. That is, the control unit 140 can control other components included in the electronic health record data analysis device 100 to perform analysis on the electronic health record data. Additionally, the control unit 140 may execute a program stored in the memory 130, read a file stored in the memory 130, or store a new file in the memory 130.

According to an embodiment, the control unit 140 may perform preprocessing on electronic health record data, which is raw data stored in the memory 130. At this time, the electronic health record data recorded as raw data is irregularly sampled time series data, and there are many missing values that occur regularly due to the non-automated manned data collection process, so it is desirable to perform sophisticated preprocessing for analysis. Details related to preprocessing of electronic health record data, which are raw data, are as follows.

The control unit 140 may form electronic health record data based on timestamps, and the electronic record data formed at this time may be formed as multivariate time series data that serially represents the patient's condition based on time for each timestamp. .

The control unit 140 may generate an embedding vector by selectively aggregating function information for each timestamp of electronic health record data formed based on timestamps. Here, the functional information may be the patient's health record information included in the electronic health record data. For example, it may be the patient's health record information such as pulse, blood pressure, hemoglobin level, etc. Meanwhile, the generated embedding vector may be a fixed-length dimensional vector.

More specifically, the control unit 140 may generate an embedding vector by converting multivariate time series data into a time series vector format for each token. Electronic health record data, which is raw data, is formed as multivariate time series data that represents the patient's condition serially based on time by timestamp. According to one embodiment, the Transformer Layer of MADE (Masked Attention-based timestamp-wise Data Embedding), which is a preset algorithm (A), as shown in Figure 2, uses input data in vector format for each token. Therefore, it is necessary to convert multivariate time series data into token-specific time series vector format. Meanwhile, according to one embodiment, the generated embedding vector may be a multidimensional embedding vector of fixed length.

The control unit 140 may perform analysis on electronic health record data by inputting the generated embedding vector into a pre-trained artificial intelligence model. The pre-trained artificial intelligence model may be a deep neural network algorithm. At this time, according to one embodiment, the deep neural network algorithm, which is a pre-trained artificial intelligence model, may be long short-term memory (LSTM) and bidirectional recurrent imputation for time series (BRITS). At this time, LSTM can be designed by referring to the paper (Hochreiter S, Schmidhuber J. Long short-term memory. Neural Computation 1997; 9(8):1735-1780.), and BRITS can be designed by referring to the paper (Cao W, Wang D , Li J, Zhou H, Li L, Li Y. BRITS: Bidirectional Recurrent Imputation for Time Series. Advances in Neural Information Processing Systems 2018.). Meanwhile, according to one embodiment, the structure of LSTM combined with MADE, a preset algorithm (A), is as shown in (a) of Figure 3, and the structure of BRITS combined with MADE, a preset algorithm (A) may be as shown in (b) of FIG. 3. The LSTM model as described above can perform output prediction of a given data instance by aggregating representative vectors and static feature values through a densely connected layer. Additionally, the BRITS model is similar to the GRU-D model in that it uses two additional pieces of information in the raw time series data, with the only difference being that GRU cells are replaced with LSTM cells. The first of the two additional pieces of information used in the GRU-D model is the masking vector ( ) to the viewpoint ( ) represents the index of the missing feature, and the second is the time point at the last observation ( ) (interval vector) representing the time interval until ( ) can be. GRU-D is an interval vector ( ) to convert feature values to average. For example, the interval vector ( ) is the decay rate vector ( ), which is a timestamp ( ) the feature values of the time interval vector ( ) can be reduced to the global average value of the feature in proportion to. The exact equation for the vector is , where: and are the weight and bias parameters, respectively. also, is another decay rate vector applied to the hidden state vector of the GRU. Is Calculated in the same way, but with different weights ( ) and bias parameters ( ) is used. vector( ) is the attenuation rate vector ( ) is used to calculate the imputation of missing data. For example, a timestamp ( ) imputed input vector ( )Is am. At this time, represents the global average value of the input vector. also, is used as input data for each timestamp of GRU-D along with the input vector.

Once trained, the BRITS model can autoregressively predict the feature value of the next timestamp state along with the basic prediction task. At this time, autoregression can iteratively replace missing values in subsequent timestamp input data until valid observation data is reached. Afterwards, when calculating the training loss, the mean square error loss between the observed and imputed values can be calculated along with the task-specific loss.

Meanwhile, since the above-described LSTM and BRITS are known technologies, detailed descriptions will be omitted.

Figure 4 is a flowchart of an electronic health record data analysis method according to one embodiment.

The electronic health record data analysis method according to the embodiment shown in FIG. 4 includes steps processed in time series in the electronic health record data analysis device 100 shown in FIGS. 1 to 3. Therefore, even if the content is omitted below, the content described above regarding the electronic health record data analysis device 100 shown in FIGS. 1 to 3 is the electronic health record data analysis method according to the embodiment shown in FIG. 4. It can also be applied.

As shown in FIG. 4, the electronic health record data analysis device 100 according to this embodiment generates an embedding vector by selectively aggregating function information for each timestamp of electronic health record data formed based on timestamps. (S410). Meanwhile, the electronic health record data analysis device 100 may generate an embedding vector by converting multivariate time series data into a time series vector format for each token. Here, the functional information may be the patient's health record information included in the electronic health record data. For example, it may be the patient's health record information such as pulse, blood pressure, hemoglobin level, etc. Meanwhile, the generated embedding vector may be a fixed-length dimensional vector.

Next, the electronic health record data analysis device 100 performs analysis on the electronic health record data by inputting the embedding vector generated in step S410 into a pre-trained artificial intelligence model (S420). At this time, the analyzed content may be information predicting prevalence or mortality rates for various diseases included in electronic health record data.

The term '~unit' used in the above embodiments refers to software or hardware components such as FPGA (field programmable gate array) or ASIC, and the '~unit' performs certain roles. However, '~part' is not limited to software or hardware. The '~ part' may be configured to reside in an addressable storage medium and may be configured to reproduce on one or more processors. Therefore, as an example, '~ part' refers to components such as software components, object-oriented software components, class components, and task components, processes, functions, properties, and procedures. , subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays and variables.

The functions provided within the components and 'parts' may be combined into a smaller number of components and 'parts' or may be separated from additional components and 'parts'.

In addition, the components and 'parts' may be implemented to regenerate one or more CPUs within the device or secure multimedia card.

Meanwhile, the electronic health record data analysis method according to an embodiment described throughout this specification may also be implemented in the form of a computer-readable medium that stores instructions and data executable by a computer. At this time, instructions and data can be stored in the form of program code, and when executed by a processor, they can generate a certain program module and perform a certain operation. Additionally, computer-readable media can be any available media that can be accessed by a computer and includes both volatile and non-volatile media, removable and non-removable media. Additionally, computer-readable media may be computer recording media, which are volatile and non-volatile implemented in any method or technology for storage of information such as computer-readable instructions, data structures, program modules, or other data. It can include both volatile, removable and non-removable media. For example, computer recording media may be magnetic storage media such as HDDs and SSDs, optical recording media such as CDs, DVDs, and Blu-ray discs, or memory included in servers accessible through a network.

Additionally, the electronic health record data analysis method according to an embodiment described throughout this specification may be implemented as a computer program (or computer program product) including instructions executable by a computer. A computer program includes programmable machine instructions processed by a processor and may be implemented in a high-level programming language, object-oriented programming language, assembly language, or machine language. . Additionally, the computer program may be recorded on a tangible computer-readable recording medium (eg, memory, hard disk, magnetic/optical medium, or solid-state drive (SSD)).

Accordingly, the electronic health record data analysis method according to an embodiment described throughout this specification can be implemented by executing the above-described computer program by a computing device. The computing device may include at least some of a processor, memory, a storage device, a high-speed interface connected to the memory and a high-speed expansion port, and a low-speed interface connected to a low-speed bus and a storage device. Each of these components is connected to one another using various buses and may be mounted on a common motherboard or in some other suitable manner.

Here, the processor can process instructions within the computing device, such as displaying graphical information to provide a graphic user interface (GUI) on an external input or output device, such as a display connected to a high-speed interface. These may include instructions stored in memory or a storage device. In other embodiments, multiple processors and/or multiple buses may be utilized along with multiple memories and memory types as appropriate. Additionally, the processor may be implemented as a chipset consisting of chips including multiple independent analog and/or digital processors.

Additionally, memory stores information within a computing device. In one example, memory may be comprised of volatile memory units or sets thereof. As another example, memory may consist of non-volatile memory units or sets thereof. The memory may also be another type of computer-readable medium, such as a magnetic or optical disk.

Additionally, the storage device can provide a large amount of storage space to the computing device. A storage device may be a computer-readable medium or a configuration that includes such media, and may include, for example, devices or other components within a storage area network (SAN), such as a floppy disk device, a hard disk device, an optical disk device, Or it may be a tape device, flash memory, or other similar semiconductor memory device or device array.

The above-described embodiments are for illustrative purposes, and those of ordinary skill in the technical field to which the above-described embodiments belong will recognize that they can be easily modified into other specific forms without changing the technical idea or essential features of the above-described embodiments. You will understand. Therefore, the embodiments described above should be understood in all respects as illustrative and not restrictive. For example, each component described as unitary may be implemented in a distributed manner, and similarly, components described as distributed may also be implemented in a combined form.

The scope sought to be protected through this specification is indicated by the patent claims described later rather than the detailed description above, and all changes or modified forms derived from the meaning and scope of the claims and their equivalent concepts are included in the scope of the present invention. It should be interpreted as being

100: Electronic health record data analysis device
110: input/output unit
120: Department of Communications
130: memory
140: control unit

Claims (10)

An electronic health record data analysis device for predicting values for various diseases by analyzing electronic health record data using a pre-trained artificial intelligence model,
a memory storing a program for analyzing electronic health record data; and
A control unit that analyzes electronic health record data by executing the program,
The control unit,
By selectively aggregating functional information for each timestamp of electronic health record data formed based on timestamps, an embedding vector is generated without replacing missing values of the electronic health record data, and the generated embedding vector is used as the pre-learned artificial intelligence. Characterized by performing analysis on electronic health record data by inputting it into an intelligence model,
The control unit,
An electronic health record data analysis device that converts the electronic health record data formed of multivariate time series data into a time series vector format for each token, inputs it to a transformer layer of the MADE algorithm, and generates the embedding vector as an output of the transformer layer. . delete According to claim 1,
The embedding vector is,
Electronic health record data analysis device, characterized in that it is a fixed-length multidimensional embedding vector. According to claim 1,
The pre-trained artificial intelligence model is,
An electronic health record data analysis device characterized by a deep neural network algorithm. In the electronic health record data analysis method, which is performed by an electronic health record data analysis device and predicts values for various diseases by analyzing the electronic health record data using a pre-trained artificial intelligence model,
Generating an embedding vector without replacing missing values of the electronic health record data formed based on timestamps by selectively aggregating function information for each timestamp; and
Comprising: performing analysis on electronic health record data by inputting the generated embedding vector into a pre-trained artificial intelligence model,
The step of generating the embedding vector is,
Electronic health, comprising the step of converting the electronic health record data formed of multivariate time series data into a time series vector format for each token, inputting it to a transformer layer of the MADE algorithm, and generating the embedding vector as an output of the transformer layer. Methods for analyzing historical data. delete According to claim 5,
The embedding vector is,
Electronic health record data analysis method, characterized in that it is a fixed-length multidimensional embedding vector. According to claim 5,
The pre-trained artificial intelligence model is,
An electronic health record data analysis method characterized by a deep neural network algorithm. A computer-readable recording medium on which a program for performing the method according to claim 5 is recorded. A computer program stored on a recording medium for performing the method described in claim 5, performed by an electronic health record data analysis device.