KR20230088327A - Prediction method for responses of biologic agent in patients with ankylosing spondylitis and analysis apparatus - Google Patents

Abstract

The analysis field to predict the reactivity of biological agents to spondylitis patients includes an input device that receives clinical information about ankylosing spondylitis patients, which is the subject of analysis, and pre-learned learning that outputs the responsiveness of biological agents based on clinical information about ankylosing spondylitis patients. It includes a storage device for storing a model and an arithmetic device for predicting the reactivity of a biological agent for an ankylosing spondylitis patient, the analysis target, by inputting the received clinical information into the learning model.

Description

Responsiveness prediction method and analysis device for biological agents for ankylosing spondylitis patients

The technique described below is a technique for predicting the reactivity of biological agents for ankylosing spondylitis patients.

Ankylosing spondylitis (AS) is a disease in which inflammation in the area where tendon attaches to bone results in calcification of the area. Drugs for the treatment of ankylosing spondylitis include traditional chemical agents and biological agents. Representative chemical agents include methotrexate and sulfasalazine.

Biological agents are usually second-line treatments used for patients who have failed chemical therapy. Biological agents show relatively good effects, but treatment is known to fail in about 20% of patients with ankylosing spondylitis.

US Patent Publication US2016-0274090 (2016.09.22)

Biological agents are recently developed drugs that are expensive, have strong immunosuppressive effects, and have various side effects. Therefore, it is necessary to accurately predict the reactivity of a biological agent as a second-line treatment for ankylosing spondylitis patients.

The technology to be described below is intended to provide a technique for predicting responsiveness to biological agents using a learning model using clinical information of ankylosing spondylitis patients.

An analysis device that predicts the reactivity of biological agents to spondylitis patients is an input device that receives clinical information about ankylosing spondylitis patients, and outputs responsiveness to multiple biological agents based on clinical information about ankylosing spondylitis patients. It includes a storage device for storing a plurality of learning models to be stored, and an arithmetic device for predicting the reactivity of a biological agent for an ankylosing spondylitis patient, the analysis target, by inputting the received clinical information into the learning model.

The clinical information includes demographic information, self-diagnosis index, and administered chemical agent information, and the self-diagnosis index includes BASDAI (Bath Ankylosing Spondylitis Disease Activity Index), BASFI (Bath Ankylosing Spondylitis Functional Index), and PtGA (patient global assessment). of disease activity).

The clinical information may further include at least one of a left sacroiliitis on X-ray (Left) image and a right sacroiliitis on X-ray image.

In the technology described below, a learning model quickly and accurately predicts the response to a biological agent based on the clinical information of an ankylosing spondylitis patient. Therefore, the technology described below can increase the treatment effect and reduce the cost by administering a biological agent only to patients with ankylosing spondylitis in which a high treatment effect is expected.

1 is an example of a system for predicting the reactivity of a biological agent for an ankylosing spondylitis patient.
Figure 2 is an example of the process of building a learning model for predicting the responsiveness of biological agents for ankylosing spondylitis patients.
3 is an example of an analysis device.
4 is a result of evaluating a learning model using verification data.

Since the technology to be described below can have various changes and various embodiments, specific embodiments will be illustrated in the drawings and described in detail. However, this is not intended to limit the technology described below to specific embodiments, and it should be understood to include all modifications, equivalents, or substitutes included in the spirit and scope of the technology described below.

Terms such as first, second, A, B, etc. may be used to describe various elements, but the elements are not limited by the above terms, and are merely used to distinguish one element from another. used only as For example, without departing from the scope of the technology described below, a first element may be referred to as a second element, and similarly, the second element may be referred to as a first element. The terms and/or include any combination of a plurality of related recited items or any of a plurality of related recited items.

In terms used in this specification, singular expressions should be understood to include plural expressions unless clearly interpreted differently in context, and terms such as “comprising” refer to the described features, numbers, steps, operations, and components. , parts or combinations thereof, but it should be understood that it does not exclude the possibility of the presence or addition of one or more other features or numbers, step-action components, parts or combinations thereof.

Prior to a detailed description of the drawings, it is to be clarified that the classification of components in the present specification is merely a classification for each main function in charge of each component. That is, two or more components to be described below may be combined into one component, or one component may be divided into two or more for each more subdivided function. In addition, each component to be described below may additionally perform some or all of the functions of other components in addition to its main function, and some of the main functions of each component may be performed by other components. Of course, it may be dedicated and performed by .

In addition, in performing a method or method of operation, each process constituting the method may occur in a different order from the specified order unless a specific order is clearly described in context. That is, each process may occur in the same order as specified, may be performed substantially simultaneously, or may be performed in the reverse order.

The technique described below is a technique for predicting in advance the reactivity of a biological agent to an ankylosing spondylitis patient.

First, a brief description of biological agents is given.

A biological agent is an antibody or a substance similar to it made using genetic recombination technology, and mainly serves to block the function of a substance or cell presumed to be involved in the development of ankylosing spondylitis. Currently used major biological agents can be divided into drugs that inhibit the function of cytokines and drugs that inhibit cell activation depending on the relevant substances and cells.

TNF (Tumor Necrosis Factor) antagonist is a drug that is first developed and used among biological agents, and reduces the production of inflammatory cytokines by inhibiting TNF function. Representative TNF antagonists include etanercept, infliximab, adalimumab, and golimumab. In the following description, the biological agents for ankylosing spondylitis are mainly described as TNF antagonists. However, the techniques described below are not necessarily applied only to predicting the reactivity of TNF antagonists.

The technique described below predicts the reactivity of a biological agent to an ankylosing spondylitis patient using a learning model. The learning model means a machine learning model. A learning model is meant to include various types of models. For example, learning models include ensemble, decision tree, random forest, K-nearest neighbor (KNN), Naive Bayes, support vector machine (SVM), artificial neural network (ANN), etc. there is

On the other hand, a device that predicts the responsiveness of biological agents to patients with spondylitis using a learning model is called an analysis device. The analysis device is a device capable of processing data and may be in the form of a PC, a smart device, or a server.

1 is an example of a system 100 for predicting the reactivity of a biological agent for an ankylosing spondylitis patient. 1 illustrates an example in which the analysis device is a computer terminal 150 and a server 180 .

In FIG. 1 , the user A can predict the reactivity of a biological agent using the computer terminal 150 . In this case, the computer terminal 150 receives input data necessary for prediction and transfers the input data to the learning model to calculate the predictive result of the reactivity of the biological agent.

The server 180 may predict the reactivity of the biological agent using the learning model. In this case, the server 180 receives input data necessary for prediction and inputs the input data to the learning model to calculate the predictive result of the reactivity of the biological agent. The user A may access the server 180 on the network and check the analysis result.

User A is a person who needs information on the reactivity of a biological agent for a specific ankylosing spondylitis patient. For example, the user A may be a doctor or researcher.

The input data is clinical information of an ankylosing spondylitis patient to be analyzed. Clinical information may be composed of information on various items of the patient. Specific clinical information applied to the learning model will be described later.

The process of obtaining clinical information is described. The patient may input self-diagnosis information about ankylosing spondylitis using the patient terminal 110 . A doctor may input medical treatment information about ankylosing spondylitis using the doctor terminal 120 . Furthermore, EMR (Electronic Medical Records, 130) of medical institutions may hold various information about patients. For example, the EMR 130 may retain patient demographic information, administered drug information, specific test results, medical images, accompanying disease information, and other information. Of course, the EMR 130 may also hold the patient's self-diagnosis information and the doctor's treatment information.

The computer terminal 150 and/or the server 180 may receive input of various clinical information obtainable through various paths. The computer terminal 150 and/or server 180 uses clinical information about a specific ankylosing spondylitis patient as input data. The computer terminal 150 and/or the server 180 inputs the input data into a previously learned learning model to predict the reactivity of the biological agent of the patient. 1 shows an example in which the computer terminal 150 and/or the server 180 use an ANN as a learning model.

The computer terminal 150 and/or the server 180 may forward the analysis result to the EMR 130 .

The process of constructing the learning model is described below. The researcher explains the process of building the learning model. Learning of the learning model is performed by researchers or developers using computer devices such as PCs and servers. Therefore, it will be described below that the computer device performs the learning process of the learning model.

2 is an example of a process 200 of constructing a learning model for predicting the reactivity of a biological agent for an ankylosing spondylitis patient.

First, the computer device acquires learning data for building a learning model (210). The process of preparing the learning data will be described below.

The researcher utilized data registered in KOBIO (Korean College of Rheumatology Biologics and Targeted Therapy) as a database (DB). KOBIO holds information obtained from 45 hospitals in Korea. The researcher extracted the data of patients who had basic clinical information and started biologic treatment from the DB. The researcher selected subjects who met the following conditions as a cohort. The researcher was a patient with ankylosing spondylitis found between December 2012 and February 2019, who started biological treatment for the first time and had progress data for more than one year. The researcher selected patients with ankylosing spondylitis who met the criteria of the Modified New York criteria for AS or ASAS (assessment of spondyloarthritis international society) as a cohort. The researcher selected patients who received biologic treatment for the first time as a cohort. As shown in Table 1 below, the researcher divided the cohort into a learning data set (611) and a verification data set (296) based on region.

The researcher built a learning model using the learning data belonging to the cohort, and used the verification data to verify the built learning model.

Training data includes input data and label values.

The input data includes clinical information of each patient belonging to the cohort. Clinical information may include various items. The features used by the researcher as input data are shown in Table 2 below. Of course, the learning model can be trained using some of the features below.

Features Category Features Category BASDAI One heart failure 6 BASFI One COPD 6 PtGA One Dyslipidemia 6 SJC 2 HBV 6 TJC 2 Hyperthyroidism 6 CRP 3 Hypothyroidism 6 ESR 3 Hyptertension 6 Hb 3 Ischemic heart disease 6 Hct 3 Liver disease 6 HLA-B27 3 Obesity 6 Plt 3 Other neurologic diseases 6 WBC 3 Osteoporosis 6 Age 4 Peptic ulcer 6 BMI 4 ILD 6 Smoking 4 weight loss 6 sex 4 DM 6 No NSAIDs use 5 History of joint replacement 6 Aceclofenac 5 History of osteoporotic fracture 6 Celecoixb 5 Headache 6 Etodolac 5 Malignancy 6 Meloxicam 5 Dactylitis 7 Morniflumate 5 Sacroiliitis 7 Nabumetone 5 Enthesitis 7 Naproxen 5 IBD 7 Nimesulide 5 History of peripheral arthritis 7 Other NSAIDs 5 Peripheral arthritis at baseline 7 Pelubiprofen 5 Psoriasis 7 Talniflumate 5 Family history of AS 7 Zaltoprofen 5 Response to NSAIDs 7 No csDMARDs use 5 Inflammatory back pain 7 Methotrexate 5 Uveitis 7 Sulfasalazine 5 Sacroiliitis on X-ray, Left 8 Alcohol 6 Sacroiliitis on X-ray, Right 8 Anemia 6 Latent TBc 9 Arrhythmia 6 Disease duration 9 Asthma 6

The categories of clinical information in Table 2 are as follows. Category 1 is patient self-diagnosis index, Category 2 is physical examination results by medical staff, Category 3 is laboratory test results, Category 4 is demographic (statistical) information, Category 5 is medication information administered, Category 6 is comorbidity information, category 7 is other joint characteristics, category 8 is medical images, and category 9 is other information.

Category 1 patient self-diagnosis index refers to the value of an index evaluated by the patient himself. The patient self-diagnosis indicator may be input by a patient through a device such as a terminal or may be obtained through a medical examination process. Furthermore, a self-diagnosis indicator may also be obtained through evaluation by a medical staff. Self-diagnosis indicators may include a Bath Ankylosing Spondylitis Disease Activity Index (BASDAI), a Bath Ankylosing Spondylitis Functional Index (BASFI), and a patient global assessment of disease activity (PtGA).

Category 2 physical examination results refer to the results of physical diagnosis or evaluation performed on a patient by medical staff. The physical examination result may include (i) a swollen joint count (SJC) and a tender joint count (TJC).

Category 3 laboratory test results include blood test results. Laboratory test results include CRP (C-reactive protein), ESR (erythrocyte sedimentation rate), Hb (hemoglobin), Hct (hematocrit), HLA-B27 (leukocyte antigen B27), Plt (platelets), WBC (leukocytes), etc. can include The laboratory test result is calculated as a numerical value or ratio of the corresponding item.

Demographic information of category 4 may include age, body mass index (BMI), smoking, sex, and the like.

Administered drug information for category 5 is information on drugs administered to the patient up to the point of analysis. Drug information is available on NSAIDs (Non-steroidal anti-inflammatory drugs), Aceclofenac, Celecoxib, Etodolac, Meloxicam, Morniflumate, Nabumetone ), naproxen, nimesulide, pelubiprofen, talniflumate, zaltoprofen, methotrexate, sulfasalazine, and the like. On the other hand, the drug information may indicate the absence of administration of a specific drug or chemical agent.

Category 6 is for comorbidity information: Alcoholism, Anemia, Arrhythmia, Asthma, Heart failure, COPD (chronic obstructive pulmonary disease), Dyslipidemia, HBV (epatitis B virus), hyperthyroidism, hypothyroidism, hypotension, ischemic heart disease, liver disease, obesity, other renal diseases ( Other neurologic disease), osteoporosis, peptic ulcer, interstitial lung disease (ILD), weight loss, diabetes mellitus (DM), history of joint replacement, osteoporosis It may include history of osteoporotic fracture, Headache, Malignancy, etc.

Category 7 other joint features include Dactylitis, Sacroiliitis, Enthesitis, inflammatory bowel disease (IBD), peripheral arthritis, Psoriasis, Family history of ankylosing spondylitis (AS), Response to NSAIDs, inflammatory back pain (Inflammatory back pain), uveitis (Uveitis), and the like.

Category 8 medical image information may include a left sacroiliitis on X-ray (Left) image, a right sacroiliitis on X-ray (Right) image, and the like.

Other information of category 9 may include latent TBc, disease duration, and the like.

The label value indicates the reactivity of the biologic to the patient with the corresponding clinical information. Label values can be used for various biological products.

The reactivity of biological agents for ankylosing spondylitis patients was determined based on ASAS20. Patients with an improvement of 20% or more based on ASAS20 after administration of biological agents were classified as high responders. Conversely, if there was an improvement of less than 20% or more based on the ASAS20 after administration of the biological agent, the response was classified as low. Of course, other values may be used as criteria for classifying high and low reactivity.

The computer device may uniformly pre-process the input data (220). For example, the computer device may change input data extracted from various DBs into the same format. The input data is a value belonging to a continuous range, and the categories can be different. Accordingly, the computer device may normalize by adjusting the scale between the input data. For example, the computer device may adjust the value of the input data to a value between 0 and 1.

The computer device performs a process of learning a learning model using the prepared input data (or preprocessed input data) (230). The learning model receives input data and outputs a certain result value (high or low responsiveness). The computer device performs a process (learning process) of optimizing parameters of the learning model by comparing a value output by the current learning model with a label value. The computer device repeatedly trains the learning model using a large amount of training data.

As described above, there are various types of biological agents. A learning model can be built regardless of the type of biological agent. Furthermore, the learning model may be separately constructed for each type of ankylosing spondylitis.

The researcher built ANN (Artificial Neural Network), RF (Random Forest), SVM (Support Vector Machine), and XGboost (Boosting) as learning models, respectively. The researcher studied various learning models of different types and individually verified their effectiveness. Of course, the learning model may use other models besides ANN, RF, SVM, and XGboost. In addition, the researcher separately constructed a regression model, a statistical approach, to compare model performance. Table 3 below shows the results of optimizing hyperparameters for RF, XGboost, and ANN.

For RF, the hyperparameters are the total number of trees, maximum depth of a tree, minimum sample split for subnode split, and minimum leaf samples for becoming a node. ). In the case of XGboost, the hyperparameters include the maximum tree depth, learning rate, and gamma value (a penalty value for deciding to split leaf nodes). In the case of ANNs, hyperparameters include the number of hidden layers, the number of hidden nodes, and the learning rate.

The researcher identified factors that greatly affect the results among the input data of the learning model. The investigators performed a feature importance analysis to determine which prenatal information was important for predicting biologics responsiveness in patients with ankylosing spondylitis. The researchers used Gini importance for RF and XGboost, and ANN selected important features using a so-called risk backpropagation technique. As a result of the analysis, BASFI, a patient self-diagnosis index, was identified as the most important factor, followed by BASDAI as an important factor.

3 is an example of the analysis device 300. The analysis device 300 corresponds to the above-described analysis devices (150 and 180 in FIG. 1). The analysis device 300 may be physically implemented in various forms. For example, the analysis device 300 may have a form of a computer device such as a PC, a network server, and a chipset dedicated to data processing.

The analysis device 300 may include a storage device 310, a memory 320, an arithmetic device 330, an interface device 340, a communication device 350, and an output device 360.

The storage device 310 may store a pre-learned learning model that outputs the reactivity of a biological agent based on clinical information about an ankylosing spondylitis patient. The learning model may vary as described above. For example, the learning model may be any one of a random forest, a support vector machine (SVM), an artificial neural network (ANN), and XGboost. Furthermore, as the learning model, individual models may be prepared for each type of biological agent. For example, as the learning model, individual models may be prepared for each biological agent, such as etanercept, infliximab, adalimumab, and golimumab.

The storage device 310 may store input clinical information.

The storage device 310 may store other programs or codes for processing input data.

The storage device 310 may store analysis results.

The memory 320 may store data and information generated in the course of the analysis device 300 predicting the reactivity of a biological agent.

The interface device 340 is a device that receives certain commands and data from the outside. The interface device 340 may receive clinical information (input data) of a patient to be analyzed from a physically connected input device or an external storage device.

The communication device 350 refers to a component that receives and transmits certain information through a wired or wireless network. The communication device 350 may receive clinical information (input data) of a patient to be analyzed from an external object. Alternatively, the communication device 350 may transmit the analysis result to an external object such as a user terminal.

Since the interface device 340 and the communication device 350 are configured to receive certain information and images from a user or other physical object, they can also be collectively referred to as input devices.

The output device 360 is a device that outputs certain information. The output device 360 may output interfaces and analysis results necessary for data processing.

The arithmetic device 330 may predict the reactivity of the biological agent to the analysis target using the learning model and program stored in the storage device 310 .

The calculation device 330 can predict the biological agent reactivity for the patient to be analyzed based on the output value calculated by inputting the input data to the learning model learned in advance.

The calculation device 330 may select a model according to the biological agent for which reactivity is to be predicted and input the input data. Alternatively, the calculation device 330 may evaluate all biological agents by inputting input data to all models. In this case, based on the evaluation results of multiple models, the computing device 330 may recommend a biological agent that is expected to be most effective for the current patient.

When clinical information is input, the arithmetic device 330 may perform certain pre-processing on the input data for input into the learning model.

The arithmetic device 330 may be a device such as a processor, an AP, or a chip in which a program is embedded that processes data and performs certain arithmetic operations.

Hereinafter, the effect of predicting the reactivity of biological agents for ankylosing spondylitis patients using the above-described learning model will be described. The researcher prepared data for verification in the process of preparing learning data, and tested the effect of the learned model using the verification data. Verification was performed using 3-fold cross validation. The performance of the learning model was evaluated by accuracy, area under curve of ROC (AUC), F1 score, and precision-recall curve.

4 is a result of evaluating a learning model using verification data. Referring to FIG. 4, it can be seen that RF, XGB (XGboost), ANN, and SVM all produced significant results.

In addition, the method for predicting the reactivity of a biological agent for ankylosing spondylitis patients as described above can be implemented as a program (or application) including an executable algorithm that can be executed on a computer. The program may be stored and provided in a temporary or non-transitory computer readable medium.

A non-transitory readable medium is not a medium that stores data for a short moment, such as a register, cache, or memory, but a medium that stores data semi-permanently and can be read by a device. Specifically, the various applications or programs described above are CD, DVD, hard disk, Blu-ray disk, USB, memory card, ROM (read-only memory), PROM (programmable read only memory), EPROM (Erasable PROM, EPROM) Alternatively, it may be stored and provided in a non-transitory readable medium such as EEPROM (Electrically EPROM) or flash memory.

Temporary readable media include static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), and enhanced SDRAM (Enhanced SDRAM). SDRAM, ESDRAM), Synchronous DRAM (Synclink DRAM, SLDRAM) and Direct Rambus RAM (DRRAM).

This embodiment and the drawings accompanying this specification clearly represent only a part of the technical idea included in the foregoing technology, and those skilled in the art can easily understand it within the scope of the technical idea included in the specification and drawings of the above technology. It will be obvious that all variations and specific examples that can be inferred are included in the scope of the above-described technology.

Claims (10)

an input device for receiving clinical information about an ankylosing spondylitis patient to be analyzed;
A storage device for storing a plurality of learning models that output responsiveness to a plurality of biological agents based on clinical information on an ankylosing spondylitis patient; and
Including an arithmetic device for predicting the reactivity of the biological agent for the ankylosing spondylitis patient, the analysis target, by inputting the input clinical information into the learning model,
The learning model is individually prepared in advance according to the type of biological agent,
The computing device recommends a biological agent for the ankylosing spondylitis patient based on the output results of the plurality of learning models,
The clinical information includes demographic information, self-diagnosis index, and administered chemical agent information, and the self-diagnosis index includes BASDAI (Bath Ankylosing Spondylitis Disease Activity Index), BASFI (Bath Ankylosing Spondylitis Functional Index), and PtGA (patient global assessment). An analysis device for predicting the reactivity of a biological agent for patients with spondylitis, including at least one of disease activity. According to claim 1,
The clinical information is an analysis device for predicting the reactivity of a biological agent for a patient with spondylitis, further including physical examination results, blood analysis results, and accompanying disease information. According to claim 2,
The physical examination result includes at least one of Swollen joint count (SJC) and Tender joint count (TJC),
The results of the blood analysis include CRP (C-reactive protein), ESR (erythrocyte sedimentation rate), Hb (hemoglobin), Hct (hematocrit), HLA-B27 (leukocyte antigen B27), Plt (platelets) and WBC (leukocytes). An analysis device for predicting the reactivity of a biological agent for a patient with spondylitis, including at least one. According to claim 2,
The clinical information is an analysis device for predicting the reactivity of a biological agent for a spondylitis patient further comprising at least one of a left sacroiliitis on X-ray image (Left) and a right sacroiliitis on X-ray image. According to claim 1,
The biologic is a TNF (Tumor Necrosis Factor) antagonist, an analysis device for predicting the responsiveness of a biological agent to a patient with spondylitis. an input device for receiving clinical information about an ankylosing spondylitis patient to be analyzed;
A storage device for storing a plurality of learning models that output responsiveness to a plurality of biological agents based on clinical information on an ankylosing spondylitis patient; and
Including an arithmetic device for predicting the reactivity of the biological agent to the ankylosing spondylitis patient, the analysis target, by inputting the input clinical information into the learning model,
The learning model is individually prepared in advance according to the type of biological agent,
The computing device recommends a biological agent for the ankylosing spondylitis patient based on the output results of the plurality of learning models,
The clinical information includes demographic information, self-diagnosis index, and administered chemical agent information, and the clinical information includes at least one of a Sacroiliitis on X-ray, Left image and a right sacroiliitis X-ray image. Analyzing device for predicting the reactivity of biological agents for patients with spondylitis further comprising a. According to claim 6,
The clinical information is an analysis device for predicting the reactivity of a biological agent for a patient with spondylitis, further including physical examination results, blood analysis results, and accompanying disease information. According to claim 7,
The physical examination result includes at least one of Swollen joint count (SJC) and Tender joint count (TJC),
The results of the blood analysis include CRP (C-reactive protein), ESR (erythrocyte sedimentation rate), Hb (hemoglobin), Hct (hematocrit), HLA-B27 (leukocyte antigen B27), Plt (platelets) and WBC (leukocytes). An analysis device for predicting the reactivity of a biological agent for a patient with spondylitis, including at least one. According to claim 6,
The self-diagnosis index predicts the reactivity of a biological agent for patients with spondylitis, including at least one of BASDAI (Bath Ankylosing Spondylitis Disease Activity Index), BASFI (Bath Ankylosing Spondylitis Functional Index), and PtGA (patient global assessment of disease activity). analysis device. According to claim 6,
The biologic is a TNF (Tumor Necrosis Factor) antagonist, an analysis device for predicting the responsiveness of a biological agent to a patient with spondylitis.

Images