Apparatus and method for predicting patient prognosis using machine learning model

Abstract

A method for analyzing patient prognosis using a machine learning model includes the steps of: a device receiving patient information; the analysis device inputting the patient information into a prediction model; and the analysis device predicting the patient's prognosis based on the output value of the prediction model. Patient information may include at least one of lactate dehydrogenase (LDH) levels, age, absolute lymphocyte counts (ALC), dyspnea (respiratory disorder), respiratory rate (RR), presence or absence of diabetes mellitus (DM), C-reactive protein (CRP) levels, absolute neutrophil counts (ANC), platelet counts (PLT), white blood cell counts (WBC), and saturation of peripheral oxygen (SPO2).

Description

Technology Field

[0001] The technology described below relates to a method for predicting a patient's prognosis using machine learning models. Background Technology

[0002] Machine learning refers to the process by which computers learn from data and perform pattern recognition. It is used to perform specific tasks or make predictions based on given data. Machine learning models are widely used for tasks such as data classification and object detection.

[0003] Many people died during the COVID-19 pandemic. Significant medical resources were deployed to prevent the rapid spread of the virus. Nevertheless, the surge in patients at hospitals placed a burden on these resources. In particular, as symptomatic and asymptomatic cases became mixed, instances of misclassification began to increase. Consequently, machine learning models began to be developed to accurately detect COVID-19. Prior art literature

[0004] Non-patent literature Deep-learning artificial intelligence analysis of clinical variables predicts mortality in COVID-19 patients (Journal of the American College of Emergency Physicians Open volume 1, Issue 6 p. 1364-1373) The problem to be solved

[0005] Conventional machine learning models have been used to diagnose COVID-19 or predict the prognosis of patients. However, these models utilized a wide variety of factors. Consequently, they required significant computing resources or took a long time to perform calculations. Alternatively, they relied on factors measured through expensive or time-consuming tests. As a result, conventional machine learning models faced difficulties in classifying patients in the early stages.

[0006] The technology described below provides a method for predicting a patient's prognosis using a machine learning model created by selecting factors that have a significant influence on predicting the patient's prognosis. means of solving the problem

[0007] A method for analyzing patient prognosis using a machine learning model includes the steps of: a device receiving patient information; the analysis device inputting the patient information into a prediction model; and the analysis device predicting the patient's prognosis based on the output value of the prediction model.

[0008] Patient information may include at least one of lactate dehydrogenase (LDH) levels, age, absolute lymphocyte counts (ALC), dyspnea (respiratory disorder), respiratory rate (RR), presence or absence of diabetes mellitus (DM), C-reactive protein (CRP) levels, absolute neutrophil counts (ANC), platelet counts (PLT), white blood cell counts (WBC), and saturation of peripheral oxygen (SPO2). Effects of the invention

[0009] Using the technology described below, it is possible to predict the patient's prognosis based on blood test results, information on the patient's underlying diseases, and factors related to the patient's current condition.

[0010] By using the technology described below, the patient's prognosis can be reliably predicted.

[0011] The technology described below can assist physicians in clinical judgment by classifying patients early. Therefore, in situations such as pandemics, it is possible to optimize and efficiently allocate medical resources. Brief explanation of the drawing

[0012] FIG. 1 shows the overall process of an analysis device (100) performing a patient prognosis prediction method using a machine learning model. FIG. 2 is a flowchart (200) of one embodiment in which an analysis device performs a patient prognosis prediction method using a machine learning model. FIGS. 3 to 10 show an example in which a researcher collects patient information, selects valid factors, builds a model based on the selected factors, and evaluates the performance of the built model. Figure 3 shows the process of classifying collected patient information. Figure 4 shows candidate variables included in the classified information on COVID patients. Figure 5 shows the process of selecting the final variables to be used for model construction and performance evaluation among the candidate variables included in the information of COVID-19 patients. Figure 6 shows some results of building a model using a subset of candidate variables and evaluating the performance of the built model. Figure 7 shows the results of analyzing the extent to which variables input into the selected DNN model have a significant influence on the result value. FIGS. 8 to 10 are embodiments of constructing a DDR Tree based on the results of predicting patient characteristics and prognosis, and interpreting the constructed DDR Tree. FIG. 11 is a configuration of one of the embodiments of an analysis device (300). Specific details for implementing the invention

[0013] The technology described below is subject to various modifications and may have various embodiments. Specific embodiments of the technology described below may be described in the drawings of the specification. However, this is for the purpose of explaining the technology described below and is not intended to limit the technology described below to specific embodiments. Accordingly, it should be understood that all modifications, equivalents, and substitutions that fall within the spirit and scope of the technology described below are included in the technology described below.

[0014] In the terms used below, singular expressions should be understood to include plural expressions unless the context clearly indicates otherwise, and terms such as "includes" should be understood to mean that the described features, number, steps, actions, components, parts, or combinations thereof exist, and not to exclude the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.

[0015] Before providing a detailed description of the drawings, it is to clarify that the classification of components in this specification is merely based on the primary function each component is responsible for. That is, two or more components described below may be combined into a single component, or a single component may be divided into two or more components based on more subdivided functions. Furthermore, each component described below may additionally perform some or all of the functions of other components in addition to its own primary function, and it is obvious that some of the primary functions of each component may be exclusively performed by other components.

[0016] Furthermore, in performing the method or operation method, each process constituting the method may occur differently from the specified order unless a specific order is clearly indicated in the 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.

[0018] Below, we examine the overall process by which the analysis device performs a patient prognosis prediction method utilizing a machine learning model.

[0019] FIG. 1 shows the overall process of an analysis device (100) performing a patient prognosis prediction method using a machine learning model.

[0020] The analysis device (100) can be physically implemented in various forms. For example, the analysis device (100) can take the form of a PC, laptop, smart device, server, or a chipset dedicated to data processing.

[0021] The analysis device (100) can receive patient information. The analysis device (100) can input patient information into a prediction model. The analysis device (100) can predict the patient's prognosis based on the output value of the prediction model.

[0022] Patient information may include at least one of lactate dehydrogenase (LDH) levels, age, absolute lymphocyte counts (ALC), dyspnea (respiratory disorder), respiratory rate (RR), presence or absence of diabetes mellitus (DM), C-reactive protein (CRP) levels, absolute neutrophil counts (ANC), platelet counts (PLT), white blood cell counts (WBC), and saturation of peripheral oxygen (SPO2).

[0024] The following describes in detail the process by which the analysis device performs a patient prognosis prediction method using a machine learning model.

[0025] FIG. 2 is a flowchart (200) of one embodiment in which an analysis device performs a patient prognosis prediction method using a machine learning model.

[0026] The analysis device can receive patient information (210).

[0027] Patient information may include information about COVID-19 patients. Patient information may include factors necessary to predict the prognosis of COVID-19 patients.

[0028] Patient information may include selected factors highly correlated with the patient's prognosis among factors measurable from COVID-19 patients.

[0029] Patient information may include at least one of lactate dehydrogenase (LDH) levels, age, absolute lymphocyte counts (ALC), dyspnea (respiratory disorder), respiratory rate (RR), presence or absence of diabetes mellitus (DM), C-reactive protein (CRP) levels, absolute neutrophil counts (ANC), platelet counts (PLT), white blood cell counts (WBC), and saturation of peripheral oxygen (SPO2).

[0030] The analysis device can input patient information into a prediction model (220).

[0031] The prediction model may be a model that receives patient information as input and calculates the information necessary to predict the patient's prognosis. The prediction model may be a trained model that is trained to predict the patient's prognosis based on training data. The prediction model may be a machine learning-based model. In one embodiment, the prediction model may be a deep neural network (DNN)-based model. Alternatively, the prediction model may be a multivariable logistic regression (MLR), random forest (RF), eXtreme gradient boosting (XGB), gradient boosting machine (GBM), or support vector machine (SVM)-based model.

[0032] The analysis device can predict the patient's prognosis based on the output value of the prediction model (230).

[0033] Analyzing the prognosis of patients may include analyzing the prognosis of COVID-19 patients.

[0034] Analyzing the patient's prognosis may include predicting whether the patient will develop severe symptoms. Severe symptoms may include at least one of mechanical ventilation required, extracorporeal membrane oxygenation required, admission to an intensive care unit (ICU), and patient death.

[0035] The analysis device can construct a DDR Tree (Discriminative dimensionality reduction via learning a tree) based on the patient's information and the results of predicting the patient's prognosis (240).

[0036] The DDR Tree (Discriminative Dimensionality Reduction via Learning a Tree) displays the result of reducing a patient's multidimensional characteristics into a two-dimensional space and visualizing the patient in a tree form on a two-dimensional plane based on the reduced characteristics. Each point displayed in the DDR Tree represents an individual patient.

[0037] The analysis device can reduce patient information into a two-dimensional space to construct a DDR Tree (Discriminative dimensionality reduction via learning a tree) and then reflect the results of predicting the patient's prognosis in the DDR Tree. In one embodiment, the analysis device can reflect the results of predicting the patient's prognosis in the DDR Tree by displaying them in color. In one embodiment, the analysis device may display a dark color when the patient's prognosis is poor and a light color when the patient's prognosis is good.

[0038] The analysis device can cluster patients based on the constructed DDR Tree and identify the characteristics of each cluster (250).

[0039] The characteristics of each cluster may represent the severity of the patients belonging to each cluster. For example, the analysis device can classify patients into four clusters based on a DDR Tree and then analyze the severity of the patients belonging to each cluster.

[0040] The characteristics of each cluster may include information on factors for each patient belonging to each cluster. For example, this could mean clustering patients into four groups based on a DDR Tree and then analyzing the values ​​of platelet count (PLT) and peripheral oxygen saturation (SPO2) for each cluster.

[0041] The analysis device can visualize and output the constructed DDR Tree, the results of clustering patients, and the characteristics of each cluster (260).

[0042] In one embodiment, the analysis device can visualize and output the DDR Tree, the results of clustering patients, and the characteristics of each cluster through a combination of a two-dimensional graph and colors.

[0043] Furthermore, the analysis device can also calculate the extent to which each factor included in the patient's information contributed to predicting the patient's prognosis.

[0044] The analysis device can calculate the degree to which each factor included in the patient's information contributes to predicting the patient's prognosis through a method of interpreting a machine learning model. In one embodiment, the analysis device may use one of MDIFI (mean decreases in Gini impurity feature importance), PFI (permutation feature importance), SHAP (shapley additive explanations), and BFI (built-in feature importance) to calculate the degree of contribution.

[0045] Alternatively, the analysis device can calculate the extent to which each factor included in the patient's information contributes to predicting the patient's prognosis through the constructed DDR Tree. For example, common factors among patients in a cluster containing a high number of severe patients can be extracted from the DDR Tree, and those factors can be analyzed as having a high correlation with severe symptoms. Or, by comparing a cluster containing a high number of severe patients with a cluster containing relatively few severe patients, factors showing a significant difference between the two clusters can be analyzed as having a high correlation with severity.

[0046] Furthermore, the analysis device can simply predict a patient's prognosis using the constructed DDR Tree. In other words, the analysis device inputs the patient's information into the constructed DDR Tree and can predict the patient's prognosis by identifying which group the patient belongs to.

[0048] Below, we examine an example in which a researcher collected patient information, selected valid factors, constructed a model based on the selected factors, and evaluated the performance of the constructed model.

[0050] Figure 3 shows the process of classifying collected patient information.

[0051] The collected patient information includes information about COVID-19 patients.

[0052] The COVID-19 patient data includes information collected from 9,199 patients across 19 hospitals. Among the collected data, information on 406 patients was excluded if they were diagnosed 15 days or more after admission or at least 1 day after admission. Additionally, information on 2,848 patients was excluded if variables of interest (factors) were missing. Therefore, a model was constructed based on the data of a total of 5,945 COVID-19 patients.

[0053] Information on 5,945 COVID-19 patients was broadly classified into four categories. The first is information on 747 patients at three general hospitals in the metropolitan area. The second is information on 260 patients at three general hospitals in non-metropolitan areas. The third is information on 1,385 patients at six tertiary hospitals in the metropolitan area. The fourth is information on 3,553 patients at seven tertiary hospitals in non-metropolitan areas.

[0054] Each of the four types of COVID-19 patient information was divided into development data to be used to build the model and validation data to be used to evaluate the model's performance. Through this process, the impact that the model might be affected by the type and location of the hospital was minimized.

[0055] Finally, information on 4,019 COVID-19 patients was used to build the model. Additionally, information on 1,926 COVID-19 patients was used to evaluate the performance of the built model.

[0057] Figure 4 shows candidate variables included in the classified information on COVID patients.

[0058] Candidate variables broadly include those related to patient characteristics, clinical symptoms, vital signs, and blood biochemistry.

[0059] It can be observed that the age of patients in the construction data (60s) is higher than that in the validation data (50s). The proportion of male patients (Male Sex) is similar in both the construction and validation data. Hypertension is common in both the construction and validation data. Cough and fever are the most common conditions in both datasets. With the exception of absolute neutrophil count (ANC), there are no significant differences between the construction and validation data for all vital signs and blood biochemistry.

[0061] Figure 5 shows the process of selecting the final variables to be used for model construction and performance evaluation among the candidate variables included in the information of COVID-19 patients.

[0062] To select the final variables, two machine learning models and three contribution calculation methods were used. Subsequently, the final variables were selected through a total of six machine learning-based (ML-based) feature engineering methods (FEM).

[0063] Six machine learning-based feature engineering methods (FEM) include RF-based mean decrease in Gini Impurity feature importance method (RF-MDIFI), RF-based permutation feature importance (RF-PFI), RF-based Shapley values ​​(RF-Shapley), XGM-based mean decrease in Gini Impurity feature importance method (XGM-MDIFI), XGB-based permutation feature importance (XGB-PFI), and XGB-based Shapley values ​​(XGB Shapley).

[0064] After calculating the contribution of each candidate variable in each FEM, the final variables were selected based on the following two criteria. The first criterion is that they will be included in the pre-set top K variables. The second criterion is that at least 50% of each FEM will be included in the pre-set top K variables.

[0066] A subset of 60 candidate variables was constructed using the method described above. Six machine learning models were built using each of the candidate variable subsets, and the performance of each model was evaluated.

[0067] Figure 6 shows some results of building a model using a subset of candidate variables and evaluating the performance of the built model.

[0068] A total of 6 models were constructed. The 6 models are based on DNN (deep neural network), MLR (multivariable logistic regression), RF (random forest), XGB (eXtreme gradient boosting), GBM (gradient boosting machine), and SVM (support vector machine), respectively.

[0069] Eight evaluation metrics were used to assess the performance of the model. The eight metrics are AUROC (area under receiver operating characteristic curve), Sensitivity, Specificity, PPV (positive predictive value), NPV (negative predictive value), LRP (likelihood ratio positive), LRN (likelihood ratio negative), and DOR (diagnostic odds ratio). Youden's Index was used to determine the cut-off value.

[0070] Among the six constructed models, DNN, MLR, and RF showed the best performance compared to other models. In particular, the DNN-based model showed higher predictive power than the MLR and RF-based models (AUROC, DOR). Therefore, the DNN-based prediction model was selected as the final model.

[0071] The final variables used by the DNN model may include at least one of lactate dehydrogenase (LDH) levels, age, absolute lymphocyte counts (ALC), dyspnea, respiratory rate (RR), presence or absence of diabetes mellitus (DM), C-reactive protein (CRP) levels, absolute neutrophil counts (ANC), platelet counts (PLT), white blood cell counts (WBC), and saturation of peripheral oxygen (SPO2).

[0073] Figure 7 shows the results of analyzing the extent to which variables input into the selected DNN model have a significant influence on the result value.

[0074] In order, it can be seen that lactate dehydrogenase (LDH) levels, age, absolute lymphocyte counts (ALC), dyspnea, respiratory rate (RR), presence of diabetes mellitus (DM), C-reactive protein (CRP) levels, absolute neutrophil counts (ANC), platelet counts (PLT), white blood cell counts (WBC), and saturation of peripheral oxygen (SPO2) have a significant impact on the results.

[0076] FIGS. 8 to 10 are embodiments of constructing a DDR Tree based on the results of predicting patient characteristics and prognosis, and interpreting the constructed DDR Tree.

[0077] As mentioned above, the DDR Tree (Discriminative dimensionality reduction via learning a tree) is used to reduce multidimensional features into a two-dimensional space and then visualize the characteristics of a patient reduced to two dimensions in a tree form. Therefore, in a DDR Tree Plot, each point represents an individual patient.

[0078] Figure 8 shows the results of calculating the severity probability for each patient and overlaying it onto a DDR Tree Plot. A darker color indicates a higher probability that the patient at that point will develop severe symptoms, while a lighter color indicates a lower probability.

[0079] Figures 9 and 10 show the residual values ​​of the selected features. These residual values ​​were obtained through linear regression analysis. These residual values ​​indicate how much each patient's selected feature value increased or decreased compared to the predicted value.

[0080] Figure 9 shows the results based on the DDR Tree Plot constructed in Figure 8, according to the presence or absence of factors obtained prior to the test results, such as age, diabetes mellitus (DM), and dyspnea. If the conditions for a corresponding factor were met, the color was maintained; if the conditions were not met, the color was changed to gray. For example, in the DDR Tree, patients with diabetes (With DM) maintained their original color, while all others were processed in gray. Conversely, in the DDR Tree, patients without diabetes (Without DM) maintained their original color, while all others were processed in gray.

[0081] Figure 10 shows the result of overlaying the concentrations of eight basic consecutive factors onto a DDR Tree. The darker the color, the higher the corresponding factor of the patient belonging to that point, and the lighter the color, the lower the corresponding factor of the patient belonging to that point.

[0083] Patients can be clustered into multiple groups based on the DDR Tree Plot. For example, based on the DDR Tree in Fig. 8, patients can be divided into four groups.

[0084] The four groups are the upper-right group (URG), the middle-right group (MRG), the lower-right group (LRG), and the lower-left group (LLG).

[0085] Looking at the results in Fig. 8, patients in the URG group are classified as high-risk, while patients in the MRG, LRG, and LLG groups can be classified from high-risk to low-risk. Therefore, if it is known where a patient belongs within the DDR Tree, the patient's risk level can be estimated. Thus, by receiving patient information and determining which group the patient belongs to within the DDR Tree, it is possible to analyze whether the current patient is in the high-risk or low-risk group.

[0086] By examining the results in Figures 8 to 10, the characteristics of each group can be identified.

[0087] For example, it can be observed that patients belonging to the URG group have a high respiratory rate (RR). Alternatively, it can be observed that the majority of patients in the URG and MRG groups have dyspnea, are over 60 years of age (Age > 60), and possess high respiratory rates (RR), C-reactive protein (CRP), and lactate dehydrogenase (LDH) levels. Alternatively, it can be observed that the majority of patients in the MRG group have high absolute neutrophil counts (ANC) and white blood cell counts (WBC). Alternatively, it can be observed that the majority of patients in the LRG and LLG groups have high absolute neutrophil counts (ANC), white blood cell counts (WBC), and platelet counts (PLT). Alternatively, it can be observed that the majority of patients in the LRG group have high C-reactive protein (CRP) levels and absolute lymphocyte counts (ALC).

[0088] Therefore, by identifying only specific factors from a new patient, it is possible to quickly determine whether the patient belongs to a high-risk or low-risk group.

[0089] For example, if a patient has a high respiratory rate (RR), they may belong to the URG group, allowing for the rapid identification of that patient as a high-risk group. Alternatively, if a patient has high C-reactive protein (CRP) levels and an absolute lymphocyte count (ALC), they may belong to the LRG group, enabling a somewhat lower assessment of their risk.

[0091] The analysis device is described below with reference to FIG. 11.

[0092] FIG. 11 is a configuration of one of the embodiments of an analysis device (300).

[0093] The analysis device (300) may correspond to the analysis device (100) described in FIG. 1. That is, the analysis device (300) may be a device that performs a patient prognosis prediction method using the machine learning model described above.

[0094] The analysis device (300) may include an input device (310), a storage device (320), a calculation device (330), an output device (340), an interface device (350), and a communication device (360).

[0095] The input device (310) may include an interface device (keyboard, mouse, touchscreen, etc.) for receiving certain commands or data. The input device (310) may include a configuration for receiving information through a separate storage device (USB, CD, hard disk, etc.). The input device (310) may receive input data through a separate measuring device or through a separate DB.

[0096] The input device (310) may receive data via wired or wireless communication through the communication device (360).

[0097] The input device (310) can receive information necessary for performing the patient prognosis prediction method using the aforementioned machine learning model. The input device (310) can receive a model necessary for performing the patient prognosis prediction method using the aforementioned machine learning model. The input device (310) can receive patient information. The input device (310) can receive a prediction model.

[0098] The storage device (320) may be a device for storing certain information. The storage device (320) may store information received through the input device (310). The storage device (320) may store information generated during the process of the computation device (330) performing computations. That is, the storage device (320) may include memory.

[0099] The storage device (320) can store information necessary for performing the patient prognosis prediction method using the machine learning model described above. The storage device (320) can store a model necessary for performing the patient prognosis prediction method using the machine learning model described above. The storage device (320) can store patient information. The storage device (320) can store a prediction model.

[0100] The computing device (330) may be a device such as a processor, AP, or a chip with an embedded program that processes data and performs certain operations. The computing device (330) may generate a control signal to control the analysis device (300). The computing device (330) may generate a control signal to control the input device (310), storage device (320), output device (340), interface device (350), and communication device (360) included in the analysis device (300). The computing device (330) may perform operations necessary to execute the patient prognosis prediction method using the aforementioned machine learning model.

[0101] The computing device (330) can input patient information into the prediction model. The computing device (330) can predict the patient's prognosis based on the output value of the prediction model.

[0102] The computing device (330) can construct a DDR Tree (Discriminative dimensionality reduction via learning a tree) based on the patient's information and the results of predicting the patient's prognosis. The computing device (330) can cluster patients based on the constructed DDR Tree and identify the characteristics of each cluster. The computing device (330) can calculate the degree to which each factor included in the patient's information contributed to predicting the patient's prognosis based on the constructed DDR Tree.

[0103] The output device (340) may be a device that outputs certain information. The output device (340) may output interfaces required for data processing, input data, analysis results, etc. The output device (340) may be implemented in various physical forms, such as a display, a document output device, a speaker, etc. The output device (340) may output information stored in the storage device (330). The output device (340) may output information generated during the process of computation by the computation device (330). The output device (340) may output the results of computation performed by the computation device (330). The output device (340) may visualize and output the DDR Tree constructed by the computation device, the results of clustering patients, and the characteristics of each cluster.

[0104] The interface device (350) may be a device that receives certain commands and data from the outside. The interface device (350) may receive control signals for controlling the analysis device (300). The interface device (350) may output the results analyzed by the analysis device (300). The interface device (350) may receive information necessary to perform the patient prognosis prediction method using the aforementioned machine learning model from a physically connected input device or an external storage device.

[0105] The communication device (360) may refer to a configuration that receives and transmits certain information via a wired or wireless network. The communication device (360) may perform network communication such as Wi-Fi (Wireless Fidelity), Wi-Fi Direct, Bluetooth, UWB (Ultra Wide Band), NFC (Near Field Communication), USB (Universal Serial Bus), HDMI (High Definition Multimedia Interface), LAN (Local Area Network), etc. The communication device (360) may receive control signals necessary to control the analysis device (300). The communication device (360) may transmit the results analyzed by the analysis device (300). The communication device (360) may receive information necessary to perform the patient prognosis prediction method using the aforementioned machine learning model. The communication device (360) may receive the model necessary to perform the patient prognosis prediction method using the aforementioned machine learning model.

[0107] The patient prognosis prediction method utilizing the aforementioned machine learning model can be implemented as a program (or application) including an executable algorithm that can be run on a computer.

[0108] The above program may be provided by storing it on a transitory or non-transitory computer-readable medium.

[0109] The above-mentioned temporary readable medium refers to various types of RAM such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synclink DRAM (SLDRAM), and Direct Rambus RAM (DRRAM).

[0110] The above-mentioned non-transient readable medium refers to a medium that stores data semi-permanently and can be read by a device, rather than a medium that stores data for a short moment, such as a register, cache, or memory. Specifically, the various applications or programs described above may be stored and provided on a non-transient readable medium such as a CD, DVD, hard disk, Blu-ray disc, USB, memory card, ROM (read-only memory), PROM (programmable read-only memory), EPROM (Erasable PROM, EPROM), EEPROM (Electrically EPROM), or flash memory.

[0111] The embodiments and drawings attached to this specification merely clearly illustrate a part of the technical ideas included in the aforementioned technology, and it is self-evident that variations and specific embodiments that can be easily inferred by a person skilled in the art within the scope of the technical ideas included in the specification and drawings of the aforementioned technology are all included within the scope of the rights of the aforementioned technology.

Claims

Claim 1 A step in which an analysis device receives patient information; a step in which the analysis device inputs the patient information into a prediction model; and the method includes the step of the analysis device predicting the prognosis of a patient based on the output value of the prediction model; wherein the patient information includes factors necessary for predicting the prognosis of a COVID-19 patient, and the patient information includes lactate dehydrogenase (LDH) levels, age, absolute lymphocyte counts (ALC), dyspnea, respiratory rate (RR), presence or absence of diabetes mellitus (DM), C-reactive protein (CRP) levels, absolute neutrophil counts (ANC), platelet counts (PLT), white blood cell counts (WBC), and saturation of peripheral oxygen (SPO2); the prediction model includes a deep neural network (DNN); and predicting the prognosis includes predicting the occurrence of severe symptoms in the patient, wherein the severe symptoms require mechanical ventilation A method for predicting a patient's prognosis using a machine learning model, comprising at least one of the cases where mechanical ventilation is required, extracorporeal membrane oxygenation is required, admission to an intensive care unit (ICU), and patient's death. Claim 2 A method for predicting a patient's prognosis using a machine learning model, further comprising: a step in which the analysis device constructs a DDR Tree (Discriminative dimensionality reduction via learning a tree) based on the patient's information and the result of predicting the patient's prognosis in claim 1; and a step in which the analysis device clusters patients based on the constructed DDR Tree and identifies the characteristics of each cluster. Claim 3 A method for predicting patient prognosis using a machine learning model, further comprising the step of the analysis device visualizing and outputting the constructed DDR Tree, the result of clustering patients, and the characteristics of each cluster in paragraph 2. Claim 4 A method for predicting a patient's prognosis using a machine learning model, further comprising the step of, in paragraph 3, calculating the degree to which each factor included in the patient's information contributes to predicting the patient's prognosis based on the DDR Tree constructed by the analysis device. Claim 5 delete Claim 6 delete Claim 7 An input device for receiving patient information; a computing device for inputting the patient information into a prediction model and predicting the patient's prognosis based on the output value of the prediction model; and a storage device for storing the patient's information and the prediction model; wherein the patient's information includes factors necessary for predicting the prognosis of a COVID-19 patient, and the patient's information includes lactate dehydrogenase (LDH) levels, age, absolute lymphocyte counts (ALC), dyspnea, respiratory rate (RR), presence or absence of diabetes mellitus (DM), C-reactive protein (CRP) levels, absolute neutrophil counts (ANC), platelet counts (PLT), white blood cell counts (WBC), and saturation of peripheral oxygen (SPO2); the prediction model includes a deep neural network (DNN); and predicting the prognosis includes predicting the occurrence of severe symptoms in the patient, wherein the severe symptoms are cases requiring mechanical ventilation A patient prognosis prediction device utilizing a machine learning model, comprising at least one of the cases where extracorporeal membrane oxygenation is required, admission to an intensive care unit (ICU), and patient's death. Claim 8 In claim 7, the computing device constructs a DDR Tree (Discriminative dimensionality reduction via learning a tree) based on the patient's information and the result of predicting the patient's prognosis, clusters patients based on the constructed DDR Tree, and identifies the characteristics of each cluster, a patient prognosis prediction device utilizing a machine learning model. Claim 9 A patient prognosis prediction device utilizing a machine learning model, further comprising, in claim 8, an output device that visualizes and outputs the DDR Tree constructed above, the result of clustering patients, and the characteristics of each cluster. Claim 10 In claim 9, the above-mentioned computing device is a patient prognosis prediction device utilizing a machine learning model, which calculates the degree to which each factor included in the patient's information contributes to predicting the patient's prognosis based on the constructed DDR Tree. Claim 11 delete Claim 12 delete Claim 13 A computer-readable recording medium storing a program for executing a method for predicting patient prognosis using a machine learning model described in paragraph 1.

Images