Disease a or b diagnosis assistance device for subject suspected of having disease a or b, trained model for diagnosis assistance, diagnosis assistance method, diagnosis assistance program, and diagnosis assistance data structure

Abstract

To provide a diagnosis assistance device for diagnosing which of two diseases a subject is suffering from, a trained model for diagnosis assistance, a diagnosis assistance method, a diagnosis assistance program, and a diagnosis assistance data structure.SOLUTION: A diagnosis assistance device provided herein is configured to use a plurality of sets of biometric data of patients with bilateral aldosteronism (disease A) and a plurality of sets of biometric data of patients with unilateral primary aldosteronism (disease B) in a database as learning data to generate a plurality of explanatory variables for learning from the plurality of sets of biometric data for learning, and then predict (diagnosis assistance) or determine which of the disease A and the disease B a subject suspected of having the disease A or disease B is suffering from.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a diagnostic support device, a trained model for diagnostic support, a diagnostic support method, a diagnostic support program, and a data structure for diagnostic support for disease A or disease B in a subject suspected of having disease A or disease B. [Background technology]

[0002] Primary aldosteronism (PA) is a hypertensive disorder that is more likely to be associated with cardiovascular disease than essential hypertension, and is estimated to affect approximately 3 million Japanese people. Approximately half of these patients have a subtype called aldosterone-producing adenoma (APA), which can be cured by adrenalectomy of the affected side. However, until now, diagnosing the disease type required screening tests, multiple confirmatory tests, CT scans, and adrenal vein catheterization, and the complexity of this diagnostic method meant that many cases were unable to reach a diagnosis and receive appropriate treatment.

[0003] Non-Patent Document 1 discloses an artificial intelligence system for predicting primary aldosteronism subtypes. However, compared to the present system, this system 1) uses different markers and 2) requires adrenal CT scan results. [Prior art documents] [Non-patent literature]

[0004] [Non-Patent Document 1] Artificial Intelligence Systems for Predicting Primary Aldosteronism Subtype. Journal of the Endocrine Society,Volume 3, Issue Supplement_1, April-May 2019, SUN-367 Summary of the Invention [Problem to be solved by the invention]

[0005] The present invention aims to provide a diagnostic support device, a diagnostic support trained model, a diagnostic support method, a diagnostic support program, and a diagnostic support data structure for diagnosing either of two diseases (particularly, bilateral aldosteronism or unilateral primary aldosteronism). [Means for solving the problem]

[0006] In order to solve the above-mentioned problems, in the present invention, multiple biometric data of patients with bilateral aldosteronism (disease A) and multiple biometric data of patients with unilateral primary aldosteronism (disease B) from a database of nearly 5,000 cases of adrenal diseases are used as training data, multiple learning explanatory variables are generated from the multiple learning biometric data, and a diagnostic support device, a diagnostic support trained model, a diagnostic support method, a diagnostic support program, and a diagnostic support data structure are constructed that can predict (diagnostic support) or determine whether a subject suspected of having disease A or disease B has either disease A or disease B. Furthermore, the present invention was completed by confirming that the diagnostic support device has high detection sensitivity and / or high specificity in assisting in the specific diagnosis of bilateral aldosteronism or unilateral primary aldosteronism without using measurement data from adrenal venous sampling (AVS) and adrenal CT scans.

[0007] That is, the present invention is as follows. 1. A diagnostic support device for disease A or disease B in a subject suspected of having disease A or disease B, a learning data storage unit that stores a plurality of biometric data sets of patients with disease A and a plurality of biometric data sets of patients with disease B as learning data; a model construction unit that constructs a learning model using a plurality of biometric data groups of patients with disease A and a plurality of biometric data groups of patients with disease B for learning, to predict whether the patient has disease A or disease B from explanatory variables including the biometric data groups; a prediction data receiving unit that receives one or more biometric data of a subject suspected of having disease A or disease B as prediction data; and a prediction unit that predicts or determines whether the subject has disease A or disease B by applying the biometric data of the prediction data accepted by the prediction data acceptance unit to the learning model constructed by the model construction unit; A diagnostic support device comprising: 2. The apparatus further includes a missing value complementation unit for training, which, when at least one piece of biometric data is missing from the group of biometric data for training stored in the training data storage unit, estimates and complements the missing piece of biometric data from other existing biometric data; 2. The diagnostic support device according to claim 1, wherein the model construction unit is a model construction unit that constructs the learning model using a plurality of biometric data sets of patients with disease A and a plurality of biometric data sets of patients with disease B that have been complemented by the learning missing value complementation unit. 3. Further comprising an imbalance data resolution unit, when an imbalance occurs between the plurality of biometric data groups of patients with disease A and the plurality of biometric data groups of patients with disease B for the learning biometric data groups stored in the learning data storage unit, 3. The diagnostic support device according to claim 1 or 2, wherein the model construction unit is a model construction unit that constructs the learning model using biometric data sets in which imbalances between the plurality of biometric data sets of the patient with disease A and the plurality of biometric data sets of the patient with disease B have been eliminated, or biometric data sets in which imbalances between the plurality of biometric data sets of the patient with disease A and the plurality of biometric data sets of the patient with disease B have been eliminated, which have been complemented by the missing value complementation unit for learning. 4. The diagnostic support device according to any one of the preceding paragraphs 1 to 3, wherein the model construction unit is constructed from a plurality of learning models. 5. A diagnostic support device according to the preceding paragraph 4, wherein the learning model is constructed from one or more of the following: 1) Random Forest 2) K - Nearest Neighbor 3) Multi-Layer Perceptron 4) Support Vector Machine 5) Logistic Regression 6) Naive Bayes 7) LightGBM 6. The diagnostic support device according to any one of items 1 to 5 above, wherein the combination of disease A and disease B is bilateral aldosteronism and unilateral primary aldosteronism. 7. The diagnostic support device according to claim 6, wherein the biometric data is a combination of any one of the following: 1) Serum K value before potassium supplementation, serum K value, K preparation dose, ARR, PAC, 2) Serum K level before potassium supplementation, serum K level, K preparation dose, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value and serum K value before potassium supplementation 5) Serum K level before potassium supplementation 8. The diagnostic support device according to claim 6 or 7, wherein the biometric data does not include adrenal venous sampling (AVS) and / or adrenal CT scans. 9. A trained model for diagnostic support of disease A or disease B in a subject suspected of having disease A or disease B, the trained model including: 1) A plurality of biometric data sets of patients with disease A and a plurality of biometric data sets of patients with disease B as training data; and 2) A trained model for predicting or determining whether a patient has either disease A or disease B from explanatory variables including a plurality of biometric data groups of patients with disease A and a plurality of biometric data groups of patients with disease B, using machine learning for the training. 10. The combination of disease A and disease B is bilateral aldosteronism and unilateral primary aldosteronism, and the biometric data is any one of the following combinations: The trained model described in Section 9 above. 1) Serum K value before potassium supplementation, serum K value, K preparation dose, ARR, PAC, 2) Serum K level before potassium supplementation, serum K level, K preparation dose, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value and serum K value before potassium supplementation 5) Serum K level before potassium supplementation 11. A diagnostic support method for disease A or disease B in a subject suspected of having disease A or disease B, the method comprising the following steps: 1) A trained model generation process for generating a trained model capable of determining whether disease A or disease B is present by performing machine learning using a plurality of biometric data sets of patients with disease A and a plurality of biometric data sets of patients with disease B. 2) A process of predicting or determining whether a subject has either disease A or disease B by applying one or more biometric data of the subject suspected of having disease A or disease B to the trained model. 12. The combination of disease A and disease B is bilateral aldosteronism and unilateral primary aldosteronism, and the biometric data is any one of the following combinations: 12. The diagnostic support method according to item 11 above. 1) Serum K value before potassium supplementation, serum K value, K preparation dose, ARR, PAC, 2) Serum K level before potassium supplementation, serum K level, K preparation dose, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value and serum K value before potassium supplementation 5) Serum K level before potassium supplementation 13. A diagnostic support program for disease A or disease B in a subject suspected of having disease A or disease B, the program comprising the following steps: 1) A trained model generation process for generating a trained model capable of determining whether disease A or disease B is present by performing machine learning using a plurality of biometric data sets of patients with disease A and a plurality of biometric data sets of patients with disease B. 2) A process of predicting or determining whether a subject has either disease A or disease B by applying one or more biometric data of the subject suspected of having disease A or disease B to the trained model. 14. The combination of disease A and disease B is bilateral aldosteronism and unilateral primary aldosteronism, and the biometric data is any one of the following combinations: 14. The diagnostic assistance program according to the preceding paragraph 13. 1) Serum K value before potassium supplementation, serum K value, K preparation dose, ARR, PAC, 2) Serum K level before potassium supplementation, serum K level, K preparation dose, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value and serum K value before potassium supplementation 5) Serum K level before potassium supplementation 15. A data structure for supporting the diagnosis of disease A or disease B in a subject suspected of having disease A or disease B, the data structure including: 1) A plurality of biometric data sets of patients with disease A and a plurality of biometric data sets of patients with disease B as training data; and 2) Data of a trained model for predicting or determining whether a patient has either disease A or disease B from explanatory variables including multiple biometric data groups of patients with disease A and multiple biometric data groups of patients with disease B used for training. 16. The combination of disease A and disease B is bilateral aldosteronism and unilateral primary aldosteronism, and the biometric data is any one of the following combinations: 16. The diagnostic support data structure according to the preceding paragraph 15. 1) Serum K value before potassium supplementation, serum K value, K preparation dose, ARR, PAC, 2) Serum K level before potassium supplementation, serum K level, K preparation dose, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value and serum K value before potassium supplementation 5) Serum K level before potassium supplementation [Effects of the Invention]

[0008] The present invention provides a diagnostic support device, a trained model for diagnostic support, a diagnostic support method, a diagnostic support program, and a diagnostic support data structure for assisting in the diagnosis of either of two diseases (particularly, bilateral aldosteronism or unilateral primary aldosteronism) with high detection sensitivity and / or specificity. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a block diagram showing an example of the functional configuration of a diagnosis support device according to a first embodiment. [Figure 2] FIG. 10 is a block diagram illustrating an example of the functional configuration of a diagnosis support device according to a second embodiment. [Figure 3] FIG. 10 is a block diagram illustrating an example of the functional configuration of a diagnosis support device according to a third embodiment. [Figure 4] FIG. 10 is a block diagram showing an example of the functional configuration of a diagnosis support device according to a fourth embodiment. [Figure 5] Selection of optimal biometric data for each training model. (A) Selection of biometric data for the screening dataset and (B) selection of biometric data for the confirmatory test dataset. [Figure 6] Receiver operating characteristic curves for predictive diagnosis of aldosterone-producing adenomas. Receiver operating characteristic curves for the model using (A) the screening dataset, (B) the screening dataset with the five most important variables, and (C) the confirmatory test dataset. DETAILED DESCRIPTION OF THE INVENTION

[0010] (Subject of the present invention) The present invention relates to a diagnostic support device, a trained model for diagnostic support, a diagnostic support method, a diagnostic support program, and a data structure for diagnostic support, which are used to assist in diagnosing two diseases (particularly, bilateral aldosteronism or unilateral primary aldosteronism).

[0011] (First embodiment) A first embodiment of the present invention will be described below with reference to the drawings. Fig. 1 is a block diagram showing an example of the functional configuration of a diagnosis support device 1 for disease A or disease B in a subject suspected of having disease A or disease B according to the first embodiment. As shown in Fig. 1, the diagnosis support device 1A according to the first embodiment includes components (functional blocks) of a learning data storage unit 2, a model construction unit 3, a trained model 4, a prediction data acceptance unit 5, and a prediction unit 6. For convenience of explanation, the components are separated, but the components may be combined or may include other components. The same applies to the other examples below. The above configuration can be configured by either hardware or software. For example, when configured by software, each of the above functional blocks is actually configured with a computer's CPU, RAM, ROM, etc., and is realized by the operation of a program stored in a recording medium such as RAM, ROM, a hard disk, or a semiconductor memory. Furthermore, each configuration does not need to be mounted on the same hardware, and part of the configuration may be mounted on another medium (e.g., cloud). The same applies to the other embodiments described below.

[0012] Learning data storage unit 2 The learning data storage unit 2 receives a plurality of biometric data sets of patients with disease A and a plurality of biometric data sets of patients with disease B as learning data. A portion of the training data (for example, about 20%) is divided as validation data. Hereinafter, the training data will also include the validation data. The division method is not particularly limited, but methods such as holdout, cross-validation, k-fold cross-validation, and leave-one-out can be used. These data can be obtained, for example, from known disease databases (e.g., the disease registry database of the Japan Research Association for Acute Adrenal Diseases (JRAS)).

[0013] Biometric data The biometric data is any data obtained from a patient with disease A and / or a patient with disease B. For example, the data in Table 1 below can be exemplified.

[0014] [Table 1]

[0015] The abbreviations in Table 1 above are as follows: The ATC / DDD index for antihypertensive drugs was calculated according to the Anatomical Therapeutic Chemical / Daily Defined Dose Index 2020. ARR: aldosterone-to-renin ratio (aldosterone concentration / renin activity ratio), BMI: body mass index, BUN: blood urea nitrogen, CCT: captopril tolerance test, CKD: chronic kidney disease, DBP: diastolic blood pressure, eGFR: estimated glomerular filtration rate, FBS: fasting blood glucose, FUT: furosemide orthostatic tolerance test, IHD: ischemic heart disease, NGSP: National Glycohemoglobin Standardization Program, PA: primary aldosteronism, PAC: plasma aldosterone level, PRA: plasma renin activity, SAS: sleep apnea syndrome, SBP: systolic blood pressure, s-Cl: serum chloride level, sK: serum potassium level, s-Na: serum sodium level, s-UA: serum uric acid level.

[0016] Model Construction Section 3 The model construction unit 3 generates multiple learning explanatory variables as needed from multiple learning biometric data of patients with disease A and multiple learning biometric data of patients with disease B received by the learning data storage unit 2. That is, the multiple biometric data are synthesized or converted to generate new explanatory variables. Methods for creating new explanatory variables include some or all of those generated by feature engineering such as standardization, factor analysis, simple regression, residual calculation using multiple regression, principal component analysis, arithmetic operations, mean, variance, and standard deviation. The model construction unit 3 uses multiple biometric data of patients with disease A and multiple biometric data of patients with disease B for training received by the training data storage unit 2, and, if necessary, the generated explanatory variables (hereinafter, the training explanatory variables may also be referred to as training data), to construct a learning model for deriving diagnostic assistance for disease A or disease B from the biometric data. Furthermore, the constructed learning model is stored in the learned model 4.

[0017] The model construction unit 3 constructs a learning model by applying known machine learning techniques using the above-mentioned learning data. For example, Random Forest, K-Nearest Neighbor, Multi-Layer Perceptron, Support Vector Machine, Logistic Regression, Naive Bayes, LightGBM, etc. can be used.

[0018] 〇Trained model 4 The trained model 4 stores and updates the trained model constructed by the model construction unit 3. The trained model 4 also has a function of updating the trained model by repeating training by the model construction unit 3.

[0019] Prediction data reception unit 5 The prediction data receiving unit 5 inputs biometric data of a subject (a test subject suspected of having disease A or disease B) as prediction data. Preferably, the subject is primarily a subject from whom biometric data has been obtained but who has not yet been diagnosed (definitively diagnosed) as having either disease A or disease B.

[0020] Prediction Section 6 The prediction unit 6 generates, as necessary, a plurality of explanatory variables for prediction from the prediction data received by the prediction data receiving unit 5. The explanatory variables can be generated using the same method as used in the model building unit 3. The prediction unit 6 predicts (assists in diagnosis) whether the subject has disease A or disease B by applying the biometric data and multiple explanatory variables for prediction received by the prediction data receiving unit 5 to the learning model stored in the learned model 4. Furthermore, the prediction unit 6 outputs the prediction results such as AUC (Area Under the Curve), detection sensitivity, specificity, PPV (Positive Predictive Value), NPV (Negative Predictive Value), F value, and accuracy rate as prediction results.

[0021] According to the first embodiment, a learning model constructed based on multiple biometric data of patients with disease A and multiple biometric data of patients with disease B can be used to predict (assist in diagnosis) whether a subject has disease A or disease B based on the biometric data of the subject.

[0022] (Second embodiment) A second embodiment of the present invention will be described with reference to Fig. 2. Fig. 2 is a block diagram showing an example of the functional configuration of a diagnostic support device 1B for disease A or disease B in a subject suspected of having disease A or disease B according to the second embodiment. In Fig. 2, components with the same reference numerals as those shown in Fig. 1 have the same functions, and therefore redundant explanations will be omitted here.

[0023] Missing value imputation unit for learning 7 The second embodiment includes a missing value complement unit for learning 7 in addition to the first embodiment. When at least one biometric measurement value is missing from the training biometric data received by the training data storage unit 2, the training missing value complementation unit 7 estimates and complements the missing biometric measurement value from the other existing biometric measurement values. That is, when there is a missing value in the training biometric data, the training missing value complementation unit 7 complements the missing value by estimating it through machine learning using other existing biometric measurements. For example, Random Forest (MissForest) can be used as the applied machine learning.

[0024] Missing value imputation for prediction 8 The system may also have a prediction missing value completion unit 7 that, when at least one biometric measurement value is missing from the biometric data of a subject (a subject suspected of having disease A or disease B), estimates and completes the missing biometric measurement value from other existing biometric measurements.

[0025] According to the second embodiment, if there are missing values ​​in the learning data, these can be complemented, and then explanatory variables can be generated, a learning model can be constructed, and disease diagnosis assistance for the subject can be performed.

[0026] (Third embodiment) A third embodiment of the present invention will be described with reference to Fig. 3. Fig. 3 is a block diagram showing an example of the functional configuration of a diagnostic support device 1C for disease A or disease B in a subject suspected of having disease A or disease B according to the third embodiment. In Fig. 3, components with the same reference numerals as those shown in Figs. 1 and 2 have the same functions, and therefore redundant explanations will be omitted here.

[0027] Imbalanced Data Resolution Unit 9 The third embodiment includes an imbalance data resolution unit 9 in addition to the first and second embodiments. The imbalance data resolving unit 9 resolves the imbalance when it occurs in the learning biometric data received by the learning data storage unit 2. That is, when there is an imbalance in the learning biometric data, the imbalance data eliminating unit 9 eliminates the imbalance by employing oversampling (for example, SMOTE), undersampling, or the like.

[0028] According to the third embodiment, if there is an imbalance in the learning data, this can be resolved, and then explanatory variables can be generated, a learning model can be constructed, and disease diagnosis assistance for the subject can be performed.

[0029] (Fourth embodiment) A fourth embodiment of the present invention will be described with reference to Fig. 4. Fig. 4 is a block diagram showing an example of the functional configuration of a diagnosis support device 1D for disease A or disease B in a subject suspected of having disease A or disease B according to the fourth embodiment. In Fig. 4, components assigned the same reference numerals as those shown in Figs. 1, 2, and 3 have the same functions, and therefore redundant explanations will be omitted here.

[0030] In the fourth embodiment, a plurality of learning models (ensemble learning models) are constructed in the model construction unit 3. Examples of the plurality of learning models include 1) Random Forest, 2) K-Nearest Neighbor, 3) Multi-Layer Perceptron, 4) Support Vector Machine, 5) Logistic Regression, 6) Naive Bayes, and 7) LightGBM. The following is an example of a method for calculating the final prediction result by combining the results of each model. 1) The predicted probabilities are averaged and classified as either Disease A or Disease B using a threshold of 0.5. The results of each learning model are added together with the same weight and averaged. 2) Divide into main learning model and sub learning model, add individual weights to each learning model, and evaluate with a total score. Although it is preferable to use the same markers (biometric data) in each learning model, it is also possible to set optimal markers for each learning model. For example, multiple markers selected in a learning model using random forests can be used in other learning models.

[0031] According to the fourth embodiment, prediction is made by integrating the results of a plurality of learning models, so that highly accurate disease diagnosis assistance for a subject can be provided.

[0032] (How to set the importance of a marker) The method for setting the importance of a marker can be a known method, and the following can be exemplified. When using a random forest, the importance of markers can be set using the Gini coefficient. The smaller the Gini coefficient, the more accurately the data can be divided. Specifically, by calculating how much the Gini coefficient drops depending on which marker is used to divide the data when building a decision tree, the more the Gini coefficient drops, the more important the marker can be determined to be.

[0033] (hyperparameter tuning) In the learning model used in this embodiment, hyperparameters can be adjusted (tuned) by known methods such as grid search, random search, Bayesian optimization, etc. For example, tuning by Bayesian optimization can be performed as follows. 1) Random Forest: Number and depth of decision trees 2) K-Nearest Neighbor: The number of neighboring samples to consider when determining the label. 3) Multi-Layer Perceptron: Number of hidden layers, number of neurons, regularization parameters 4) Support Vector Machine: Regularization strength, RBF kernel coefficients 5) Logistic Regression: Strength of regularization 7) LightGBM: Regularization strength, number and depth of decision trees

[0034] (Combination of disease A and disease B) The combination of disease A and disease B in this specification is primarily intended for cases where it is doubtful whether a subject (test subject) has disease A or disease B. In this example, the subject was diagnosed as a primary aldosteronism patient, and bilateral aldosteronism (disease A) and unilateral primary aldosteronism (disease B) are exemplified, but are not particularly limited.

[0035] (Specific diagnostic aid for bilateral aldosteronism or unilateral primary aldosteronism) The following example confirms that specific diagnostic assistance for bilateral aldosteronism or unilateral primary aldosteronism can be achieved using the following biometric data. 1) Serum K value before potassium supplementation, serum K value, K preparation dose, ARR, PAC, 2) Serum K level before potassium supplementation, serum K level, K preparation dose, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value and serum K value before potassium supplementation 5) Serum K level before potassium supplementation The serum K value before supplementation with a potassium preparation (particularly a potassium preparation for treating hypokalemia caused by the treatment of primary aldosteronism) (untreated serum K value) means, for example, the serum K value before treatment of primary aldosteronism. The dose of potassium preparation means, for example, the dose of potassium preparation used in the treatment of primary aldosteronism. If potassium preparation is not administered, the value is 0. The serum K value may be measured at any time, and may be measured when a disease is suspected (for example, when aldosteronism is diagnosed), when a biopsy is taken (at screening), or after potassium supplementation. The PAC (plasma aldosterone level) may be measured at any time without limitation, and may be measured when a disease is suspected (for example, when aldosteronism is diagnosed), when a biopsy is taken (at screening), or after potassium supplementation. Preferably, the PAC is measured when oral medications that affect renin or aldosterone are discontinued or changed, in accordance with the hypertension treatment guidelines of the Japanese Society of Hypertension. The ARR (aldosterone concentration / renin activity ratio) is not limited to a specific time period, and may be a measurement taken when a disease is suspected (for example, a measurement taken at the time of diagnosis of aldosteronism), a measurement taken at the time of biopsy (at screening), or a measurement taken after potassium supplementation. Preferably, the measurement taken when oral medications that affect renin or aldosterone are discontinued or changed is used, in accordance with the hypertension treatment guidelines of the Japanese Society of Hypertension. The above measurement data does not include measurement data from adrenal vein sampling and adrenal CT scans, which have traditionally been used for definitive diagnosis.

[0036] (A trained model for supporting the diagnosis of disease A or disease B in subjects suspected of having disease A or disease B) A trained model for diagnostic support of disease A or disease B in a subject suspected of having disease A or disease B includes the following: 1) As training data, a group of multiple biometric data sets of patients with disease A and a group of multiple biometric data sets of patients with disease B 2) A trained model for predicting whether a patient has disease A or disease B from explanatory variables including a plurality of biometric data groups of patients with disease A and a plurality of biometric data groups of patients with disease B, using machine learning to train the trained model. In a specific embodiment of the trained model, the combination of disease A and disease B is bilateral aldosteronism and unilateral primary aldosteronism, and the biometric data is any one of the following combinations: 1) Serum K value before potassium supplementation, serum K value, K preparation dose, ARR, PAC, 2) Serum K level before potassium supplementation, serum K level, K preparation dose, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value and serum K value before potassium supplementation 5) Serum K level before potassium supplementation

[0037] (Method for supporting diagnosis of disease A or disease B in subjects suspected of having disease A or disease B) A method for supporting the diagnosis of disease A or disease B in a subject suspected of having disease A or disease B includes the following steps. 1) A trained model generation process for generating a trained model capable of determining whether disease A or disease B is present by performing machine learning using a plurality of biometric data sets of patients with disease A and a plurality of biometric data sets of patients with disease B. 2) predicting whether a subject has either disease A or disease B by applying one or more biometric data of the subject suspected of having disease A or disease B to the trained model; In a specific embodiment of the diagnostic assistance method, the combination of disease A and disease B is bilateral aldosteronism and unilateral primary aldosteronism, and the biometric data is any one of the following combinations: 1) Serum K value before potassium supplementation, serum K value, K preparation dose, ARR, PAC, 2) Serum K level before potassium supplementation, serum K level, K preparation dose, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value and serum K value before potassium supplementation 5) Serum K level before potassium supplementation

[0038] (Diagnosis support program for disease A or disease B in subjects suspected of disease A or disease B) A diagnostic support program for disease A or disease B in a subject suspected of having disease A or disease B includes the following steps. 1) A trained model generation process for generating a trained model capable of determining whether disease A or disease B is present by performing machine learning using a plurality of biometric data sets of patients with disease A and a plurality of biometric data sets of patients with disease B. 2) A process of predicting whether a subject has either disease A or disease B by applying one or more biometric data of the subject suspected of having disease A or disease B to the trained model. In a specific embodiment of the diagnostic assistance program, the combination of disease A and disease B is bilateral aldosteronism and unilateral primary aldosteronism, and the biometric data is any one of the following combinations: 1) Serum K value before potassium supplementation, serum K value, K preparation dose, ARR, PAC, 2) Serum K level before potassium supplementation, serum K level, K preparation dose, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value and serum K value before potassium supplementation 5) Serum K level before potassium supplementation

[0039] (Data structure for supporting diagnosis of disease A or disease B for subjects suspected of disease A or disease B) A data structure for supporting the diagnosis of disease A or disease B in a subject suspected of having disease A or disease B includes the following: 1) As training data, a group of multiple biometric data sets of patients with disease A and a group of multiple biometric data sets of patients with disease B 2) Data of a trained model for predicting whether a patient has disease A or disease B from explanatory variables including a plurality of biometric data groups of patients with disease A and a plurality of biometric data groups of patients with disease B used for training. In a specific embodiment of the diagnostic support data structure, the combination of disease A and disease B is bilateral aldosteronism and unilateral primary aldosteronism, and the biometric data is any one of the following combinations: 1) Serum K value before potassium supplementation, serum K value, K preparation dose, ARR, PAC, 2) Serum K level before potassium supplementation, serum K level, K preparation dose, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value and serum K value before potassium supplementation 5) Serum K level before potassium supplementation

[0040] The present invention will be described below with reference to examples, but the present invention is not limited to these examples. The following examples have been approved by the ethics committee of the National Hospital Organization Kyoto Medical Center (lead institution) and the ethics committees of JRAS participating institutions. [Example]

[0041] In this example, a high-precision ensemble learning model was constructed to assist in diagnosing which of the two diseases a patient has, using biometric data from a patient who has been definitively diagnosed with bilateral aldosteronism and a patient who has been definitively diagnosed with unilateral primary aldosteronism as training data.

[0042] (Patient data used) Male or female patients aged 20 to 90 years who were diagnosed with primary aldosteronism (PA) by adrenal venous sampling (AVS) between January 2006 and December 2018 were enrolled in JRAS. Clinical characteristics, biochemical findings, and confirmatory test results were collected electronically using an online registration system. System construction, data security, and maintenance of the registry data were outsourced to EPS Corporation. This example was performed using a dataset validated in March 2020. Patients with data including clinical characteristics, biochemistry, and AVS were included. PA was diagnosed according to the guidelines of the Japan Endocrine Society and the Japanese Society of Hypertension.

[0043] (Training dataset used) The JRAS database includes PA diagnostic criteria: physical findings; medical history; electrolyte, renal function, glucose metabolism, and lipid metabolism parameters; aldosterone-to-renin ratio (ARR) at screening; and confirmatory test results.

[0044] (statistical analysis) Data were expressed as mean ± standard deviation (SD) or percentage. Differences between APA and BAH were analyzed. The Shapiro-Wilk test was used to test for normality. If normality was rejected, the Mann-Whitney U test was used. If normality was accepted, the Bartlett test was used to test for equality of variances. Student's t-test was used when equal variances were found, and Welch's t-test was used when equal variances were not found. P<0.05 was considered statistically significant.

[0045] (Building a learning model) The algorithm development process includes the general processes of 1) data creation and 2) model building and evaluation. A total of seven machine learning programs (Random Forest, K-Nearest Neighbor, Multi-Layer Perceptron, Support Vector Machine, Logistic Regression, Naive Bayes, and LightGBM) were used. The six algorithms other than LightGBM were developed by imputing missing values ​​in the database using MissForest. These seven machine learning programs were combined into a single high-precision ensemble learning model.

[0046] (result) 1) Patient characteristics A total of 4,057 PA patients were enrolled in the JRAS database. Based on the inclusion and exclusion criteria, data from 545 and 1,352 APA and BAH patients, respectively, were used in this study. Table 2 below shows the baseline characteristics and percentage of missing data for APA and BAH patients.

[0047] [Table 2]

[0048] The data in Table 2 above are presented as mean (SD) or percentage. The ATC / DDD index of antihypertensive drugs was calculated according to the Anatomical Therapeutic Chemical / Daily Defined Dose Index 2020. *, p<0.05 vs. BAH; **, p<0.01 vs. BAH; ***, p<0.001 vs. BAH. APA: aldosterone-producing adenoma, ARR: aldosterone-to-renin ratio, BAH: bilateral adrenal hyperplasia, BMI: body mass index, BUN: blood urea nitrogen, CCT: captopril tolerance test, CKD: chronic kidney disease, DBP: diastolic blood pressure, eGFR: estimated glomerular filtration rate, FBS: fasting blood glucose, FUT: furosemide orthostatic tolerance test, IHD: ischemic heart disease, NGSP: National Glycohemoglobin Standardization Program, PA: primary aldosteronism, PAC: plasma aldosterone level, PRA: plasma renin activity, SAS: sleep apnea syndrome, SBP: systolic blood pressure, s-Cl: serum chloride, sK: serum potassium, s-Na: serum sodium, s-UA: serum uric acid.

[0049] 2) Feature Importance Analysis Table 3 below shows the feature importance rankings and importance scores in the RF (Random Forest) model for the screening and confirmatory test datasets. The five most important variables (explanatory variables) for the screening test dataset were serum potassium level at the first visit, serum potassium level at the time of supplementation, PAC, ARR, and potassium supplement dose, while the five most important variables (explanatory variables) for the confirmatory test dataset were serum potassium level at the first visit, PAC at 60 and 90 minutes in the captopril loading test, PAC at 0 minutes in the furosemide orthostatic loading test, and serum potassium level at the time of supplementation. Figure 5 shows the relationship between the number of variables and the average AUC for the seven MLAs in the two datasets. In the screening dataset, RF, MLP, LightGBM, SVM, LR, KNN, and NB showed the best average AUC with 19, 12, 13, 12, 12, 7, and 9 variables, respectively. In the validation dataset, RF, MLP, LightGBM, SVM, LR, KNN, and NB showed the best average AUC with 23, 19, 25, 21, 21, 7, and 10 variables, respectively. Therefore, we selected the number of variables obtained in the validation test dataset to develop the best performing predictive model.

[0050] [Table 3]

[0051] 3) Results of ensemble learning Three ensemble learning models were developed by combining seven MLAs and three datasets (screening data, data on five important variables, and validation data). Figure 6 shows the median receiver operating characteristic curves for the APA prediction model using the screening test dataset, the screening dataset with the five most important variables, and the confirmatory test dataset. The AUC for the screening test dataset was 0.898 ± 0.040, and the AUC value for the five most important variables model was 0.886 ± 0.044 (see Table 4). The model built on the validation dataset performed better than those on the other two datasets, with an AUC value of 0.914 ± 0.041. The predictive performance of only the seven MLAs is shown in Table 5. The highest average AUC was 0.914 ± 0.042 for RF. Based on the above, we constructed an ensemble learning model with high accuracy and specificity to assist in the specific diagnosis of two diseases. In particular, the specific diagnostic aid for bilateral aldosteronism or unilateral primary aldosteronism can be used without the use of measurement data from adrenal venous sampling (AVS) and adrenal CT scans. It was confirmed that it has high detection sensitivity and / or high specificity.

[0052] [Table 4]

[0053] [Table 5] [Example]

[0054] In this example, the following highly important markers (measurement data) obtained in Example 1 were used to make predictions using the highly accurate ensemble learning model obtained in Example 1. 1) Serum K value before potassium supplementation, serum K value, K preparation dose, ARR, PAC, 2) Serum K level before potassium supplementation, serum K level, K preparation dose, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value and serum K value before potassium supplementation 5) Serum K level before potassium supplementation

[0055] (result) The prediction results are shown in Table 6 below. The highly accurate ensemble learning model obtained in this Example 1 can make predictions with high accuracy and specificity not only using the five markers but also using a portion of the five markers.

[0056] [Table 6] [Industrial Applicability]

[0057] The present invention makes it possible to provide a diagnostic support device, a diagnostic support trained model, a diagnostic support method, a diagnostic support program, and a diagnostic support data structure for assisting in the diagnosis of either of two diseases (particularly, bilateral aldosteronism or unilateral primary aldosteronism) with high detection sensitivity and / or specificity.

Claims

1. A diagnostic support device for disease A or disease B in a subject suspected of disease A or disease B, a learning data storage unit that stores a plurality of biometric data groups of patients with disease A and a plurality of biometric data groups of patients with disease B as learning data; A model construction unit that constructs a learning model for predicting whether a patient has disease A or disease B from explanatory variables including a plurality of biometric data groups of patients with disease A and a plurality of biometric data groups of patients with disease B for learning; A prediction data receiving unit that receives one or more pieces of biometric data of a subject suspected of having disease A or disease B as prediction data; and a prediction unit that predicts or determines whether the subject has disease A or disease B by applying the biometric data of the prediction data accepted by the prediction data accepting unit to the learning model constructed by the model constructing unit; A diagnosis support device comprising: Wherein, the combination of the disease A and the disease B is bilateral aldosteronism and unilateral primary aldosteronism; Diagnostic Support Devices

2. a missing value complementation unit for learning that, when at least one piece of biometric data is missing from the group of biometric data for learning stored in the learning data storage unit, estimates and complements the missing piece of biometric data from other existing biometric data; 2. The diagnostic support device according to claim 1, wherein the model construction unit constructs the learning model using a plurality of biometric data groups of patients with disease A and a plurality of biometric data groups of patients with disease B that have been complemented by the learning missing value complementation unit.

3. The method further includes an imbalance data resolution unit when an imbalance occurs between the plurality of biometric data groups of a patient with disease A and the plurality of biometric data groups of a patient with disease B for the learning biometric data groups stored in the learning data storage unit, 3. The diagnostic support device according to claim 1, wherein the model construction unit is a model construction unit that constructs the learning model using a biometric data group in which imbalance between the multiple biometric data groups of the patient with disease A and the multiple biometric data groups of the patient with disease B is eliminated, or a biometric data group in which imbalance between the multiple biometric data groups of the patient with disease A and the multiple biometric data groups of the patient with disease B is eliminated, the biometric data group being complemented by the learning missing value complementation unit.

4. The diagnosis support device according to claim 1 , wherein the model construction unit is constructed from a plurality of learning models.

5. The diagnosis support device according to claim 4 , wherein the learning model is constructed from one or more of the following: 1) Random Forest 2) K-Nearest Neighbor 3) Multi-Layer Perceptron 4) Support Vector Machine 5) Logistic Regression 6) Naive Bayes 7) LightGBM

6. The diagnosis support device according to claim 5 , wherein the biometric data is any one of the following combinations: 1) Serum K value before potassium supplementation, serum K value, K preparation dosage, ARR, PAC, 2) Serum K value before potassium supplementation, serum K value, K preparation dosage, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value before potassium supplementation and serum K value 5) Serum K value before potassium supplementation

7. The diagnostic support device of claim 6 , wherein the biometric data does not include adrenal venous sampling (AVS) and / or adrenal CT scans.

8. A trained model for diagnostic support of disease A or disease B in a subject suspected of having disease A or disease B, the trained model including: 1) A plurality of biometric data sets of patients with disease A and a plurality of biometric data sets of patients with disease B as learning data; and 2) A trained model for predicting or determining whether a patient has disease A or disease B from explanatory variables including a plurality of biometric data groups of patients with disease A and a plurality of biometric data groups of patients with disease B, using machine learning to train the trained model.

9. The combination of the disease A and the disease B is bilateral aldosteronism and unilateral primary aldosteronism, and the biometric data is any one of the following combinations: The trained model according to claim 8. 1) Serum K value before potassium supplementation, serum K value, K preparation dosage, ARR, PAC, 2) Serum K value before potassium supplementation, serum K value, K preparation dosage, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value before potassium supplementation and serum K value 5) Serum K value before potassium supplementation

10. A diagnostic support method for disease A or disease B in a subject suspected of disease A or disease B, the method comprising the steps of: 1) A trained model generation process for generating a trained model capable of determining whether a patient has disease A or disease B by performing machine learning using a plurality of sets of biometric data of patients with disease A and a plurality of sets of biometric data of patients with disease B. 2) A process of predicting or determining that a subject suspected of having disease A or disease B is suffering from one or more biometric data of the subject to the trained model.

11. The combination of the disease A and the disease B is bilateral aldosteronism and unilateral primary aldosteronism, and the biometric data is any one of the following combinations: The diagnostic support method according to claim 10. 1) Serum K value before potassium supplementation, serum K value, K preparation dosage, ARR, PAC, 2) Serum K value before potassium supplementation, serum K value, K preparation dosage, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value before potassium supplementation and serum K value 5) Serum K value before potassium supplementation

12. A diagnostic support program for disease A or disease B in a subject suspected of disease A or disease B, the program comprising the following steps: 1) A trained model generation process for generating a trained model capable of determining whether a patient has disease A or disease B by performing machine learning using a plurality of sets of biometric data of patients with disease A and a plurality of sets of biometric data of patients with disease B. 2) A process of predicting or determining that a subject suspected of having disease A or disease B is suffering from one or more biometric data of the subject to the trained model.

13. The combination of the disease A and the disease B is bilateral aldosteronism and unilateral primary aldosteronism, and the biometric data is any one of the following combinations: The diagnostic support program according to claim 12. 1) Serum K value before potassium supplementation, serum K value, K preparation dosage, ARR, PAC, 2) Serum K value before potassium supplementation, serum K value, K preparation dosage, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value before potassium supplementation and serum K value 5) Serum K value before potassium supplementation

14. A data structure for supporting the diagnosis of disease A or disease B in a subject suspected of having disease A or disease B, the data structure including: 1) A plurality of biometric data sets of patients with disease A and a plurality of biometric data sets of patients with disease B as learning data; and 2) Data of a trained model for predicting or determining whether the patient has disease A or disease B from explanatory variables including a plurality of biometric data groups of patients with disease A and a plurality of biometric data groups of patients with disease B used for the training.

15. The combination of the disease A and the disease B is bilateral aldosteronism and unilateral primary aldosteronism, and the biometric data is any one of the following combinations:

15. A data structure for diagnostic support according to claim 14. 1) Serum K value before potassium supplementation, serum K value, K preparation dosage, ARR, PAC, 2) Serum K value before potassium supplementation, serum K value, K preparation dosage, ARR 3) Serum K value, serum K value, and K preparation dosage before potassium supplementation 4) Serum K value before potassium supplementation and serum K value 5) Serum K value before potassium supplementation

Images