Selection assistance device, selection assistance method, and recording medium

Abstract

This selection assistance device includes a matching degree calculation unit, a generation unit, a conversion unit, an extraction unit, and an output unit. The matching degree calculation unit calculates the matching degree of information on treatment for each patient with respect to a selection criterion of a clinical trial patient. The generation unit generates a graph indicating the relationship between the patients on the basis of the attributes of the patients and the data relating to the treatment. The conversion unit converts the matching degree with respect to the selection criterion and the graph into feature vectors of each patient. The extraction unit extracts, on the basis of the feature vectors of each patient, a patient matching the clinical trial from among the patients as a candidate for the clinical trial patient. The output unit outputs information on the extracted candidate for the clinical trial patient. By using the selection assistance device, decision making regarding extraction of a clinical trial patient can be appropriately performed.

Description

Selection support device, selection support method, and recording medium

[0001] The present disclosure relates to a selection support device and the like.

[0002] In clinical trials of pharmaceuticals, the person in charge of selecting candidate clinical trial patients extracts patients who meet the selection criteria for clinical trial patients, for example, by checking the descriptions in medical records. For example, the person in charge determines whether the selection criteria match the contents of the medical records. The person in charge then extracts patients whose selection criteria match the contents of the medical records as candidate clinical trial patients. In addition, a system that supports the extraction of candidate clinical trial patients may be used to extract candidate clinical trial patients.

[0003] The subject candidate extraction system of Patent Document 1 extracts reference information defined in a clinical trial protocol, and then performs an extraction process using the reference information to extract subject candidates that conform to the clinical trial protocol from among multiple patients.

[0004] International Publication No. 2019 / 225678

[0005] The technology described in Patent Document 1 may not be accurate enough to extract patients suitable for clinical trials.

[0006] In order to solve the above-mentioned problems, the present disclosure aims to provide a selection support device etc. that can improve the accuracy of extracting patients suitable for clinical trials.

[0007] In order to solve the above problems, the selection support device disclosed herein comprises a compatibility calculation means for calculating the compatibility of information related to the treatment of each patient with the selection criteria for clinical trial patients; a generation means for generating a graph showing the relationships between patients based on data related to each patient's attributes and treatment; a conversion means for converting the compatibility with the selection criteria and the graph into a feature vector for each patient; an extraction means for extracting patients who are compatible with the clinical trial from among the patients based on the feature vector for each patient as candidate clinical trial patients; and an output means for outputting information related to the extracted candidate clinical trial patients.

[0008] The selection support method disclosed herein calculates the suitability of information about each patient's treatment against the selection criteria for clinical trial patients, generates a graph showing the relationships between patients based on data about each patient's attributes and treatment, converts the suitability against the selection criteria and the graph into a feature vector for each patient, extracts patients who are suitable for the clinical trial from among the patients based on each patient's feature vector as candidate clinical trial patients, and outputs information about the extracted candidate clinical trial patients.

[0009] The recording medium of the present disclosure non-temporarily records a selection support program that causes a computer to execute the following processes: calculating the degree of suitability of information regarding each patient's treatment with respect to the selection criteria for clinical trial patients; generating a graph showing the relationships between patients based on data regarding each patient's attributes and treatment; converting the degree of suitability with respect to the selection criteria and the graph into a feature vector for each patient; extracting patients who are suitable for the clinical trial from among the patients as candidate clinical trial patients based on each patient's feature vector; and outputting information regarding the extracted candidate clinical trial patients.

[0010] According to the present disclosure, it is possible to improve the accuracy of extracting patients suitable for clinical trials.

[0011] FIG. 1 is a diagram illustrating an example of the configuration of a clinical trial support system according to the present disclosure. FIG. 2 is a diagram illustrating an example of the configuration of a selection support device according to the present disclosure. FIG. 3 is a diagram illustrating an example of conditions for including patients in a clinical trial according to the present disclosure. FIG. 4 is a diagram illustrating an example of conditions for excluding patients from a clinical trial according to the present disclosure. FIG. 5 is a diagram illustrating an example of compatibility and a graph according to the present disclosure. FIG. 6 is a diagram illustrating an example of a feature vector according to the present disclosure. FIG. 7 is a diagram illustrating an example of an extraction result of clinical trial candidates according to the present disclosure. FIG. 8 is a diagram illustrating an example of an extraction result of clinical trial candidates according to the present disclosure. FIG. 9 is a diagram illustrating an example of the operation flow of a selection support device according to the present disclosure. FIG. 10 is a diagram illustrating an example of the hardware configuration of a selection support device according to the present disclosure.

[0012] Embodiments of the present disclosure will be described in detail with reference to the drawings. FIG. 1 is a diagram illustrating an example of the configuration of a clinical trial support system. The clinical trial support system includes a selection support device 10, a terminal device 20, and a data management device 30. The selection support device 10 is connected to the terminal device 20, for example, via a network. The selection support device 10 is connected to the data management device 30, for example, via a network. There may be multiple terminal devices 20 and multiple data management devices 30. The number of terminal devices 20 and data management devices 30 can be set as appropriate.

[0013] A clinical trial support system is, for example, a system that supports the selection of clinical trial patients. A clinical trial patient is, for example, a patient who is the subject of a clinical trial of a new drug. The clinical trial support system outputs, for example, information indicating whether a patient receiving treatment at a medical institution is suitable for a clinical trial. Suitable for a clinical trial means, for example, that the patient's condition is suitable for verifying the effectiveness of a drug. A patient's suitability for a clinical trial is determined, for example, based on the degree to which information about the patient's treatment matches the selection criteria for clinical trial patients. In other words, clinical trial patients are selected from patients who match the selection criteria for clinical trial patients. The selection criteria are conditions for selecting a patient as a clinical trial patient. The selection criteria include, for example, inclusion criteria and exclusion criteria. The inclusion criteria indicate, for example, the criteria for including a patient in a clinical trial. Furthermore, the exclusion criteria indicate, for example, the criteria for excluding a patient from a clinical trial. In other words, the exclusion criteria are, for example, the criteria for not selecting a patient as a clinical trial patient.

[0014] In the example of FIG. 1 , the selection support device 10 can be accessed by, for example, a person in charge of a medical institution. The person in charge of a medical institution is, for example, a doctor. The person in charge of a medical institution may be a staff member other than a doctor. The person in charge of a medical institution may include a person in charge of an institution that has been entrusted with work by the medical institution. The person in charge of a medical institution is not limited to the above.

[0015] In the example of Figure 1, for example, a person in charge at an institution entrusted with a clinical trial by a pharmaceutical company operating a terminal device sends selection criteria to a person in charge at a medical institution, thereby inquiring about information on patients who meet the selection criteria. The institution entrusted with the clinical trial by the pharmaceutical company is, for example, a CRO (Contract Research Organization). The person in charge at the institution entrusted with the clinical trial by the pharmaceutical company is, for example, a CRA (Clinical Research Associate). The CRA may be an employee of the pharmaceutical company.

[0016] For example, the CRA inquires of a medical institution's personnel about information regarding patients who meet the selection criteria for patients to be included in the clinical trial. For example, the CRA inquires about information regarding patients who meet the selection criteria for patients to be included in the clinical trial at the stage of creating a clinical trial protocol. The clinical trial protocol is used, for example, to explain the contents of the clinical trial to and negotiate with the medical institution, and to submit notifications to related organizations. The uses of the clinical trial protocol are not limited to the above. Furthermore, the timing for inquiring about information regarding patients who meet the selection criteria is not limited to the stage of creating the clinical trial protocol.

[0017] The CRA is the person in charge at the medical institution, for example, the person in charge at the institution to which the hospital entrusts the clinical trial. The institution to which the hospital entrusts the clinical trial is, for example, an SMO (Site Management Organization). Also, the person in charge at the institution to which the hospital entrusts the clinical trial is, for example, a CRC (Clinical Research Coordinator).

[0018] The CRC, for example, operates the terminal device 20 to access the selection support device 10. The selection support device 10 is a device that, for example, references data stored in the data management device 30 and extracts candidate clinical trial patients based on the selection criteria and information on patients suitable for the clinical trial. The CRC, for example, accesses the selection support device 10 to extract information on patients suitable for the selection criteria sent from the CRA. The CRC then returns the extracted patient information to the CRA. The CRA, for example, references the patient information included in the response from the CRC to create a clinical trial protocol.

[0019] 1 illustrates an example in which the selection support device 10 is operated by a medical institution, but the selection support device 10 may also be operated by a third party. A third party refers to an entity other than a pharmaceutical company, an institution commissioned by a pharmaceutical company, a medical institution, or an institution commissioned by a medical institution. In this case, the entity operating the selection support device 10 receives data on the patient's condition from, for example, the medical institution. The selection support device 10 extracts and outputs candidate clinical trial patients based on a request from, for example, a terminal device 20 used by a medical institution's staff member or a terminal device used by a pharmaceutical company's staff member.

[0020] The selection support device 10 may also be operated by a pharmaceutical company or an institution commissioned by a pharmaceutical company. In this case, the selection support device 10 extracts and outputs information about candidate clinical trial patients by, for example, referencing pseudonymized information about treatments stored in the data management device 30. Examples of the configuration of the clinical trial support system are not limited to those described above. The configuration of the clinical trial support system may be set as appropriate.

[0021] Here, a specific example of the configuration of the selection support device 10 will be described. Fig. 2 is a diagram showing an example of the configuration of the selection support device 10. The selection support device 10 basically includes a compatibility calculation unit 12, a generation unit 13, a conversion unit 14, an extraction unit 15, and an output unit 17. The selection support device 10 further includes, for example, an acquisition unit 11, a model generation unit 16, and a storage unit 18.

[0022] The acquisition unit 11 acquires, for example, information about the attributes and treatment of each patient. The acquisition unit 11 also acquires, for example, selection criteria for clinical trial patients. The information about treatment is, for example, patient attributes and records of medical procedures in treating the patients. The acquisition unit 11 acquires, for example, the attributes of each patient and information about the patient's treatment from the data management device 30.

[0023] Patient attributes are, for example, information indicating the characteristics of a patient that is related to the patient's condition and does not change with treatment. Patient attributes are, for example, information on one or more of the patient's gender, age, family structure, family medical history, nationality, and race. Patient attributes are not limited to the above.

[0024] The information regarding the patient's treatment is, for example, a medical record of the patient's treatment. The information regarding the patient's treatment is, for example, a record of one or more items of diagnosis, examination, medication, surgery, follow-up, and the patient's condition. The information regarding the treatment may also include information about the person who performed the medical procedure. The information about the person who performed the medical procedure is, for example, the name or identifier of a medical professional, i.e., a doctor, nurse, pharmacist, or physical therapist. The information regarding the patient's treatment is, for example, data for each patient recorded in an electronic medical record. The information regarding the treatment may be test data. The information regarding the patient's treatment may also be data written on a medical receipt.

[0025] The record of the patient's condition includes information on one or more items of, for example, disease information, complication information, biomarkers, disease status, guideline score, effect, and test results. The effect is, for example, the effect of treatment, drug administration, and surgery. The effect is not limited to the above. The test result is, for example, the result of a biopsy, an imaging diagnostic test, and a genomic test. The test result is not limited to the above. Furthermore, the information regarding the patient's treatment is not limited to the above. Furthermore, the information regarding the treatment is not limited to the above.

[0026] The information regarding the patient's treatment may be time-series data indicating the patient's condition. For example, if the information regarding the patient's condition is test results using biomarkers, the information regarding the patient's treatment is time-series data of test results using the biomarkers performed at different times. Furthermore, for example, if the information regarding the patient's treatment is information whose changes can be estimated based on the patient's age, the information regarding the patient's treatment does not have to be time-series data.

[0027] The acquisition unit 11 acquires at least one of the criteria for inclusion of patients in a clinical trial and the criteria for exclusion from the clinical trial as selection criteria. The selection criteria include, for example, multiple criteria used to select clinical trial patients.

[0028] The inclusion criteria is, for example, information indicating the conditions of patients who are eligible to be clinical trial patients. The conditions of patients who are eligible to be clinical trial patients are indicated using the attributes of patients who are suitable as clinical trial patients and medical records of their treatment. Furthermore, the exclusion criteria is information indicating the conditions for excluding patients from the list of eligible clinical trial patients. In other words, the exclusion criteria is, for example, information indicating the conditions of patients who are not selected as clinical trial patients. The exclusion criteria is indicated using the attributes of patients who are not suitable as clinical trial patients and medical records of their treatment.

[0029] Figures 3 and 4 show examples of selection criteria when conducting a clinical trial of a lung cancer drug. Figure 3 shows an example of a sentence indicating the criteria for inclusion of patients in the clinical trial, among the selection criteria. In the example of the inclusion criteria for the clinical trial in Figure 3, the inclusion criteria include multiple criteria. Also, Figure 4 shows an example of a sentence indicating the criteria for exclusion of patients from the clinical trial, among the selection criteria. In the example of the exclusion criteria for the clinical trial in Figure 4, the exclusion criteria include multiple criteria.

[0030] The compatibility calculation unit 12 calculates the compatibility of the treatment-related information for each patient with the selection criteria for clinical trial patients. For example, the compatibility calculation unit 12 calculates the compatibility of the treatment-related information with each criterion included in the selection criteria for each patient. The compatibility with the selection criteria is, for example, an index indicating the degree to which the treatment-related information matches the clinical trial conditions set forth in the selection criteria. The compatibility with the selection criteria may be the percentage of criteria included in the selection criteria that match the treatment-related information with the clinical trial conditions set forth in each criterion.

[0031] The relevance calculation unit 12 calculates the relevance of the treatment information for each patient to the selection criteria for the clinical trial patients using, for example, a relevance calculation model. The relevance calculation model is, for example, a machine learning model that calculates the relevance based on a feature vector converted from the selection criteria and a feature vector converted from the data for each patient recorded in the electronic medical record. The relevance calculation model, for example, converts each condition included in the selection criteria into a feature vector. For example, the relevance calculation model uses a dictionary to extract treatment-related terms from the selection criteria. The dictionary includes, for example, terms in the medical field. The relevance calculation model then converts, for example, the terms extracted from the selection criteria and the relationships between the terms into feature vectors. In other words, the feature vector converted from the selection criteria reflects the content of the selection criteria. The relevance calculation model also converts the treatment information into a feature vector. For example, the relevance calculation model uses a dictionary to extract treatment-related terms from the treatment information. The relevance calculation model then converts, for example, the terms extracted from the treatment information into a feature vector. In other words, the feature vector converted from the treatment information reflects the content of the treatment information. The selection criteria and information on treatment are converted into feature vectors using, for example, Word2Vec. The algorithm for converting the selection criteria and information on treatment into feature vectors is not limited to the above. The fitness calculation model calculates the fitness for the selection criteria based on, for example, the Euclidean distance or cosine similarity between the feature vector converted from the conditions included in the selection criteria and the feature vector converted from the information on treatment.

[0032] The compatibility calculation unit 12 may calculate the compatibility with the selection criteria based on information extracted using a language model. For example, the compatibility calculation unit 12 extracts information corresponding to the selection criteria from the treatment information for each patient using a language model. Then, for example, the compatibility calculation unit 12 calculates the compatibility with the selection criteria based on the extracted information corresponding to the selection criteria. For example, the compatibility calculation unit 12 extracts sentences similar to each of the criteria included in the selection criteria from the treatment information. Then, the compatibility calculation unit 12 calculates the compatibility of the extracted treatment information with each of the criteria included in the selection criteria. For example, the compatibility calculation unit 12 determines whether or not a term or numerical value indicated by the criterion included in the selection criteria matches a term or numerical value included in the extracted treatment information. Then, for example, the compatibility calculation unit 12 calculates the compatibility with the selection criteria as the number of criteria whose terms or numerical values ​​match those of the treatment information relative to the total number of criteria included in the selection criteria. Matching may include similarity.

[0033] The compatibility calculation unit 12 may calculate the compatibility based on data in which the selection criteria are structured. For example, the compatibility calculation unit 12 generates data in which the selection criteria are structured using a language model. The language model in this case may be the same as the language model used when calculating the compatibility with the selection criteria based on information extracted using the language model, as described above. Furthermore, the language model in this case may be different from the language model used when calculating the compatibility with the selection criteria based on information extracted using the language model, as described above. Furthermore, the compatibility calculation unit 12 extracts information corresponding to the items of the selection criteria from the information about the treatment. Then, the compatibility calculation unit 12 calculates the compatibility based on whether the information about the treatment matches the conditions associated with the items of the selection criteria. Matching may also include similarity. The items of the selection criteria are information indicating what each criterion included in the selection criteria is a criterion.

[0034] The structured data of the selection criteria is, for example, data associating each item of the criteria included in the selection criteria with a condition corresponding to the item. For example, in the example of the inclusion criteria in Figure 3, the items are "age," "diagnosis," "lesion size," "lesion count," and "white blood cell count." Also, in the example of the inclusion criteria in Figure 3, for example, when the item is "age," the condition corresponding to the item is "20 years of age or older."

[0035] The language model may be, for example, GPT-2 (Generative Pre-trained Transformer-2), GPT-3, or GPT-4. Alternatively, the language model may be T5 (Text-to-Text Transfer Transformer), BERT (Bidirectional Encoder Representations from Transformers), RoBERTa (Robustly optimized BERT approach), or ELECTRA (Efficiently Learning an Encoder that Classifies Token Replacements Accurately).

[0036] The compatibility calculation unit 12 may calculate the compatibility using a compatibility calculation model that calculates the compatibility from information about conditions and treatments associated with the items. In this case, the compatibility calculation model is generated by learning the relationship between the information about conditions and treatments associated with each item of the selection criteria and the compatibility of the information about treatments with the selection criteria. The compatibility calculation model is generated, for example, by deep learning using a neural network. The algorithm for generating the compatibility calculation model is not limited to the above. Furthermore, the compatibility calculation model may be generated, for example, in a device external to the selection support device 10. Furthermore, the method by which the compatibility calculation unit 12 calculates the compatibility is not limited to the above.

[0037] The generation unit 13 generates a graph showing the relationships between patients based on each attribute of the patients and information about their treatments. A graph showing the relationships between patients is also called a knowledge graph. The generation unit 13 may generate a graph showing the relationships between patients based on either each attribute of the patients or information about their treatments. The generation unit 13 generates a graph showing the relationships between patients based on, for example, items among the attributes of each patient and information about their treatments that have a significant impact on the onset of disease, disease progression, and drug effectiveness. For example, if a blood relative has developed a disease, there may be a high possibility that the patient will develop the same disease. In such a case, the generation unit 13 generates a graph showing the relationships between patients based on the blood relationships and the medical histories of the blood relatives.

[0038] Among the items included in the information on each patient's attributes and treatment, the items used to generate the graph may be set for each drug undergoing clinical trials. For example, for a drug used to treat a disease known to be influenced by genetic factors on the onset and progression of the disease, the items used to generate the graph are set to include items related to blood relationships. Furthermore, for a drug used to treat a disease known to be significantly influenced by lifestyle habits on the onset and progression of the disease, the items used to generate the graph are set to include, for example, weight, BMI (Body Mass Index), smoking history, alcohol consumption, and exercise volume. Among the items included in the information on each patient's attributes and treatment, the items used to generate the graph are set, for example, by the user of the selection support device 10.

[0039] The generation unit 13 generates a graph showing the relationships between patients, for example, by the following process. For example, the generation unit 13 converts attribute data for each patient into an embedding vector using a language model. Furthermore, for example, the generation unit 13 converts information about the patient's treatment for each patient into an embedding vector using a language model. The language model converts the information about the attributes and treatment into an embedding vector using, for example, a dictionary. Then, the generation unit 13 converts the information about the attributes and treatment of each patient into a feature vector using, for example, the converted embedding vector. In this case, the feature vector is a multidimensional vector that reflects the information about the attributes and treatment of each patient. In other words, the feature vector in this case is a multidimensional vector that represents the features of each patient.

[0040] The language model converts information about patient attributes and treatment into embedding vectors using, for example, a dictionary generated based on terms in the field of use. The language model may convert information about attributes and treatment into embedding vectors using multiple dictionaries. In this case, the language model converts information about attributes and treatment into embedding vectors using, for example, a dictionary corresponding to the medical field in general and a dictionary corresponding to the medical department in which the drug under clinical trial is used. The language model used by the generation unit 13 may be the same as the language model used by the compatibility calculation unit 12. Alternatively, the language model used by the generation unit 13 may be different from the language model used by the compatibility calculation unit 12. For example, Word2Vec may be used as the language model. GPT-2, GPT-3, or GPT-4 may be used as the language model. Alternatively, T5, BERT, RoBERTa, or ELECTRA may be used as the language model. The language model used for conversion into embedding vectors is not limited to the above.

[0041] The generation unit 13 calculates the similarity between patients, for example, using the distance between the feature vectors of each patient. The generation unit 13 calculates the similarity between patients, for example, by calculating the distance between the feature vectors. The distance between the vectors is, for example, Euclidean distance. The distance between the feature vectors is not limited to Euclidean distance. The generation unit 13 may also calculate the similarity between patients by calculating the cosine similarity between the feature vectors. The generation unit 13 generates a graph showing the relationship between patients by connecting nodes with edges based on the similarity between patients. Here, the nodes represent, for example, each patient. The edges represent, for example, the relationship between patients. In other words, the graph showing the relationship between patients is a graph in which nodes representing each patient are connected with edges connecting similar patients. "Similar" means, for example, that information regarding attributes and treatments has a relationship. "Relationship" between information regarding attributes and treatments means, for example, that if one meets the selection criteria for a clinical trial, there is a high probability that the other also meets the selection criteria for a clinical trial. That is, "similar" means that, for example, if one patient meets a certain selection criterion, there is a high possibility that the other patient will also meet the criterion. Also, "similar" can include "identical."

[0042] The generator 13 may use a generative model to generate a graph showing the relationships between patients. The generative model is, for example, a machine learning model that generates a graph showing the relationships between patients from information about the attributes and treatments of each patient. The generator 13 converts, for example, data included in the attributes of each patient into an embedding vector using a language model. Then, the generator 13 generates a graph showing the relationships between patients, for example, using the embedding vector and the generative model as input. The generative model is generated, for example, by learning the relationship between the information about the attributes and treatments of each patient and the graph showing the relationships between patients. The generative model is generated, for example, by deep learning using a neural network. The learning algorithm used to generate the generative model is not limited to the above. The generative model may also be generated, for example, in a device external to the selection support device 10.

[0043] FIG. 5 is a diagram schematically illustrating an example of a graph showing patient relationships and the degree of conformance of information about patient treatments with the criteria included in the selection criteria. In the example of FIG. 5, in the graph showing patient relationships, edges connect "Patient A" and "Patient B," "Patient A" and "Patient C," "Patient B" and "Patient D," and "Patient C" and "Patient D." Each patient corresponds to a node. In the example of the graph shown in FIG. 3, patients connected by edges have, for example, similar attributes. For example, "Patient A" and "Patient B," "Patient A" and "Patient C," "Patient B" and "Patient D," and "Patient C" and "Patient D" have similar attributes.

[0044] 5 also shows the suitability of each of "Patient A," "Patient B," "Patient C," and "Patient D." For each patient, the example of FIG. 5 shows the suitability of the information about the patient's treatment for each criterion included in the selection criteria.

[0045] The conversion unit 14 converts a graph showing the degree of suitability for the selection criteria and the relationships between patients into a feature vector for each patient. The conversion unit 14 converts, for example, a graph showing the degree of suitability for each criterion included in the selection criteria and the relationships between patients into a feature vector for each patient. The conversion unit 14 converts, for example, a graph showing the degree of suitability for the selection criteria and the relationships between patients into a feature vector for each patient using a conversion model that converts the graph showing the degree of suitability and the relationships between patients into a single feature vector. The conversion model is, for example, a machine learning model that converts a graph showing the degree of suitability for each criterion included in the selection criteria and the relationships between patients into a feature vector for each patient.

[0046] The conversion model is generated, for example, as follows. In the first stage of generating the conversion model, a learning device that generates the conversion model learns a vector representation for each piece of information related to attributes and treatments, for example, using a message passing technique in which messages are exchanged via edges in a graph showing relationships between patients. In the second stage, the learning device learns a vector representation for each node that combines vectors for each piece of information related to fitness, attributes, and treatments. The nodes correspond, for example, to each patient. The conversion model generated in this manner can convert the graph showing fitness and relationships between patients into a feature vector. This learning technique is also referred to as embedding propagation. The conversion model is generated, for example, in a system external to the selection support device 10. The conversion model may also be generated by a learning means (not shown) included in the selection support device 10.

[0047] FIG. 6 shows an example of feature vectors converted from a graph showing the fitness and patient relationships using a conversion model. The feature vectors shown in the example of FIG. 6 are converted, for example, from the graph showing the fitness and patient relationships for each patient shown in FIG. 5. In the example of FIG. 6, the graph showing the fitness and patient relationships for "Patient A" is converted into a feature vector [a1, a2, a3, a4, a5, a6]. In the example of FIG. 6, the graph showing the fitness and patient relationships for "Patient B" is converted into a feature vector [b1, b2, b3, b4, b5, b6]. Also, in the example of FIG. 6, the graph showing the fitness and patient relationships for "Patient C" is converted into a feature vector [c1, c2, c3, c4, c5, c6].

[0048] The extraction unit 15 extracts patients suitable for the clinical trial as candidate clinical trial patients based on the feature vectors of each patient. The feature vectors of each patient are feature vectors converted by the conversion unit 14 from a graph showing the suitability of each patient with respect to the selection criteria and the relationships between the patients. The extraction unit 15 extracts, for example, patients whose suitability for the clinical trial is equal to or exceeds a standard as patients suitable for the clinical trial. The standard for suitability for the clinical trial is set, for example, so that patients whose suitability exceeds the standard are suitable as clinical trial patients.

[0049] The extraction unit 15 extracts candidate clinical trial patients using, for example, an extraction model. The extraction model is used, for example, to extract patients suitable for the clinical trial based on the feature vectors of each patient. The extraction model is, for example, a machine learning model that estimates the suitability for the clinical trial using the feature vectors of each patient as input. The suitability for the clinical trial is, for example, the probability that the patient will be suitable for the clinical trial. The extraction unit 15 extracts patients suitable for the clinical trial based on, for example, the suitability for the clinical trial of each patient estimated by the extraction model. In other words, the extraction unit 15 extracts, for example, patients whose suitability for the clinical trial of each patient estimated by the extraction model is equal to or exceeds a standard as patients suitable for the clinical trial.

[0050] The extraction unit 15 may, for example, extract whether or not each patient is suitable for the clinical trial. The extraction unit 15 may also extract the suitability of each patient for the clinical trial in multiple stages. By setting the suitability in multiple stages, it becomes possible to extract patients who are candidates for clinical trial patients, for example, when some of the criteria are relaxed. The extraction unit 15, for example, extracts information indicating which of multiple stages the suitability for the clinical trial estimated by the extraction model is for each patient. The extraction unit 15 may also extract the number of patients suitable for the clinical trial. When multiple stages are set according to the suitability for the clinical trial, the extraction unit 15 may extract the number of patients for each stage.

[0051] The model generation unit 16 generates, for example, an extraction model. The model generation unit 16 generates, for example, the extraction model by machine learning the relationship between each patient's feature vector and whether or not they are suitable for the clinical trial. The feature vector for each patient is a feature vector converted from a graph showing the suitability and the relationship between patients using a conversion model by the conversion unit 14. The suitability for the clinical trial is performance data on whether or not each patient is suitable for the clinical trial criteria. The model generation unit 16 generates the extraction model by, for example, deep learning using a neural network. The extraction model may be generated by a device external to the selection support device 10.

[0052] The model generation unit 16 may generate a fitness calculation model that the fitness calculation unit 12 uses to calculate fitness for the selection criteria. The model generation unit 16 may generate a generation model that the generation unit 13 uses to generate a graph. The model generation unit 16 may also generate a conversion model that the conversion unit 14 uses to convert into a feature vector. The model generation unit 16 may also re-train each of the learning models.

[0053] The output unit 17 outputs information about the candidate clinical trial patients extracted by the extraction unit 15. The information about the candidate clinical trial patients is, for example, information identifying patients who are suitable for the clinical trial. The information identifying patients who are suitable for the clinical trial is, for example, the patient's name. The information identifying patients who are suitable for the clinical trial is not limited to the patient's name. The output unit 17 may also output the suitability of each patient for the clinical trial as information about the candidate clinical trial patient. If multiple levels of suitability are set, the output unit 17 may also output the level of suitability for the clinical trial for each patient as information about the candidate clinical trial patient. The output unit 17 may also output information identifying patients in order of the patients with the highest suitability for the clinical trial.

[0054] The output unit 17 may output a graph showing the relationships between patients generated by the generation unit 13, in addition to the information about the candidate clinical trial patients extracted by the extraction unit 15. Furthermore, the extraction unit 15 may output the graph showing the relationships between patients generated by the generation unit 13, with the suitability of each patient for the clinical trial superimposed on it.

[0055] The output unit 17 may output the number of candidate clinical trial patients as information related to the candidate clinical trial patients. The number of candidate clinical trial patients is, for example, the number of patients who are suitable for the clinical trial. The output unit 17 outputs, for example, the number of patients whose suitability for the clinical trial meets or exceeds a standard. Furthermore, when multiple levels of suitability are set, the output unit 17 may output, for each level of suitability for the clinical trial, the number of patients included in each level of suitability.

[0056] The output unit 17 may output the degree of conformance of each patient with respect to the selection criteria. The output unit 17 may also output the degree of conformance of each patient with respect to each criterion included in the selection criteria. The output unit 17 outputs information about the candidate clinical trial patients extracted by the extraction unit 15 to, for example, the terminal device 20.

[0057] FIG. 7 is an example of the extraction results for candidate clinical trial patients. In the example of the extraction results in FIG. 7, the suitability of each clinical trial patient as a clinical trial patient is indicated. In the example of the extraction results in FIG. 7, patients who are suitable for the clinical trial are represented by an "O". Patients who are suitable for the clinical trial are, for example, patients whose suitability for the clinical trial meets or exceeds a standard. The standard for suitability for the clinical trial is set so that it can be determined whether a patient is suitable as a subject for the clinical trial. In the example of the extraction results in FIG. 7, patients who are not suitable for the clinical trial are represented by an "X". Patients who are not suitable for the clinical trial are, for example, patients whose suitability for the clinical trial is below the standard.

[0058] FIG. 8 is an example showing the extraction results of candidate clinical trial patients using the patient's suitability for the clinical trial. In the example of the extraction results in FIG. 8 , the suitability as a clinical trial patient is shown for each clinical trial patient. The suitability is, for example, the probability that a patient is a candidate clinical trial patient when the extraction model extracts candidate clinical trial patients. A person in charge of selecting clinical trial patients can select patients who are suitable as clinical trial patients by, for example, referring to the extraction results shown in the example of FIG. 8 . The output unit 17 may output the suitability indicated by multiple levels. Furthermore, the output unit 17 may output the suitability for the clinical trial by changing the display format based on the value of the suitability for the clinical trial or the level of suitability for the clinical trial. For example, the output unit 17 outputs the suitability by changing the color of the numerical value according to the level of suitability for the clinical trial.

[0059] FIG. 9 is an example of a display screen that displays the extraction results of candidate clinical trial patients along with a graph showing the relationships between patients. In the example of the display screen in FIG. 9, the suitability of each clinical trial patient, which corresponds to a node in the graph, as a clinical trial patient, is shown. In the example of the display screen in FIG. 9, patients who are suitable for the clinical trial are represented by a "○". Furthermore, in the example of the display screen in FIG. 9, patients who are not suitable for the clinical trial are represented by an "X". In the example of the display screen in FIG. 9, each patient is shown as a node. Furthermore, in the example of the display screen in FIG. 9, patients with similar attributes are connected by edges in the graph.

[0060] In the example display screen of FIG. 9 , for example, the suitability of "Patient A" to the selection criteria is 0.9, and the suitability of "Patient B," "Patient C," and "Patient D" to the selection criteria is approximately 0.7. In this case, for example, "Patient A" is highly likely to be suitable for the clinical trial. Therefore, in the example display screen of FIG. 9 , the suitability of "Patient A" as a clinical trial patient is displayed as "○." Furthermore, "Patient B" and "Patient C," who are different in relationship to "Patient A," are displayed as "○" for their suitability to the selection criteria, even though their suitability to the selection criteria is lower than that of "Patient A." On the other hand, in the example display screen of FIG. 9 , the suitability of "Patient D" to the selection criteria is similar to that of "Patient B" and "Patient C," but is more distantly related to "Patient A" than "Patient B" and "Patient C," so their suitability to the clinical trial patient is displayed as "○." In this way, by extracting patients suitable for the clinical trial based on not only the suitability to the selection criteria but also the relationships between patients, the accuracy of extracting patients suitable for the clinical trial can be improved. Furthermore, a person in charge of selecting candidate clinical trial patients can refer to the display screen of FIG. 9, for example, to select candidate clinical trial patients by referring to the similarity of patient attributes.

[0061] The memory unit 18 stores, for example, data related to the extraction of candidate clinical trial patients. The memory unit 18 stores, for example, information related to the attributes and treatment of each patient. The memory unit 18 also stores, for example, the extraction results of candidate clinical trial patients. The memory unit 18 stores, for example, language models and dictionaries used by the language models. The memory unit 18 stores, for example, compatibility calculation models. The memory unit 18 stores, for example, generative models. The memory unit 18 stores, for example, conversion models. The memory unit 18 also stores, for example, extraction models. The language models, compatibility calculation models, generative models, conversion models, and extraction models may each be stored in a storage means other than the memory unit 18.

[0062] The terminal device 20 is, for example, a terminal device that accesses the selection support device 10 and is used for processing to extract candidate patients suitable for the clinical trial. The terminal device 20 is, for example, a terminal device used by a person in charge of a medical institution or a person in charge of an institution entrusted with work related to the clinical trial by the medical institution. The person in charge of an institution entrusted with work related to the clinical trial by the medical institution is, for example, a CRC.

[0063] The terminal device 20 outputs the selection criteria to, for example, the acquisition unit 11 of the selection support device 10. Then, the terminal device 20 acquires information about candidate clinical trial patients from the output unit 17 of the selection support device 10. For example, the terminal device 20 acquires information about the suitability of the patient for the clinical trial from the output unit 17 of the selection support device 10. Then, the terminal device 20 outputs the information about the suitability of the patient for the clinical trial to a display device (not shown).

[0064] The data management device 30 is a device that stores information related to a patient's treatment. The information related to a patient's treatment is, for example, data in an electronic medical record entered by a doctor. The information in the electronic medical record may also be data entered by a nurse or a laboratory technician. The information related to a patient's treatment may also be information on one or more items of the patient's disease information, complication information, biomarkers, disease status, guideline score, medical history, effects, and test results other than those described in the electronic medical record.

[0065] The operation of the process for extracting candidate clinical trial patients will now be described. Fig. 10 is a diagram showing an example of the operational flow of the process for extracting candidate clinical trial patients.

[0066] The acquisition unit 11 acquires, for example, the attributes of each patient and information related to the treatment (step S11). The own information related to the treatment is, for example, information about each patient recorded in an electronic medical record. The acquisition unit 11 acquires, for example, the attributes of each patient and information related to the treatment from the data management device 30.

[0067] Once the attributes and treatment-related information for each patient have been acquired, the compatibility calculation unit 12 calculates the compatibility of the treatment-related information for each patient with the selection criteria for clinical trial patients (step S12). The selection criteria for clinical trial patients are input to the selection support device 10, for example, by a user of the selection support device 10.

[0068] When the degree of conformance to the selection criteria for all patients is calculated (Yes in step S13), the generating unit 13 generates a graph showing the relationships between patients based on the attributes and treatment information of each patient (step S14).

[0069] When the graph showing the relationships between patients is generated, the conversion unit 14 converts the degree of conformance to the selection criteria and the graph showing the relationships between patients into feature vectors for each patient (step S15).

[0070] Once all target patients have been converted into feature vectors (Yes in step S16), the extraction unit 15 extracts patients who are suitable for the clinical trial from among the patients as candidate clinical trial patients based on the feature vectors of each patient (step S17).

[0071] When candidate clinical trial patients are extracted, the output unit 17 outputs information about the candidate clinical trial patients extracted by the extraction unit 15 (step S18). The output unit 17 outputs the candidate clinical trial patients extracted by the extraction unit 15, for example, to the terminal device 20. If no candidate clinical trial patients are extracted in step S17, the output unit 17 outputs, for example, information indicating that there are no candidate clinical trial patients.

[0072] In step S13, if there is a patient for whom the suitability against the selection criteria has not been calculated (No in step S13), the process returns to step S13, and the suitability calculation unit 12 calculates the suitability of the information regarding each patient's treatment against the selection criteria for clinical trial patients.

[0073] In step S16, if there are any patients for which conversion to feature vectors has not been completed (No in step S16), the process returns to step S15, and the conversion unit 14 converts the graph showing the conformance to the selection criteria and the relationships between patients into feature vectors for each patient.

[0074] The selection support device 10 calculates the degree of conformance between the selection criteria for clinical trial patients and information about each patient's treatment. The selection support device 10 also generates a graph showing the relationships between patients based on the attributes and treatment information for each patient. The selection support device 10 then converts the graph showing the conformance and the relationships between patients into a feature vector for each patient. The selection support device 10 extracts patients who are suitable for the clinical trial from among the patients based on each patient's feature vector. In this way, by extracting candidate clinical trial patients based on the conformance to the selection criteria and the relationships between patients, the selection support device 10 can improve the accuracy of extracting patients suitable for the clinical trial.

[0075] The selection support device 10 generates a graph showing the relationships between patients based on, for example, items from information about each patient's attributes and treatment that have a significant impact on the onset of disease, disease progression, and drug effectiveness. The selection support device 10 then extracts patients suitable for the clinical trial by estimating their suitability for the clinical trial based on feature vectors converted from the graph showing the suitability for the selection criteria and the relationships between patients. In this way, extracting patients based on relationships between patients based on items that have a significant impact on disease is likely to enable efficient verification of drug effectiveness in clinical trials. Extracting multiple patients with similar relationships between patients may, for example, enable verification of drug effectiveness while suppressing the influence of fluctuations. Therefore, using the selection support device 10 can improve the accuracy of extracting patients suitable for clinical trials.

[0076] Furthermore, by re-learning the extraction model based on the track record indicating whether a patient has become a clinical trial subject, the selection support device 10 can optimize the extraction model according to, for example, the environment in which clinical trial candidate patients are actually extracted. As a result, the selection support device 10 can extract patients suitable for clinical trials with higher accuracy.

[0077] Furthermore, by outputting a graph showing the relationships between patients in addition to information about the extracted clinical trial patient candidates, the person in charge of extracting clinical trial patient candidates can make appropriate decisions regarding the selection of clinical trial patients, for example, by referring to the relationships between patients. Furthermore, by outputting the suitability of information about each patient's treatment with respect to the selection criteria, the person in charge of extracting clinical trial patient candidates can make appropriate decisions regarding the selection of clinical trial patients, for example, by referring to the suitability of information about each patient's treatment with respect to the selection criteria. Thus, the selection support device 10 can support decision-making regarding the selection of clinical trial patient candidates.

[0078] The processes in the selection support device 10 may be distributed and executed among a plurality of information processing devices connected via a network. For example, the processes in the compatibility calculation unit 12, the generation unit 13, the conversion unit 14, and the extraction unit 15 may be executed in different information processing devices. It can be set as appropriate which information processing device executes each process in the selection support device 10.

[0079] Each process in the selection support device 10 can be realized by executing a computer program on a computer. Fig. 11 shows an example of the configuration of a computer 100 that executes a computer program that performs each process in the selection support device 10. The computer 100 includes a CPU (Central Processing Unit) 101, a memory 102, a storage device 103, an input / output I / F (Interface) 104, and a communication I / F 105.

[0080] The CPU 101 reads and executes computer programs for performing each process from the storage device 103. The CPU 101 may be configured as a combination of multiple CPUs. The CPU 101 may also be configured as a combination of a CPU and another type of processor. For example, the CPU 101 may be configured as a combination of a CPU and a graphics processing unit (GPU). The memory 102 is configured with a dynamic random access memory (DRAM) or the like, and temporarily stores computer programs executed by the CPU 101 and data being processed. The storage device 103 stores computer programs executed by the CPU 101. The storage device 103 is configured with, for example, a non-volatile semiconductor storage device. Other storage devices such as a hard disk drive may also be used for the storage device 103. The input / output I / F 104 is an interface that accepts input from an operator and outputs display data, etc. The communication I / F 105 is an interface that transmits and receives data between the terminal device 20, the data management device 30, and other information processing devices. The terminal device 20 and the data management device 30 may also have the same configuration as the computer 100 .

[0081] The computer program used to execute each process can also be stored and distributed on a computer-readable recording medium that non-temporarily stores data. Examples of recording media that can be used include magnetic tapes for recording data and magnetic disks such as hard disks. Optical disks such as CD-ROMs (Compact Disc Read Only Memory) can also be used as recording media. Non-volatile semiconductor storage devices can also be used as recording media.

[0082] Some or all of the above-described embodiments can be described as, but are not limited to, the following supplementary notes.

[0083] [Supplementary Note 1] A selection support device comprising: a fitness calculation means for calculating the fitness of information relating to the treatment of each patient with respect to the selection criteria for clinical trial patients; a generation means for generating a graph showing the relationships between patients based on the attributes of each patient and data relating to the treatment; a conversion means for converting the fitness for the selection criteria and the graph into a feature vector for each patient; an extraction means for extracting patients who are suitable for the clinical trial from among patients based on the feature vector for each patient as candidates for the clinical trial patients; and an output means for outputting the extracted candidates for the clinical trial patients.

[0084] [Supplementary Note 2] The selection support device according to Supplementary Note 1, wherein the fitness calculation means calculates the fitness for each criterion included in the selection criteria for each of the patients, and the conversion means converts the fitness for each criterion included in the selection criteria and the graph into a feature vector for each of the patients.

[0085] [Supplementary Note 3] The selection support device according to Supplementary Note 1 or 2, wherein the compatibility calculation means converts information about the treatment of each of the patients into features using a language model, and calculates the compatibility based on the converted features.

[0086] [Supplementary Note 4] The selection support device according to Supplementary Note 3, wherein the compatibility calculation means calculates the compatibility based on a distance between a feature vector converted from the selection criteria and a feature vector converted from information on the treatment of each of the patients.

[0087] [Supplementary Note 5] The selection support device according to any one of Supplementary Notes 1 to 4, wherein the conversion means converts the degree of conformance with the selection criteria and the graph into a feature vector for each of the patients using a conversion model that converts the degree of conformance with the selection criteria and the graph into one feature vector.

[0088] [Supplementary Note 6] The selection support device according to any one of Supplementary Notes 1 to 5, wherein the generating means converts the attributes of each patient and the data related to the treatment into features using a language model, and generates the graph based on the similarity between patients calculated from the features of each patient.

[0089] [Supplementary Note 7] The selection support device according to any one of Supplementary Notes 1 to 6, wherein the output means further outputs a graph showing the relationships between the patients.

[0090] [Supplementary Note 8] The selection support device according to any one of Supplementary Notes 1 to 7, wherein the output means further outputs a degree of suitability of each of the patients with respect to the selection criteria.

[0091] [Supplementary Note 9] The selection support device according to any one of Supplementary Notes 1 to 8, wherein the selection criteria include at least one of conditions for inclusion of patients in the clinical trial and conditions for exclusion of patients from the clinical trial.

[0092] [Supplementary Note 10] The selection support device according to any one of Supplementary Notes 1 to 9, wherein the extraction means extracts the candidate clinical trial patients using an extraction model that extracts patients who meet selection criteria based on the feature vectors of each of the patients.

[0093] [Supplementary Note 11] The selection support device according to Supplementary Note 10, further comprising a model generation means for generating the extraction model by machine learning the relationship between the feature vector of each of the patients and whether or not they are suitable for the clinical trial.

[0094] [Supplementary Note 12] A selection support method comprising: calculating the suitability of information on the treatment of each patient to selection criteria for clinical trial patients; generating a graph showing the relationships between patients based on each patient's attributes and data on the treatment; converting the suitability to the selection criteria and the graph into a feature vector for each patient; extracting patients who are suitable for the clinical trial from among the patients as candidates for the clinical trial based on the feature vector for each patient; and outputting the extracted candidates for the clinical trial patients.

[0095] [Supplementary Note 13] A recording medium that non-temporarily records a selection support program that causes a computer to execute the following processes: a process of calculating the degree of suitability of information regarding the treatment of each patient with respect to the selection criteria for clinical trial patients; a process of generating a graph showing the relationships between patients based on the attributes of each patient and data regarding the treatment; a process of converting the degree of suitability with respect to the selection criteria and the graph into a feature vector for each patient; a process of extracting patients who are suitable for the clinical trial from among the patients based on the feature vector for each patient as candidates for the clinical trial patients; and a process of outputting the extracted candidates for the clinical trial patients.

[0096] Furthermore, some or all of the configurations described in Supplementary Notes 2 to 11, which are dependent on Supplementary Note 1, may also be dependent on Supplementary Notes 12 and 13 in the same dependent relationship as Supplementary Notes 2 to 11. Furthermore, not limited to Supplementary Notes 1, 12, and 13, some or all of the configurations described as Supplements may be made dependent on various hardware, software, various recording means for recording software, or systems, within the scope of each of the above-mentioned embodiments.

[0097] Although the present disclosure has been described above with reference to the embodiments, the present disclosure is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure. Furthermore, each embodiment can be combined with other embodiments as appropriate.

[0098] REFERENCE SIGNS LIST 10 Selection support device 11 Acquisition unit 12 Goodness of fit calculation unit 13 Generation unit 14 Conversion unit 15 Extraction unit 16 Model generation unit 17 Output unit 18 Storage unit 20 Terminal device 30 Data management device 100 Computer 101 CPU 102 Memory 103 Storage device 104 Input / output I / F 105 Communication I / F

Claims

1. A selection support device comprising: a fitness calculation means for calculating the fitness of the treatment information of each patient with respect to the selection criteria of the clinical trial patients; a generation means for generating a graph showing the relationship between patients based on the attributes of each patient and the treatment information; a conversion means for converting the fitness with respect to the selection criteria and the graph into the feature vectors of each patient; an extraction means for extracting, as candidates for the clinical trial patients, the patients who meet the clinical trial based on the feature vectors of each patient; and an output means for outputting information on the extracted candidates for the clinical trial patients.

2. The selection support device according to claim 1, wherein the fitness calculation means calculates the fitness of each patient with respect to each of the criteria included in the selection criteria, and the conversion means converts the fitness with respect to each of the criteria included in the selection criteria and the graph into the feature vectors of each patient.

3. The selection support device according to claim 1 or 2, wherein the fitness calculation means converts the treatment information of each patient into features using a language model, and calculates the fitness based on the converted features.

4. The selection support device according to claim 3, wherein the fitness calculation means calculates the fitness based on the distance between the feature vector converted from the selection criteria and the feature vector converted from the treatment information of each patient.

5. The selection support device according to any one of claims 1 to 4, wherein the conversion means converts the fitness with respect to the selection criteria and the graph into the feature vectors of each patient using a conversion model that converts the fitness with respect to the selection criteria and the graph into one feature vector.

6. The selection support device according to any one of claims 1 to 5, wherein the generation means converts at least one of the attributes of each patient and the treatment information into features using a language model, and generates the graph based on the similarity between patients calculated from the features of each patient.

7. The selection support device according to any one of claims 1 to 6, wherein the output means further outputs a graph showing the relationship between the patients.

8. The selection support device according to any one of claims 1 to 7, wherein the output means further outputs the fitness of each patient with respect to the selection criteria.

9. The selection support device according to any one of claims 1 to 8, wherein the selection criteria include at least one of the conditions for including patients in the clinical trial and the conditions for excluding patients from the clinical trial.

10. The selection support device according to any one of claims 1 to 9, wherein the extraction means extracts candidates for the clinical trial patients using an extraction model that extracts patients suitable for the clinical trial based on the feature vectors of the respective patients.

11. The selection support device according to claim 10, further comprising model generation means for generating the extraction model by machine learning the relationship between the feature vector of each patient and the presence or absence of suitability for the clinical trial.

12. A selection support method, which calculates the degree of fitness of the information on the treatment of each patient with respect to the selection criteria for clinical trial patients, generates a graph showing the relationship between patients based on the attributes of each patient and the data on the treatment, converts the degree of fitness and the graph with respect to the selection criteria into the feature vectors of the respective patients, extracts patients suitable for the clinical trial as candidates for the clinical trial patients based on the feature vectors of the respective patients, and outputs information on the extracted candidates for the clinical trial patients.

13. The selection support method according to claim 12, wherein for each of the patients, the degree of fitness for each of the criteria included in the selection criteria is calculated, and the degree of fitness for each of the criteria included in the selection criteria and the graph are converted into the feature vectors of the respective patients.

14. The selection support method according to claim 12 or 13, wherein the information on the treatment of each patient is converted into feature quantities using a language model, and the degree of fitness is estimated based on the converted feature quantities.

15. The selection support method according to claim 14, wherein the degree of fitness is calculated based on the distance between the feature vector converted from the selection criteria and the feature vector converted from the information on the treatment of each patient.

16. The selection support method according to any one of claims 12 to 15, wherein the degree of fitness and the graph are converted into the feature vectors of the respective patients using a conversion model that converts the degree of fitness and the graph with respect to the selection criteria into one feature vector.

17. The method for selection support according to any one of claims 12 to 16, wherein attributes of each patient and data related to the treatment are converted into features using a language model, and the graph is generated based on the similarity between patients calculated from the features of each patient.

18. The method for selection support according to any one of claims 12 to 17, further outputting a graph showing the relationship between the patients.

19. The method for selection support according to any one of claims 12 to 18, further outputting the degree of fitness of each patient with respect to the selection criteria.

20. A recording medium non-temporarily recording a selection support program that causes a computer to execute a process of calculating the degree of fitness of information related to the treatment of each patient with respect to the selection criteria for clinical trial patients, a process of generating a graph showing the relationship between patients based on the attributes of each patient and the data related to the treatment, a process of converting the degree of fitness with respect to the selection criteria and the graph into feature vectors of each patient, a process of extracting, as candidates for the clinical trial patients, patients who meet the clinical trial from among the patients based on the feature vectors of each patient, and a process of outputting information related to the extracted candidates for the clinical trial patients.

Images