Information processing device, information processing method, and recording medium

Abstract

In this information processing device, a document feature vector conversion means converts a plurality of pieces of medical record information into a document feature vector for each patient by natural language processing. A similarity level calculation means calculates levels of similarity among the patients from the document feature vectors. A graph construction means constructs a graph including nodes representing the patients and edges each connecting the nodes on the basis of the levels of similarity. A node feature vector calculation means calculates node feature vectors of the nodes on the basis of the graph. The information processing device can assist decision making of a user.

Description

Information processing device, information processing method, and recording medium

[0001] The present disclosure relates to prediction techniques using graph data.

[0002] Conventionally, prediction techniques using graph data have been known. For example, Non-Patent Document 1 proposes a method for estimating node attributes from graph data including nodes and edges by using EP (Embedding Propagation). In Non-Patent Document 1, graph data is generated based on the type of admission, location of admission, and initial diagnosis of a patient, and EP is used to estimate three attributes of ICU-admitted patients: in-hospital mortality rate, length of stay, and discharge destination.

[0003] Brandon Malone, Alberto Garcia-Duran, Mathias Niepert, "Learning Representations of Missing Data for Predicting Patient Outcomes" 2018, arXiv:1811.04752v1.

[0004] In Non-Patent Document 1, graph data is generated by selecting multiple attributes that represent the characteristics of patients and calculating the similarity between patients. Because the selected attributes only represent a small portion of the patient's condition, the graph data in Non-Patent Document 1 does not reflect the patient's comprehensive condition. Therefore, the method in Non-Patent Document 1 requires an expert to select patient attributes each time according to the purpose, which is costly and time-consuming.

[0005] One object of the present disclosure is to provide an information processing device capable of generating feature vectors to be used in downstream tasks.

[0006] In one aspect of the present disclosure, an information processing device comprises: a document feature vector conversion means for converting multiple pieces of medical record information into a document feature vector for each patient using natural language processing; a similarity calculation means for calculating similarities between patients from the document feature vectors; a graph construction means for constructing a graph including nodes representing patients and edges connecting the nodes based on the similarities; and a node feature vector calculation means for calculating node feature vectors of the nodes based on the graph.

[0007] In another aspect of the present disclosure, an information processing method includes: performing document feature vector conversion, which converts multiple pieces of medical record information into a document feature vector for each patient using natural language processing; performing similarity calculation, which calculates similarities between patients from the document feature vectors; performing graph construction, which constructs a graph including nodes representing patients and edges connecting the nodes, based on the similarities; and performing node feature vector calculation, which calculates node feature vectors of the nodes based on the graph.

[0008] In yet another aspect of the present disclosure, a recording medium records a program that causes a computer to execute the following processes: performing document feature vector conversion to convert multiple pieces of medical record information into document feature vectors for each patient using natural language processing; performing similarity calculation to calculate similarities between patients from the document feature vectors; performing graph construction to construct a graph including nodes representing patients and edges connecting the nodes based on the similarities; and performing node feature vector calculation to calculate node feature vectors of the nodes based on the graph.

[0009] According to the present disclosure, it is possible to provide an information processing device capable of generating feature vectors to be used in downstream tasks.

[0010] 1 is a diagram conceptually illustrating an information processing device; FIG. 2 is a block diagram illustrating a hardware configuration of the information processing device; FIG. 3 is a block diagram illustrating a functional configuration of the information processing device; FIG. 4 is an example of a property graph; FIG. 5 is a flowchart of a node feature vector calculation process; FIG. 6 is a block diagram illustrating a functional configuration of an information processing device of Modification 1; FIG. 7 is a block diagram illustrating a functional configuration of an information processing device of Modification 2; FIG. 8 is a diagram for explaining a use case; FIG. 9 is a block diagram illustrating a functional configuration of an information processing device of a second embodiment; and FIG. 10 is a flowchart of processing by the information processing device of the second embodiment.

[0011] Preferred embodiments of the present disclosure will be described below with reference to the drawings. First Embodiment Overall Configuration FIG. 1 is a conceptual diagram of an information processing device. The information processing device 10 calculates a feature vector for each patient based on electronic medical records (medical document data) and medical information. The medical information in this embodiment is assumed to be information related to clinical trials, such as patient test results and medication information. The information processing device 10 then uses the calculated feature vector in downstream tasks. The downstream tasks include, for example, attribute estimation using regression analysis and classification.

[0012] 2 is a block diagram showing the hardware configuration of the information processing device 10. As shown in the figure, the information processing device 10 includes a processor 11, an interface (IF) 12, a read-only memory (ROM) 13, a random access memory (RAM) 14, a storage device 15, and an input unit 16. The components are connected to each other via a bus 18, for example.

[0013] The processor 11 is a computer such as a CPU (Central Processing Unit), and executes a program prepared in advance to control the entire information processing device 10. Specifically, the processor 11 may be a CPU, a GPU (Graphics Processing Unit), a DSP (Digital Signal Processor), an MPU (Micro Processing Unit), an FPU (Floating Point number Processing Unit), a PPU (Physics Processing Unit), a TPU (Tensor Processing Unit), a quantum processor, a microcontroller, or a combination thereof.

[0014] The processor 11 also loads programs stored in the ROM 13, the storage device 15, etc., and executes each process coded in the program. The processor 11 also functions as part or all of the information processing device 10. The processor 11 then executes the node feature vector calculation process described below.

[0015] The IF 12 inputs and outputs data to and from external devices. Specifically, electronic medical records and medical information are input to the information processing device 10 through the IF 12.

[0016] The ROM 13 stores various programs executed by the processor 11. The RAM 14 is used as a working memory while the processor 11 is executing various processes.

[0017] The storage device 15 is a non-volatile, non-transitory storage device such as a disk-shaped recording medium or a semiconductor memory. The storage device 15 may be configured to be detachable from the information processing device 10. The storage device 15 stores various programs executed by the processor 11. The storage device 15 may also store electronic medical records and medical information input from an external device.

[0018] The input unit 16 is, for example, a mouse, a keyboard, etc., and is used by the user to input data.

[0019] 3 is a block diagram showing the functional configuration of the information processing device 10 according to the first embodiment. Functionally, the information processing device 10 includes an electronic medical record combining unit 111, a document feature vector calculation unit 112, a similarity calculation unit 113, a graph construction unit 114, a node feature vector calculation unit 115, and a downstream task processing unit 116.

[0020] An electronic medical record is input to the information processing device 10 via the IF 12. The electronic medical record is input to an electronic medical record combining unit 111 and a graph constructing unit 114. In addition, medical information of a patient is input to the information processing device 10 via the IF 12. The medical information of a patient is input to the graph constructing unit 114.

[0021] The electronic medical record combining unit 111 outputs the input electronic medical records to the document feature vector calculation unit 112. The electronic medical records are assumed to comprehensively include information about the patient and their condition. If there are multiple electronic medical records for the same patient, the electronic medical record combining unit 111 combines the multiple electronic medical records into one. The electronic medical record combining unit 111 then outputs the combined electronic medical record to the document feature vector calculation unit 112.

[0022] The document feature vector calculation unit 112 receives the electronic medical record of each patient from the electronic medical record combination unit 111. The document feature vector calculation unit 112 converts each patient's electronic medical record into a feature vector using natural language processing technology. The feature vector for the electronic medical record is hereinafter also referred to as a "document feature vector." Note that an example of a natural language processing technology used for vectorization is BERT (Bidirectional Encoder Representations from Transformers). The document feature vector calculation unit 112 outputs the document feature vector of each patient to the similarity calculation unit 113.

[0023] The similarity calculation unit 113 receives the document feature vector of each patient from the document feature vector calculation unit 112. The similarity calculation unit 113 receives the document feature vectors as input to a similarity calculation function and calculates the similarity between the document feature vectors. The similarity calculation function is a function that derives the similarity between a first document feature vector and a second document feature vector, and may be, for example, cosine similarity or Euclidean distance. The similarity between document feature vectors will hereinafter also be referred to as "similarity between patients." The similarity calculation unit 113 outputs the calculated similarity to the graph construction unit 114.

[0024] The graph construction unit 114 receives the similarity between patients from the similarity calculation unit 113. The graph construction unit 114 also receives the electronic medical records and the patient's medical information. The electronic medical records and the patient's medical information input to the graph construction unit 114 are structured data and include patient attributes. The patient attributes are, for example, composed of pairs of attribute names and attribute values.

[0025] The graph construction unit 114 constructs a property graph in which patients are treated as nodes and similar nodes are connected by edges based on the similarities between patients. For example, if the cosine similarity between patients is equal to or greater than a predetermined threshold, the graph construction unit 114 connects the nodes representing those patients with an edge. On the other hand, if the cosine similarity between patients is less than the predetermined threshold, the graph construction unit 114 does not connect the nodes representing those patients with an edge. Furthermore, the nodes in the property graph have the attributes of the patient corresponding to that node. The graph construction unit 114 extracts patient attributes from the electronic medical records and medical information and adds the extracted attributes to the corresponding nodes.

[0026] FIG. 4 is an example of a property graph constructed by the graph construction unit 114. The property graph includes a patient node 41, an edge 42, and an attribute 43. The patient node 41 represents a patient. The edge 42 indicates the connection between the patient nodes. In FIG. 4, similar patient nodes are connected by an edge. The attribute 43 indicates the patient attribute of the patient node 41. In FIG. 4, the patient node 41 has attributes (pairs of attribute name and attribute value) such as "(Age, 2)" and "(Gender, Female)." The above attributes are extracted from structured data such as electronic medical records or medical information and added to the attribute 43.

[0027] Returning to FIG. 3, the graph constructing unit 114 outputs the constructed property graph to the node feature vector calculating unit 115 .

[0028] The property graph is input to the node feature vector calculation unit 115 from the graph construction unit 114. The node feature vector calculation unit 115 calculates the feature vector of each node included in the property graph using graph neural network technology such as EP. The feature vector of a node is hereinafter also referred to as a "node feature vector."

[0029] For example, the EP exchanges messages via edges in the property graph to learn the vector representation of the attributes of each node. If a node has missing attribute values, the EP can learn a vector representation of the missing data based on the attribute values ​​of adjacent nodes. The EP then combines the learned vector representations of the attributes to calculate the vector representation of the node (node ​​feature vector). Note that vector 44 in Figure 4 represents the node feature vector of the patient node 41.

[0030] The node feature vector calculation unit 115 outputs the node feature vector of each patient to the downstream task processing unit 116 .

[0031] The node feature vector of each patient is input to the downstream task processing unit 116. The downstream task processing unit 116 uses the node feature vector of each patient to execute downstream tasks such as attribute estimation.

[0032] In the above configuration, the electronic medical record combination unit 111 and the document feature vector calculation unit 112 are examples of a document feature vector conversion means, the similarity calculation unit 113 is an example of a similarity calculation means, the graph construction unit 114 is an example of a graph construction means, the node feature vector calculation unit 115 is an example of a node feature vector calculation means, and the downstream task processing unit 116 is an example of a patient selection means and an attribute estimation means.

[0033] [Node Feature Vector Calculation Processing] Next, a description will be given of information processing by the information processing device 10. Fig. 5 is a flowchart of node feature vector calculation processing by the information processing device 10. This processing is realized by the processor 11 shown in Fig. 2 executing a program prepared in advance and operating as each element shown in Fig. 3.

[0034] First, an electronic medical record and medical information of a patient are input to the information processing device 10 via the IF 12. The electronic medical record is input to the electronic medical record combining unit 111 and the graph constructing unit 114. The medical information of the patient is input to the graph constructing unit 114 (step S111).

[0035] Next, the electronic medical record combining unit 111 outputs the input electronic medical records to the document feature vector calculation unit 112. If there are multiple electronic medical records for the same patient, the electronic medical record combining unit 111 combines the multiple electronic medical records into one and outputs the combined electronic medical record to the document feature vector calculation unit 112 (step S112).

[0036] Next, the document feature vector calculation unit 112 converts each patient's electronic medical record into a document feature vector using natural language processing technology (step S113). The document feature vector calculation unit 112 outputs each patient's document feature vector to the similarity calculation unit 113. Next, the similarity calculation unit 113 inputs the document feature vectors into a similarity calculation function and calculates the similarity between patients (step S114). The similarity calculation unit 113 outputs the calculated similarity to the graph construction unit 114.

[0037] Next, the graph construction unit 114 constructs a property graph based on the similarities between patients, the electronic medical records, and the medical information (step S115). Specifically, the graph construction unit 114 constructs a property graph in which patients are treated as nodes and similar nodes are connected by edges based on the similarities between patients. The graph construction unit 114 also extracts patient attributes from the electronic medical records and medical information and adds the extracted attributes to the corresponding nodes. The graph construction unit 114 outputs the constructed property graph to the node feature vector calculation unit 115.

[0038] Next, the node feature vector calculation unit 115 calculates a node feature vector for each node included in the property graph using a graph neural network technique such as EP (step S116). The node feature vector calculation unit 115 outputs the calculated node feature vector to the downstream task processing unit 116. The downstream task processing unit 116 uses the node feature vector to execute downstream tasks such as attribute estimation. Then, the processing ends.

[0039] [Modifications] Next, modifications of the present embodiment will be described. The following modifications may be applied to the above embodiment in any combination.

[0040] 6 shows the functional configuration of an information processing device 10a according to Modification 1. Modification 1 differs from the information processing device 10 in that it includes an electronic medical record portion extraction unit 111a instead of the electronic medical record combination unit 111.

[0041] The electronic medical record portion extraction unit 111a extracts predetermined information from the electronic medical record and outputs it to the document feature vector calculation unit 112. The predetermined information is, for example, information predetermined by the user. The user can instruct the information processing device 10a to extract only the necessary information from the electronic medical record depending on the content of the downstream task. In this way, by using a portion of the information in the electronic medical record, the information processing device 10a can calculate a more appropriate node feature vector.

[0042] 7 shows the functional configuration of an information processing device 10b according to Modification 2. The information processing device 10b according to Modification 2 differs from the information processing device 10 in that it calculates the similarity between patients using electronic medical records and medical information in addition to document feature vectors.

[0043] The similarity calculation unit 113b receives the document feature vector of each patient from the document feature vector calculation unit 112. The similarity calculation unit 113b also receives the electronic medical record and the patient's medical information. The electronic medical record and the patient's medical information are structured data and include patient attributes. Hereinafter, the electronic medical record and the patient's medical information will also be referred to as "patient attribute data."

[0044] The similarity calculation unit 113b calculates the similarity between patients based on the document feature vector and attribute data of each patient. For example, the similarity calculation unit 113b converts each patient's attribute data into a feature vector using natural language processing technology. The feature vector for the patient's attribute data is hereinafter also referred to as an "attribute feature vector." The similarity calculation unit 113b then generates a composite feature vector by concatenating the document feature vector and the attribute feature vector of the same patient. The similarity calculation unit 113b inputs a group of composite feature vectors into a similarity calculation function and calculates the similarity between the composite feature vectors to determine the similarity between the patients.

[0045] The method for calculating the similarity between patients is not limited to the above. For example, the similarity calculation unit 113b may calculate the similarity between the attribute data of each patient and integrate it with the similarity between the document feature vectors of each patient to calculate the similarity between patients.

[0046] In this way, by using the attribute data of patients when calculating the similarity between patients, it is possible to calculate a more appropriate similarity between patients.

[0047] [Use Case] ​​Next, a description will be given of a use case of the first embodiment. The information processing device 10 of this embodiment can be used to select subject candidates for clinical trials.

[0048] The user specifies the attributes of the subject to the information processing device 10. The attribute may be, for example, the efficacy of a new drug. The downstream task processing unit 116 of the information processing device 10 estimates the specified attribute using the node feature vector of each patient input from the node feature vector calculation unit 115. Then, based on the attribute estimation results, the information processing device 10 can extract, for example, a patient group for which the new drug is effective.

[0049] The information processing device 10 may construct a property graph for each clinical trial and calculate a node feature vector. For example, the information processing device 10 adds information related to the clinical trial to the attributes of each node. FIG. 8 is a diagram illustrating a use case. In FIG. 8, the information processing device 10 adds, as information related to the clinical trial, the matching score between the clinical trial requirements and the electronic medical record, and the feature vector of the examination image to the node attributes, and calculates the node feature vector for each patient. Then, the information processing device 10 uses the node feature vector for each patient to estimate the efficacy of the new drug for each patient and extracts a patient group for whom the new drug is effective and a patient group for whom the standard drug is effective. This allows the user to efficiently select candidate subjects for the clinical trial.

[0050] 9 is a block diagram showing the functional configuration of an information processing apparatus according to the second embodiment. The information processing apparatus 200 includes a document feature vector conversion unit 201, a similarity calculation unit 202, a graph construction unit 203, and a node feature vector calculation unit 204.

[0051] 10 is a flowchart of processing by the information processing apparatus of the second embodiment. The document feature vector conversion means 201 converts multiple pieces of medical record information into document feature vectors for each patient using natural language processing (step S201). The similarity calculation means 202 calculates similarities between patients from the document feature vectors (step S202). The graph construction means 203 constructs a graph including nodes representing patients and edges connecting the nodes based on the similarities (step S203). The node feature vector calculation means 204 calculates node feature vectors for the nodes based on the graph (step S204).

[0052] According to the information processing device 200 of the second embodiment, it is possible to provide an information processing device capable of generating feature vectors to be used in downstream tasks. Furthermore, the information processing device 200 can support user decision-making.

[0053] A part or all of the above-described embodiments can be described as, but not limited to, the following supplementary notes.

[0054] (Supplementary Note 1) An information processing device comprising: a document feature vector conversion means for converting multiple pieces of medical record information into a document feature vector for each patient using natural language processing; a similarity calculation means for calculating similarities between patients from the document feature vectors; a graph construction means for constructing a graph including nodes representing patients and edges connecting the nodes based on the similarities; and a node feature vector calculation means for calculating node feature vectors of the nodes based on the graph.

[0055] (Supplementary Note 2) The information processing device according to Supplementary Note 1, further comprising: an attribute information acquisition unit that acquires attribute information of a patient; and the graph construction unit that associates the attribute information with a node that represents the patient.

[0056] (Supplementary Note 3) The information processing device according to Supplementary Note 2, wherein the similarity calculation means calculates similarity between patients based on the document feature vector and the attribute information.

[0057] (Supplementary Note 4) The information processing device according to Supplementary Note 2, further comprising: an attribute estimation means for estimating attribute information of a patient based on the node feature vector.

[0058] (Appendix 5) An information processing device according to appendix 4, comprising an extraction means for extracting patients having predetermined attribute information, wherein the attribute estimation means estimates the attribute information of the patient designated by the user based on the user's designation, and the extraction means extracts a group of patients having the predetermined attribute information based on the estimation result.

[0059] (Supplementary Note 6) The information processing device according to Supplementary Note 1, further comprising: a combining unit that combines a plurality of pieces of medical record information of the same patient; and the document feature vector converting unit that converts the combined medical record information into a document feature vector.

[0060] (Supplementary Note 7) The information processing device according to Supplementary Note 1, further comprising: an attribute information extraction unit that extracts predetermined attribute information from medical record information; and the document feature vector conversion unit that converts the extracted attribute information into a document feature vector.

[0061] (Supplementary Note 8) An information processing method that performs document feature vector conversion to convert multiple pieces of medical record information into document feature vectors for each patient using natural language processing, performs similarity calculation to calculate similarities between patients from the document feature vectors, performs graph construction to construct a graph including nodes representing patients and edges connecting the nodes based on the similarities, and performs node feature vector calculation to calculate node feature vectors of the nodes based on the graph.

[0062] (Supplementary Note 9) The information processing method according to Supplementary Note 8, further comprising: acquiring attribute information of a patient; and constructing the graph by linking the attribute information to a node representing the patient.

[0063] (Supplementary Note 10) The information processing method according to Supplementary Note 9, wherein the similarity calculation calculates a similarity between patients based on the document feature vector and the attribute information.

[0064] (Supplementary Note 11) The information processing method according to Supplementary Note 9, further comprising: performing attribute estimation to estimate attribute information of a patient based on the node feature vector.

[0065] (Appendix 12) An information processing method according to Appendix 11, further comprising: performing an extraction process to extract patients having predetermined attribute information; the attribute estimation process estimates attribute information of patients designated by a user based on the user's designation; and the extraction process extracts a group of patients having the predetermined attribute information based on the estimation result.

[0066] (Supplementary Note 13) The information processing method according to Supplementary Note 8, further comprising: performing a combining process for combining a plurality of medical record information of the same patient; and converting the document feature vector into the document feature vector of the combined medical record information.

[0067] (Supplementary Note 14) The information processing method according to Supplementary Note 8, further comprising: extracting predetermined attribute information from the medical record information; and converting the extracted attribute information into a document feature vector.

[0068] (Appendix 15) A recording medium having recorded thereon a program that causes a computer to execute the following processes: performing document feature vector conversion to convert multiple pieces of medical record information into document feature vectors for each patient using natural language processing; performing similarity calculation to calculate similarities between patients from the document feature vectors; performing graph construction to construct a graph including nodes representing patients and edges connecting the nodes based on the similarities; and performing node feature vector calculation to calculate node feature vectors of the nodes based on the graph.

[0069] (Supplementary Note 16) The recording medium according to Supplementary Note 15, further comprising: acquiring attribute information of a patient; and constructing the graph by linking the attribute information to a node representing the patient.

[0070] (Supplementary Note 17) The recording medium according to Supplementary Note 16, wherein the similarity calculation calculates a similarity between patients based on the document feature vector and the attribute information.

[0071] (Supplementary Note 18) The recording medium according to Supplementary Note 16, further comprising: performing attribute estimation to estimate attribute information of a patient based on the node feature vector.

[0072] (Appendix 19) The recording medium according to Appendix 18, further comprising: performing an extraction process for extracting patients having predetermined attribute information; the attribute estimation process estimates attribute information of patients designated by a user based on the user's designation; and the extraction process extracts a group of patients having the predetermined attribute information based on the estimation result.

[0073] (Supplementary Note 20) The recording medium according to Supplementary Note 15, further comprising: a combining process for combining a plurality of medical record information of the same patient; and the document feature vector conversion for converting the combined medical record information into a document feature vector.

[0074] (Supplementary Note 21) The recording medium according to Supplementary Note 15, further comprising: extracting predetermined attribute information from the medical record information; and converting the extracted attribute information into a document feature vector.

[0075] Although the present disclosure has been described above with reference to the embodiments and examples, the present disclosure is not limited to the above-described embodiments and examples. Various modifications that can be understood by a person skilled in the art can be made to the configuration and details of the present disclosure within the scope of the present disclosure.

[0076] 10 Information processing device 111 Electronic medical record combining unit 112 Document feature vector calculation unit 113 Similarity calculation unit 114 Graph construction unit 115 Node feature vector calculation unit 116 Downstream task processing unit

Claims

1. An information processing apparatus comprising: document feature vector conversion means for converting a plurality of medical record information into document feature vectors for each patient by natural language processing; similarity calculation means for calculating the similarity between patients from the document feature vectors; graph construction means for constructing a graph including nodes representing patients and edges connecting the nodes based on the similarity; and node feature vector calculation means for calculating node feature vectors of the nodes based on the graph.

2. The information processing apparatus according to claim 1, further comprising attribute information acquisition means for acquiring attribute information of a patient, wherein the graph construction means associates the attribute information with the node representing the patient.

3. The information processing apparatus according to claim 2, wherein the similarity calculation means calculates the similarity between patients based on the document feature vectors and the attribute information.

4. The information processing apparatus according to claim 2, further comprising attribute estimation means for estimating attribute information of a patient based on the node feature vectors.

5. The information processing apparatus according to claim 4, further comprising extraction means for extracting patients having predetermined attribute information, wherein the attribute estimation means estimates attribute information of a patient designated by a user based on the designation of the user, and the extraction means extracts a group of patients having the predetermined attribute information based on the estimation result.

6. The information processing apparatus according to claim 1, further comprising combination means for combining a plurality of medical record information of the same patient, wherein the document feature vector conversion means converts the combined medical record information into a document feature vector.

7. The information processing apparatus according to claim 1, further comprising attribute information extraction means for extracting predetermined attribute information from medical record information, wherein the document feature vector conversion means converts the extracted attribute information into a document feature vector.

8. Perform document feature vector conversion that converts a plurality of medical record information into document feature vectors for each patient by natural language processing, perform similarity calculation that calculates the similarity between patients from the document feature vectors, perform graph construction that constructs a graph including nodes representing patients and edges connecting the nodes based on the similarity, and perform node feature vector calculation that calculates node feature vectors of the nodes based on the graph. An information processing method.

9. Further perform attribute information acquisition that acquires attribute information of a patient, and the graph construction is the information processing method according to claim 8 that associates the attribute information with a node representing the patient.

10. The similarity calculation is the information processing method according to claim 9 that calculates the similarity between patients based on the document feature vector and the attribute information.

11. Further perform attribute estimation that estimates attribute information of a patient based on the node feature vector. The information processing method according to claim 9.

12. Further perform extraction processing that extracts patients having predetermined attribute information, the attribute estimation estimates attribute information of a patient specified by the user based on the user's specification, and the extraction processing extracts a group of patients having predetermined attribute information based on the estimation result. The information processing method according to claim 11.

13. Further perform combination processing that combines a plurality of medical record information of the same patient, and the document feature vector conversion is the information processing method according to claim 8 that converts the combined medical record information into a document feature vector.

14. Further perform attribute information extraction that extracts predetermined attribute information from medical record information, and the document feature vector conversion is the information processing method according to claim 8 that converts the extracted attribute information into a document feature vector.

15. Perform document feature vector conversion that converts a plurality of medical record information into document feature vectors for each patient by natural language processing, perform similarity calculation that calculates the similarity between patients from the document feature vectors, perform graph construction that constructs a graph including nodes representing patients and edges connecting the nodes based on the similarity, and record a program for causing a computer to execute a process of performing node feature vector calculation that calculates node feature vectors of the nodes based on the graph on a recording medium.

16. Further perform attribute information acquisition that acquires attribute information of a patient, and the graph construction is the recording medium according to claim 15, which associates the attribute information with a node representing the patient.

17. The similarity calculation is the recording medium according to claim 16, which calculates the similarity between patients based on the document feature vectors and the attribute information.

18. Further perform attribute estimation that estimates the attribute information of a patient based on the node feature vector, and the recording medium according to claim 16.

19. Further perform extraction processing that extracts patients having predetermined attribute information, the attribute estimation estimates the attribute information of the patient designated by the user based on the designation of the user, and the extraction processing extracts a group of patients having predetermined attribute information based on the estimation result, and the recording medium according to claim 18.

20. Further perform combination processing that combines a plurality of medical record information of the same patient, and the document feature vector conversion is the recording medium according to claim 15, which converts the combined medical record information into a document feature vector.

Images