JP2020042758A - Device and method for supporting guidance of medication - Google Patents

Abstract

To provide a device and a method for supporting guidance of medication which can teach a patient to take medicines properly by extracting necessary information for the patient from information on medicines.SOLUTION: The device for supporting guidance of medication includes: a vector generation unit 16 for generating a patient vector and a plurality of character-string vectors on the basis of patient data on patients and plural pieces of character-strings for guidance of medication according to medicines; a learning model generation unit 17 for generating a learning model on which learning processing has been performed so that the learning model will output a predetermined character-string vector in response to input of the patient vector and the plurality of character-string vectors; and a character-string extraction unit 19 for extracting a character string corresponding to the character-string vector output from the learning model in response to input of new medicine and a new patient vector.SELECTED DRAWING: Figure 3

Description

  The present disclosure relates to a medication instruction support device and a medication instruction support system that support creation of a medication instruction document to be provided to a patient together with a drug.

  When a pharmacist delivers a medicine to a patient based on a prescription, it is obliged to provide information on the medicine. Such an act is generally called medication instruction. The information on the medicine includes various information such as a medicinal effect, a medicine taking method, a storage method, significance of the medicine taking, contraindications, and the like. In the medication instruction by a pharmacist, it is not necessary that all the information on the medicine be transmitted to the patient, and it is desirable that only the information necessary for the patient is extracted and provided.

  At the time of medication instruction, a pharmacist who has received a prescription from a patient needs to prepare a medicine based on the prescription and quickly extract necessary information for the patient from various information related to the medicine. However, it is not easy to quickly extract information necessary for a patient from various kinds of information on medicines, and depending on the skill of a pharmacist, the patient may not always be able to perform optimal medication instruction. Under these circumstances, it is demanded to extract information necessary for a patient from various information regardless of the skill of a pharmacist, and to provide suitable medication instruction.

  As a system for supporting medication instruction, for example, there is a technique disclosed in Patent Document 1. Patent Literature 1 discloses a medication instruction support system capable of quickly outputting optimal medication instruction policy data to be performed on a patient and the certainty thereof together with a rule ID for medication instruction.

JP-A-8-272882

  According to the technology disclosed in Patent Document 1, a medication instruction policy is inferred based on various data concerning a patient, and certainty data indicating the validity of the inferred medication instruction policy is generated. Has not been considered. For this reason, if the inferred medication instruction policy is not appropriate, an appropriate medication instruction policy cannot be output.

  An object of the present disclosure is to provide a medication instruction support device and a medication instruction support system that can extract information necessary for a patient from various information related to medicine and can perform medication instruction suitable for the patient. .

  A medication instruction support device of the present invention is a vector generation unit that generates a patient vector and a plurality of character string vectors based on patient data on a patient and a plurality of character strings for medication instruction corresponding to a medicine, A learning model generating unit configured to generate a learning model subjected to a learning process so as to output a predetermined character string vector with respect to the input of the plurality of character string vectors, and the plurality of character strings using the learning model. And a character string extracting unit that extracts a predetermined character string from the character string.

  A medication support system according to the present invention is a vector generation unit that generates a patient vector and a plurality of character string vectors based on patient data relating to a patient and a plurality of character strings for medication guidance corresponding to medicines, A learning model generating unit configured to generate a learning model subjected to a learning process so as to output a predetermined character string vector with respect to the input of the plurality of character string vectors, and the plurality of character strings using the learning model. And a character string extraction unit that extracts a predetermined character string from the character string corresponding to the predetermined character string based on whether or not the extracted predetermined character string is used in a medication instruction document created by a pharmacist. A plurality of medication instruction support devices having a weighting unit for determining a weight value for the character string vector; a server device; and the plurality of medication instruction support devices. A predetermined network connected to the server device, wherein the server device is used in each of the medication support devices based on the weight values transmitted from the plurality of medication support devices. Generate or update the learning model.

  ADVANTAGE OF THE INVENTION According to this invention, the information required for a patient can be extracted from the various information regarding a medicine, and the medication instruction | indication suitable for a patient can be performed.

The figure which shows the example of a structure of the medication support system containing a medication support apparatus. The figure which shows an example of the hardware constitutions of the medication instruction support device FIG. 2 is a block diagram illustrating an example of a functional configuration of a medication instruction support device according to the first embodiment. Figure illustrating the attached document Diagram illustrating generation of patient vector based on patient data Diagram illustrating generation of instruction sentence vector based on instruction sentence Figure exemplifying the display screen by the guidance sentence display unit Diagram for explaining an example of weighting by a weighting unit FIG. 9 is a block diagram showing an example of a functional configuration of a medication instruction support device according to the second embodiment. FIG. 9 is a block diagram showing an example of a functional configuration of a medication support system according to a third embodiment. The figure which illustrated the teacher data used in 3rd Embodiment Conceptual diagram of a learning model according to the third embodiment The figure which illustrated the teacher data in 4th Embodiment. FIG. 13 is a block diagram showing a functional configuration of a medication instruction support device according to a fifth embodiment. FIG. 13 is a block diagram showing a functional configuration of a medication instruction support device according to a sixth embodiment. The figure which shows the example of the collection data for producing | generating the teacher data of 6th Embodiment. The figure which shows the example of the prescription teacher data generated based on the collection data shown in FIG. The figure which shows the example of the prescription teacher data generated based on the collection data shown in FIG. The figure which shows the example of the prescription teacher data generated based on the collection data shown in FIG. Diagram showing another example of prescription teacher data Diagram showing another example of prescription teacher data Diagram showing a display example of a button for displaying a check column and a guidance sentence

  Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. However, an unnecessary detailed description, for example, a detailed description of a well-known matter or a redundant description of substantially the same configuration may be omitted. The following description and the referenced drawings are provided to enable those skilled in the art to understand the present invention, and not to limit the scope of the present invention.

(First Embodiment)
First, a medication instruction support device 10 according to a first embodiment of the present invention will be described.

<System configuration>
FIG. 1 is a diagram illustrating a configuration example of a medication instruction support system 1 including a medication instruction support device 10. As shown in FIG. 1, the medication support system 1 includes a medication support device 10 installed in a pharmacy P, a medication support server 20 connected to the medication support device 10 via a communication network N, Having. As shown in FIG. 1, in the medication support system 1, it is assumed that a plurality of medication support devices 10 are installed in a plurality of pharmacies P, respectively. The present invention does not particularly limit the number of the medication instruction support devices 10 included in the medication instruction support system 1 and the number of pharmacies where the medication instruction support devices 10 are installed.

  The medication instruction support device 10 is a terminal operated by a pharmacist working at the pharmacy P, and is a general computer such as a personal computer (PC) or a tablet terminal. The medication support device 10 provides support for the pharmacist to prepare a medication guidance document to be provided to the patient together with the drug, based on the information on the prescription of the patient who has visited the pharmacy P and the information on the patient. In the present specification, the medication instruction means that a pharmacist provides a patient with information on a drug prescribed to the patient, and the medication instruction document is a document containing information on the drug. Etc. means a document in which a character string is described. Hereinafter, a character string such as a sentence described in the medication instruction document is referred to as an instruction sentence.

  The medication guidance support server 20 is connected to the medication guidance support devices 10 installed in a plurality of pharmacies P via a communication network N such as a LAN (Local Area Network), a WAN (Wide Area Network), or an Internet line. Server device. The medication administration support server 20 is, for example, a cloud server. The medication support server 20 receives the information used at the time of creating the medication guidance document in the plurality of medication support devices 10 from the medication support device 10 via the communication network N, and stores the information in a storage device (not shown). In addition, in response to a request from the medication guidance support device 10, the medication guidance support server 20 transmits the information used when the medication guidance document is created in the medication guidance support device 10 to the medication guidance support device 10 via the communication network N. Send to

<Configuration of medication compliance support device 10>
FIG. 2 is a diagram illustrating an example of a hardware configuration of the medication instruction support device 10.

  The medication taking support device 10 is, for example, a computer including a CPU 101, a ROM 102, a RAM 103, an external storage device 104, a communication interface 105, an input device 106, an output device 107, and the like. Examples of the external storage device 104 include an HDD, an SSD, and a flash memory. Examples of the communication interface 105 include a communication controller for a LAN line. Examples of the input device 106 include a keyboard, a mouse, a touch panel, a scanner, and a barcode reader. Examples of the output device 107 include a display device such as a CRT and a liquid crystal, and a printer.

  Each function described later of the medication instruction support device 10 is realized by, for example, the CPU 101 referring to a processing program and various data stored in the ROM 102, the RAM 103, the external storage device 104, and the like. However, some or all of the above-described functions may be realized by processing by a dedicated hardware circuit instead of or together with the processing by software.

  FIG. 3 is a block diagram illustrating an example of a functional configuration of the medication instruction support device 10 according to the first embodiment.

  As shown in FIG. 3, the medication instruction support device 10 includes a prescription data acquisition unit 11, a medicine information document data acquisition unit 12, a medicine information document database (DB) 13, a patient data acquisition unit 14, a patient DB 15, a vector generation unit 16. , A learning model generation unit 17, a learning model DB 18, a guidance sentence extraction unit 19, a ranking unit 110, a guidance sentence display unit 111, a selection reception unit 112, a medication instruction document data generation unit 113, and a weighting unit 114.

  As shown in FIG. 1, the medication instruction support device 10 is installed in a pharmacy P, and is operated by a pharmacist working at the pharmacy P. When a patient having a prescription obtained from a doctor visits the pharmacy P, the pharmacist uses the input device 106 (see FIG. 2) to take information on the prescription described in the prescription (hereinafter, referred to as prescription data). Input to the guidance support device 10. The method of inputting prescription data to the medication instruction support device 10 is, for example, a method in which a pharmacist directly inputs using a mouse, a keyboard, a touch panel, and the like, and a barcode or a two-dimensional code that represents prescription data described in a prescription. A method of scanning with a scanner (reader device) or the like can be adopted.

  The prescription data acquisition unit 11 acquires the input prescription data. The prescription data includes, for example, information on the patient such as the name of the patient having the prescription, date of birth, gender, insurer number, and the like, and the name, amount, and use site of the drug prescribed for the patient. Information about the medicine, such as the time of use and the amount used, is included.

  The medicine information document data obtaining unit 12 obtains, from the medicine information document DB 13, data on all guidance sentences corresponding to the medicine prescribed to the patient based on the prescription data. The guidance sentence is a sentence described in a medicine information document for providing information on a medicine. A drug information document is a package insert that describes the items specified in Article 52 of the Pharmaceutical Affairs Law, a medicine bookmark (registered trademark), an optional drug information report, etc., for example, from a drug manufacturer that manufactures drugs. Or provided by volunteer pharmacists.

  The medicine information document includes a plurality of items, and each item may include a small item. Each item or sub-item of the drug information document includes one or more sentences or a character string such as a list of words. The character string included in the medicine information document is described as a guidance sentence in this specification. This instruction sentence is an example of a character string for medication instruction in the present invention. Package inserts as examples of drug information documents include, for example, `` name '', `` contraindications '', `` efficacy / effect '', `` dosage / dose '', `` careful administration '', `` important basic precautions '', `` interactions , "Side effects" and the like. Each of these items may include a small item, an instruction sentence explaining the nature of the drug, and / or an instruction sentence on how to use the drug, precautions, and the like.

  FIG. 4 is a diagram illustrating an example of an attached document. In the example shown in FIG. 4, each of the items such as “contraindications”, “cautious administration”, “important basic precautions”, “interaction”, “side effects”, etc., has a plurality of instruction sentences ((1 ), (2), (3), etc.). Since these plural instruction sentences include contents that are not necessary for a specific patient, the pharmacy P prepares a medication instruction document provided for medication instruction for a specific patient. It is desirable that a guidance sentence of the content necessary for the patient be extracted.

  Therefore, the medication instruction support device 10 according to the present embodiment extracts and provides an instruction sentence including information necessary for a patient, as described below. A specific method of extracting a guidance sentence including information necessary for a patient will be described later.

  Returning to the description of FIG. The patient data acquisition unit 14 acquires information about a patient (hereinafter, referred to as patient data) based on the prescription data and the patient DB 15. The patient data may include information about the patient, such as the patient's name, date of birth, gender, disease, and insurer number, and drug history information, such as past dispensing contents and guidance history for the patient. The patient DB 15 is a database in which information on the patient and drug history information of the patient are associated with each other. However, the patient data used in the medication instruction support device 10 does not need to include all information about the patient and the drug history information of the patient, and may use only a part of the information. In the present embodiment, the patient data includes at least information such as gender and age. When a new patient who is not registered in the patient DB 15 visits the pharmacy P, the pharmacist may perform, for example, an operation of listening to the patient data from the patient and registering the new patient data in the patient DB 15.

  The vector generation unit 16 generates a guidance sentence vector and a patient vector based on the data related to the guidance sentence acquired by the medicine information document data acquisition unit 12 and the patient data acquired by the patient data acquisition unit 14. The guidance sentence vector is obtained by vector-converting each guidance sentence included in the medicine information document, and the patient vector is obtained by vector-converting the contents of each item of the patient data. As a specific example of the vector conversion of the guidance sentence and the patient data, for example, Bag of Words for counting the number of appearances of a word included in a sentence or patient data, Word2Vec for performing a vector conversion of the word itself, or the like can be adopted. For more accurate vector conversion, a learning model learned using a medical term dictionary or the like may be used. The instruction sentence vector is an example of the character string vector of the present invention.

  FIG. 5 is a diagram for describing an example of vector conversion of patient data and instruction text. FIG. 5A is a diagram illustrating generation of a patient vector based on patient data, and FIG. 5B is a diagram illustrating generation of a guidance sentence vector based on a guidance sentence.

  In FIG. 5A, the items of the patient data include “age”, “sex”, “pregnant woman”, “lactation”, “new”, “allergy”, “concomitant medication”, “hypertension”, “renal dysfunction”, “ Hepatic dysfunction "," The same guidance was given last time ", etc. are provided. The “data content” row indicates the content corresponding to each item, and the data content obtained by performing vector conversion is shown in the “vector value” row. In the example shown in FIG. 5A, excluding the age and gender, the vector value of the item whose data content is simply “Yes” or “YES” is “1”, and the vector value of the item whose data content is “No” or “NO” is The case where the vector value is “0” is illustrated. According to such a method, a patient vector having the same dimension as the number of items of the patient data is generated.

  Here, the patient vector is generated so as to have the same dimension as the number of items of the patient data, but the dimension may be extended for more unique expression. For example, if a technique called Entity Embeddings (Entity Embeddings of Categorical Variables) is used, each item of patient data can be extended to any dimension and expressed.

  FIG. 5B shows an example in which the instruction sentences (1) to (3) exemplified in the item “important basic caution” of the medicine information document illustrated in FIG. 4 are vector-converted. As shown in FIG. 5B, words such as “bilateral”, “renal artery stenosis”, “patient”, “use”, and “avoid” appear in the instruction sentence (1). The number of appearances of each word is counted, and the count value corresponds to the vector value. By such a method, a guidance sentence vector having the same dimension as the number of words is generated. Although FIG. 5B shows only a part of the words appearing in the guidance sentence, the number of appearances is actually counted for all the words.

  As a method of extracting a word from each instruction sentence, for example, a known morphological analysis technique can be applied. In addition, in order to prevent a word to be treated as one word from being divided into a plurality of words, it is desirable to perform morphological analysis using, for example, a medical term dictionary or the like. Specifically, for example, if the word “interstitial pneumonia” is divided into three words such as “between”, “characteristics”, “pneumonia”, etc., in order to generate an accurate instruction sentence vector, Since it is not preferable, there is an example in which “interstitial pneumonia” is prevented by using a medical term dictionary registered as one word.

  In the example shown in FIG. 5B, a method (Bag-of-Words) of using the number of appearances of each word as a vector value as it is is adopted, but the present invention is not limited to this, and another method is adopted. May be. For example, a method such as TF-IDF that evaluates more characteristic words in a sentence can be adopted. In addition, vector conversion may be performed based on the meaning of each word, such as Word2vec, and an instruction sentence vector may be generated based on the vectors of all words appearing in the instruction sentence.

  Returning to the description of FIG. The learning model generation unit 17 performs machine learning using the patient vector and the guidance sentence vector generated by the vector generation unit 16 as teacher data, and generates a learning model. More specifically, the teacher data used by the learning model generation unit 17 includes a patient vector corresponding to a plurality of patients and an instruction sentence vector corresponding to an instruction sentence provided to each patient in the past. This is the attached teacher data.

  In the first embodiment, the learning model generated by the learning model generating unit 17 is configured such that the instruction sentence vector corresponding to the instruction sentence provided to a certain patient is similar to the patient vector of the patient. This is a model that has been subjected to learning processing. At the same time, this learning model is a model that has been subjected to learning processing so that the instruction sentence vector corresponding to the instruction sentence not provided to a certain patient is not similar to the patient vector of the patient. In other words, the learning model generated by the learning model generation unit 17 rearranges and outputs the instruction sentence vectors in the order of the degree of similarity with the patient vector in response to the input of the patient vector and the plurality of instruction sentence vectors. Is a model that was learned as follows.

  The degree of similarity between the patient vector and the instruction sentence vector is determined, for example, by calculating the degree of similarity between the vectors. As a method of calculating the similarity, for example, a known method such as cosine similarity can be adopted. The cosine similarity is a value that is calculated using the inner product of the normalized vectors and indicates the closeness of the angle formed by the two vectors. In the cosine similarity, the closer the directions of the two vectors are, the closer the value is to 1, indicating that these vectors are similar. The learning model generated by the learning model generation unit 17 is registered in the learning model DB 18 and read out by the guidance sentence extraction unit 19 as necessary.

  A learning model newly created or updated by re-learning, which will be described later, is transmitted from the medication administration support server 20 (see FIG. 1) to the learning model DB 18 via the communication network N. Upon receiving a newly created or updated learning model, the learning model DB 18 updates the registered learning model with the newly received learning model as needed.

  It is desirable that the generation of the learning model by the learning model generation unit 17 is performed before the operation of the medication instruction support device 10 in the pharmacy P, for example. The generation does not have to be performed. That is, in the state where the medication instruction support device 10 is operating in the pharmacy P, the learning model is used as a new teacher data with the patient vector of the new patient and the instruction sentence vector of the instruction sentence created by the pharmacist for the patient as new teacher data. May be continuously updated (additional learning or re-learning is performed at an arbitrary timing).

  In the above description, the example in which the medication instruction support device 10 includes the learning model generation unit 17 has been described, but the learning model generation unit 17 may not be provided in the medication instruction support device 10. That is, the learning model generation unit 17 may exist in another device that can communicate via the communication network N, for example, in the medication support server 20. In this case, the medication instruction support device 10 transmits information necessary for model generation (patient vector and instruction sentence vector) to the medication instruction support server 20 via the communication network N to generate a learning model. May be adopted.

  Using the learning model read from the learning model DB 18, the instruction sentence extracting unit 19 extracts an instruction sentence to be included in the medication instruction document provided to the patient for each item in the drug information document. Specifically, the guidance sentence extracting unit 19 acquires a new patient vector and the guidance sentence vector of all the guidance sentences included in the medicine information document of the medicine prescribed for the patient from the vector generation unit 16, The instruction sentence vectors of all instruction sentences are classified into instruction sentence vectors for each item. Then, the instruction sentence extracting unit 19 inputs the patient vector and the instruction sentence vector for each item to the learning model, and obtains an output of the instruction sentence vector rearranged in descending order of the similarity with the patient vector. Then, the guidance sentence extracting unit 19 extracts a guidance sentence corresponding to a predetermined number (for example, one) of guidance instruction vectors that are higher than the output guidance sentence vector. Thereby, the guidance sentence extracting unit 19 can extract, from the guidance sentence included in the medicine information document, a guidance sentence that is highly relevant to the patient for each item. For example, in the case of the medicine information document illustrated in FIG. 4, the instruction sentence extracting unit 19 outputs the instruction sentence such as one item from the item “contraindications”, one item from “effect / effect”, and one item from “usage / dose”. Can be extracted.

  The ranking unit 110 uses the learning model read from the learning model DB 18 to rank the guidance sentences extracted by the guidance sentence extraction unit 19 for each item of the drug information document. Specifically, the ranking unit 110 inputs the patient vector and the instruction sentence vector extracted for each item to the learning model, and outputs the instruction sentence vector sorted in descending order of similarity with the patient vector. Get output. Accordingly, the ranking unit 110 can rank the instruction sentences of items that are more relevant to the patient in a higher rank.

  The guidance sentence display unit 111 causes the output device 107 (see FIG. 2) to display the guidance sentences for each item extracted by the guidance sentence extraction unit 19 in the order determined by the ranking unit 110. FIG. 6 is a diagram exemplifying a display screen of the guidance sentence display unit 111 when the output device 107 is a display device.

  FIG. 6 shows an example in which a guidance sentence display column 201 is displayed on the display screen 200. In the guidance sentence display column 201, guidance sentences of a plurality of items are displayed in the order of ranking. In FIG. 6, only items selected by, for example, a pharmacist are displayed by the filter function for each item. In the guidance sentence display column 201, a “keyword” column is provided, and the word on which the guidance sentence is extracted is displayed. As a method of extracting such a keyword, for example, an instruction sentence vector of an extracted instruction sentence and an item name (“sex”, “pregnant woman”, etc.) of patient data used for teacher data for generating a learning model are used. A method of calculating the similarity with the vector-converted one and using the item name having a high similarity with the instruction sentence vector as a keyword may be adopted. The extraction of the keyword using such a method may be executed by, for example, the instruction sentence display unit 111.

  The selection receiving unit 112 receives a selection of a plurality of instruction sentences displayed by the instruction sentence display unit 111 that the pharmacist determines to provide to the patient. The selection by the pharmacist may be performed by, for example, the input device 106 (mouse, keyboard, touch panel, etc.) shown in FIG.

  The medication instruction document data generating unit 113 generates medication instruction document data including the selected instruction based on the pharmacist's selection received by the selection receiving unit 112. The medication instruction document data is document data that describes the instruction sentence selected by the pharmacist among the instruction sentences extracted for each item by the instruction sentence extraction unit 19 and ranked by item by the ranking unit 110. . The medication instruction document data generating unit 113 outputs the generated medication instruction document data to the output device 107 (see FIG. 2, for example, a printer). As a result, for example, the medication instruction document data is printed and provided to the patient together with the medicine. The medication instruction document data may not be printed but may be displayed on, for example, a portable terminal of the patient.

  The medication instruction document data generation unit 113 may register the generated medication instruction document data in the patient DB 15 in association with the medication history data of the patient.

  In the present embodiment, the instruction text included in the medication instruction document data generated by the medication instruction document data generation unit 113 is a sentence (or a character string) based on an attached document, a bookmark, or a drug information provision book. Yes, it is not always easy for patients to understand. In order to provide a medication instruction document in which a sentence that is easy for the patient to understand is described, the medication instruction document data generation unit 113 may perform a colloquial style conversion process of converting the text format of the instruction sentence into, for example, a colloquial style.

  For the colloquial style conversion processing, for example, a machine learning model such as sequence to sequence (seq2seq) can be used. It is preferable that, for example, a medical term dictionary is used as teacher data for generating a learning model used in the colloquial style conversion process. By such a process, it is possible to provide the patient with a medication instruction document in which a plain and easy-to-understand instruction sentence is described. Here, in order to facilitate the design of the machine learning in the medication instruction support device 10, it is preferable to perform the colloquial style conversion processing one sentence at a time in the order of the ranking by the ranking unit 110.

  On the other hand, the weighting unit 114 sets a weighting value for the teacher data used when the learning model generation unit 17 generates the learning model based on the selection or non-selection of the guidance sentence by the pharmacist, which is received by the selection reception unit 112. I do. FIG. 7 is a diagram for describing an example of weighting by the weighting unit 114.

  FIG. 7 illustrates the teacher data before weighting and the teacher data after weighting by the weighting unit 114. In the example shown in FIG. 7, a case is shown where sentence 1 is not selected and sentence 2 and sentence 3 are selected among the three instruction sentences (sentence 1, sentence 2, and sentence 3) included in the teacher data. . In the example shown in FIG. 7, since sentence 1 is not selected, the vector value (the above-mentioned instruction sentence vector) corresponding to all the words appearing in sentence 1 is weighted by 0.5 times. Thereby, for example, the vector value “5” of the word 1 included in the sentence 1 before the weighting is “2.5” after the weighting. Similarly, the vector values of other words included in sentence 1 are all multiplied by 0.5.

  On the other hand, since sentences 2 and 3 have been selected, all words appearing in sentences 2 and 3 are weighted twice. Thus, for example, the vector value “3” of the word 2 included in the sentence 2 before weighting is “6” after weighting. Similarly, the vector values of all the other words included in the sentences 2 and 3 are also doubled.

  As described above, the weighting unit 114 weights the vector value of the instruction sentence vector based on whether the instruction sentence is selected by the pharmacist. In the above example, weighting is performed twice when selected and 0.5 times when not selected. However, the present invention is not limited to this, and the amount of weighting may be set as appropriate.

  The teacher data after the weighting performed by the weighting unit 114 is input to the learning model generation unit 17. The learning model generation unit 17 performs re-learning using the weighted teacher data, updates the learning model, and registers a new learning model in the learning model DB 18. Further, the teacher data after the weighting performed by the weighting unit 114 may be transmitted to the medication instruction support server 20 (see FIG. 1) via the communication network N. Information regarding the weighting performed by the weighting unit 114 and unweighted teacher data may be transmitted to the medication administration support server 20.

  The medication instruction support server 20 stores the transmitted teacher data (and information about the weighting), and generates a new learning model by re-learning using the weighted teacher data, or replaces the existing learning model with the weighted teacher data. Be updated. The learning model newly created or updated in the medication administration support server 20 is transmitted to the above-described learning model DB 18 via the communication network N as needed.

  The medication support server 20 is preferably a server device that collects weighting information and teacher data from a plurality of pharmacies P and updates learning data. The medication compliance support server 20 may generate a learning model that can be used in all stores based on information and teacher data from all stores, or based on information and teacher data collected from some pharmacies P. , A learning model suitable for some pharmacies P may be generated. Some pharmacies P may be, for example, only one store or a plurality of stores existing in a certain area. Alternatively, the identification information of the pharmacist may be collected together with the information, and the learning model may be generated for each pharmacist. By generating a plurality of learning models in this way, the medication instruction support device 10 creates a medication instruction document more suitable for a patient using a learning model optimized for a store, a region, or a pharmacist. Will be able to do it.

  As described above, the medication instruction support device 10 according to the first embodiment of the present invention generates a learning model based on a patient vector and an instruction sentence vector, and uses the learning model to generate a drug information document. For each item, an instruction sentence corresponding to an instruction sentence vector having a large similarity to the patient vector is extracted. Then, based on the similarity between the corresponding instruction text vector and the patient vector, the extracted instruction texts are ranked for each item.

  Therefore, a guidance sentence that is highly relevant to the patient and includes important contents is extracted for each item. In addition, since the ranking is set for each item, it is easy to understand which of the extracted instruction sentences is more important. Furthermore, since guidance sentences are extracted for each item, guidance sentences containing important information for each item are not extracted and omitted compared to, for example, extracting important items for the patient from all guidance sentences. Is easy to avoid. Thereby, the medication instruction support device 10 can suitably support the pharmacist in preparing the medication instruction document.

  In the above description, the learning model generation unit 17 generates a learning model that calculates a similarity between vectors and outputs a similar instruction sentence vector, but the present invention is not limited to this. For example, a learning model may be generated in which clustering is performed on a patient vector and a guidance sentence vector in advance to generate a patient cluster and a guidance sentence cluster, and a similarity between the clusters is calculated to output a similar cluster. Good. In this case, for example, the instruction sentence extracting unit 19 may extract an instruction sentence corresponding to, for example, an instruction sentence vector located at the center of the output instruction sentence cluster. The present invention does not particularly limit which instruction sentence vector is extracted from the instruction sentence cluster obtained by using the learning model by the instruction sentence extracting unit 19.

(Second embodiment)
Next, a second embodiment of the present invention will be described. FIG. 8 is a block diagram illustrating an example of a functional configuration of the medication instruction support device 10A according to the second embodiment. In the second embodiment, the learning model generated by the learning model generator differs from the first embodiment described above. Therefore, in the following, an operation of the learning model generation unit 17A according to the second embodiment that is different from that of the learning model generation unit 17 according to the first embodiment will be described, and the same operation as that of the first embodiment will be described. The description of the configuration to be performed is omitted.

  In the second embodiment, the learning model generated by the learning model generation unit 17A is such that, in response to the input of a patient vector of a certain patient, a guidance sentence vector previously associated with a patient vector similar to the patient vector is used. This is a model that has been subjected to learning processing so as to be output. The instruction sentence vector previously associated with a patient vector means an instruction sentence vector corresponding to an instruction sentence provided in the past to a patient corresponding to a certain patient vector.

  As in the first embodiment, the learning model generation unit 17A uses the patient vectors corresponding to a plurality of patients and the instruction sentence vector of the instruction sentence previously provided to each patient as teacher data. What is necessary is just to generate such a learning model. The degree of similarity between patient vectors may be determined by a known method such as cosine similarity, as in the first embodiment. The learning model thus generated is registered in the learning model DB 18A.

  Using the learning model generated in this way, the subsequent processing in the medication administration support device 10A is executed. That is, the guidance sentence extracting unit 19 extracts a guidance sentence to be provided to a new patient for each item by using a learning model, the ranking unit 110 ranks the guidance sentences for each item, and the guidance sentence display unit 111 Display instruction sentences according to the ranking. Then, the selection receiving unit 112 receives the selection of the pharmacist, and the medication instruction document data generating unit 113 generates medication instruction document data including the selected instruction text. The weighting unit 114 weights the teacher data based on the pharmacist's selection, and transmits the weighting information or the teacher data to the medication instruction support server 20.

  As described above, in the medication support system 10A according to the second embodiment of the present invention, a learning process is performed so as to output a guidance sentence vector previously associated with a patient vector similar to the input patient vector. Is generated. For this reason, according to the medication guidance support device 10A, it is possible to create a medication guidance document including an optimal guidance sentence for a new patient by utilizing the accumulation of medication guidance documents provided to various patients in the past. Can provide assistance.

  In the above-described second embodiment, the learning process is performed such that, in response to the input of the patient vector, a guidance sentence vector previously associated with a patient vector similar to the patient vector is output. It has been described that a learning model is generated. However, for example, a learning model may be generated in which a learning process is performed so that a patient vector similar to the patient vector is output in response to the input of the patient vector. In this case, a database in which instruction sentences provided in the past for a plurality of patients are registered in advance is prepared, and the instruction sentence extracting unit 19 refers to this database to extract a patient vector of a new patient. The guidance sentence provided to the patient corresponding to the patient vector output when input to the learning model can be extracted as a guidance sentence to be provided to a new patient.

(Third embodiment)
Next, a third embodiment of the present invention will be described. FIG. 9 is a block diagram illustrating an example of a functional configuration of a medication instruction support device 10B according to the third embodiment.

  In the first embodiment described above, using the learning model generated by the learning model generating unit 17, the instruction sentence extracting unit 19 extracts an instruction sentence for each item, and the ranking unit 110 extracts the extracted instruction sentence. Was being ranked. In the third embodiment described below, the extraction and the ranking of the guidance sentences for each item are performed at one time (one step). To realize this, the medication support system 10B according to the third embodiment includes a vector generation unit 16B, a learning model generation unit 17B, a guidance sentence extraction unit 19B, and a weighting unit 114B. Further, the medication instruction support device 10B according to the third embodiment does not include the ranking unit 110 included in the medication instruction support device 10 according to the first embodiment.

  In the third embodiment, the learning model generated by the learning model generation unit 17B is used to input a patient vector of a certain patient and a guidance sentence vector of all the guidance sentences corresponding to the medicine prescribed for the patient. On the other hand, it is a model that has been subjected to a learning process so as to output the certainty factor for each instruction sentence vector. In the present specification, the certainty factor is a value indicating a degree (probability) that a certain instruction sentence includes information necessary for a certain patient. Details of the certainty factor will be described later. The learning model thus generated is registered in the learning model DB 18B.

  In order to generate such a learning model, the third embodiment includes a patient vector of a plurality of patients and a guidance sentence vector of all guidance sentences related to drugs prescribed for each patient. Teacher data to which information indicating whether or not each instruction sentence was used in the provided medication instruction document is used in advance.

  FIG. 10 is a diagram illustrating teacher data used in the third embodiment. As shown in FIG. 10, the teacher data includes information on medicine, patient data, instruction text, and importance. The importance level is information (label) indicating whether or not each instructional sentence was used in the medication instruction document provided to the patient. The importance level is “0” for the instruction text that was not provided.

  The teacher data as shown in FIG. 10 is vector-converted by the vector generation unit 16B. In the third embodiment, Word2vec, which is a vector conversion for each word, is used for the vector conversion of the guidance sentence by the vector generation unit 16B. Specifically, the vector generation unit 16B extracts words from the instruction sentence by morphological analysis, and performs word vector conversion using Word2vec. As a result, a word vector obtained by converting a word included in the instruction sentence is generated. The dimension of this word vector does not depend on the number of words, but is a predetermined fixed dimension. It is known that when the dimension is increased, the performance of expressing a guidance sentence is improved. Then, the vector generation unit 16B generates a guidance sentence vector based on all word vectors included in the guidance sentence. In order to generate an instruction sentence vector based on a word vector, for example, an average of the word vectors included in the instruction sentence may be obtained.

  Using such teacher data, the learning model generator 17B generates the following learning model. In the third embodiment, the learning model generation unit 17B generates a learning model as a set of a plurality of prediction models. FIG. 11 is a conceptual diagram of a learning model according to the third embodiment.

  In the example shown in FIG. 11, four prediction models are generated. In the example illustrated in FIG. 11, each prediction model includes a decision tree based on teacher data, for example, a word vector included in each item of patient data or an instruction sentence as an explanatory variable, and importance as an objective variable. included. For example, in the prediction model 1, the root node classifies “sex”, and the child nodes below it classify “concomitant medication” or “age”. In the prediction model 2, the root node classifies “component [n]” (the n-th component (element) vector in the instruction sentence vector), and the child nodes below it classify “hepatic dysfunction” or “liver dysfunction”. Gender ".

  As illustrated in FIG. 11, the learning model generated by the learning model generation unit 17B calculates an average value of the importance levels output by the plurality of prediction models, and outputs the average value as a certainty factor. As described above, since the importance is a value of 0 or 1, the certainty takes a value of 0 or more and 1 or less. In the example of FIG. 11, the output degrees of the four prediction models are 1, 0, 1, and 1, so the confidence is 0.75.

  The importance output by the plurality of prediction models is the importance predicted in consideration of only some of the patient data and the instruction sentence (items of the patient data or words of the instruction sentence). For this reason, the certainty factor, which is the average value of the importance levels output by the plurality of prediction models, is predicted in consideration of all the elements of the patient data and the instruction sentence, and a certain instruction sentence is necessary for a certain patient. A value indicating the degree (probability) of including information.

  In the example shown in FIG. 11, four prediction models are included in the learning model. However, it is desirable that more prediction models are actually generated. In order to ensure the accuracy of the certainty factor, it is desirable to generate a prediction model that includes at least all the items of the patient data as explanatory variables. In the example shown in FIG. 11, the number of nodes included in the decision tree of each prediction model is three, but each decision tree may include more nodes.

  The instruction sentence extracting unit 19B extracts an instruction sentence to be provided to the patient using the learning model generated in this manner. Specifically, when the instruction sentence extracting unit 19B acquires a new patient vector and instruction instruction vectors of all instruction sentences included in the medicine information document of the medicine prescribed for the patient, the new instruction vector and each of the new patient vectors are acquired. The instruction sentence vector is input to the learning model, and an output of the degree of certainty of the instruction sentence vector is obtained.

  Here, when the certainty degree of a certain guidance sentence vector is equal to or more than a predetermined threshold, the guidance sentence extraction unit 19B determines that the guidance sentence corresponding to the input guidance sentence vector is a patient corresponding to the input patient vector. Is determined to be important to the user, and the instruction sentence is selected. When the certainty factor of a certain instruction sentence vector is smaller than a predetermined threshold, the instruction sentence extraction unit 19B determines that the instruction sentence corresponding to the input instruction sentence vector is important for the patient corresponding to the input patient vector. Is not determined, and the instruction sentence is not selected. The guidance sentence extracting unit 19B performs such a determination for all the guidance sentence vectors, and extracts a guidance sentence corresponding to the selected guidance sentence vector. The instruction sentence extracted in this manner is an instruction sentence that is highly likely to be provided (important to the patient) to the patient.

  Note that, for example, a value such as 0.5 can be adopted as the predetermined threshold. The threshold may be dynamically determined by using a statistical analysis method. Specifically, there is a method of using the value of the valley of the frequency distribution as a threshold value using the discriminant analysis binarization method (Otsu's method) or the MaxEntropy method for the frequency distribution of the certainty factor output from the learning model. . As a result, an appropriate instruction sentence can be extracted according to the output characteristics of the learning model.

  The guidance sentence extracting unit 19B extracts the above-described guidance text for each item of the medicine information document. Then, the guidance sentence extracting unit 19B ranks the extracted guidance sentences based on the degree of certainty. Thereby, in the medication instruction support device 10B according to the third embodiment, the instruction sentence extraction unit 19 in the medication instruction assistance device 10 (see FIG. 1) according to the first embodiment extracts the instruction sentence and ranks the instruction sentence. The ranking by the attaching unit 110 can be performed at once.

  In the above description, the random forest classifier shown in FIG. 11 is used to obtain the output of the degree of certainty of the instruction sentence, but another method may be adopted. Specifically, a classifier configured by any one of SGD (Stochastic Gradient Descent), SVM (Support Vector Machine), NN (Neural Network) and CNN (Convolutional Neural Network), or a combination thereof may be used.

  On the other hand, the weighting unit 114B sets weighting for the teacher data used when the learning model generation unit 17B generates the learning model based on the selection result of the guidance sentence by the pharmacist received by the selection reception unit 112.

  The weighting unit 114B registers the instruction text selected by the pharmacist and the information on the drug related to the instruction text, the patient data, and the importance multiple times in the teacher data, thereby realizing the weight of the relevant instruction text. . That is, the weighting unit 114B acquires the instruction sentence newly selected by the pharmacist, information on the medicine corresponding to the instruction sentence, patient data, and importance. Then, when the importance level of the instructional sentence is “1” (when the instructional sentence is already selected), the weighting unit 114B adds the instructional sentence to the teacher data again. As a result, the information, patient data, and importance related to the same instruction text and the drug related to the instruction text are duplicated in the teacher data. Therefore, it is possible to perform weighting so that the learning model generated using the teacher data outputs a certainty of the instruction text that exists redundantly.

  The number of instruction sentences to be added may be the same as the number of times selected by the pharmacist, or may be proportional to the number of times selected. When the number of instruction sentences to be added is made proportional to the number of times selected, the number of times selected may be separately managed for each instruction sentence. In addition, when gradual weighting is performed, the number of times of selection and the instruction sentence may be managed in association with each other, and the number of times of selection may be proportional to the common logarithm of the number of times of selection. That is, for example, when the instruction is selected ten times, the instruction sentence may be added to the teacher data. If the importance of the instruction sentence selected by the pharmacist is “0”, the weighting unit 114B does not add the instruction sentence to the teacher data but overwrites the importance of the corresponding instruction sentence to “1”.

  The teacher data after the weighting performed in the weighting unit 114B is input to the learning model generation unit 17B. The learning model generation unit 17B performs re-learning using the weighted teacher data, updates the learning model, and registers a new learning model in the learning model DB 18B. The teacher data updated by the weighting unit 114B may be transmitted to the medication instruction support server 20. The operation of the medication instruction support server 20 that has received the updated teacher data is the same as in the first embodiment described above.

  As described above, in the medication support system 10B according to the third embodiment of the present invention, a patient vector of a certain patient, a guidance sentence vector of all the guidance sentences corresponding to the medicine prescribed for the patient, In response to the input, a model that has been subjected to a learning process so as to output a certainty factor for each instruction sentence vector is generated. Then, when the new patient vector and the instruction sentence vector of all instruction sentences included in the drug information document of the drug prescribed for the patient are input, the new patient vector and each instruction sentence vector are used as a learning model. Input and obtain the output of the certainty factor of the instruction sentence vector. This makes it possible to extract a guidance sentence that includes a large degree of information necessary for the patient, and to rank the extracted guidance sentences based on the certainty factor.

(Fourth embodiment)
Next, a fourth embodiment of the present invention will be described. In the fourth embodiment, it is possible to prevent a plurality of different instruction sentences having the same content from being extracted. It should be noted that the functional configuration of the medication instruction support device of the fourth embodiment is the same as that of the medication instruction support device 10B of the third embodiment described with reference to FIG.

  Hereinafter, a characteristic operation of the medication administration support device according to the fourth embodiment will be described with a specific example. For example, in the item of "side effect" of the drug information document, depending on the source of the instruction text (attached document, medicine bookmark, drug information provision book, etc.), for example, there is an instruction text with substantially the same content but different notation as follows. sell.

-The main side effects reported include rash, anorexia, nausea, vomiting, and diarrhea.
-The main side effects are anorexia, nausea, vomiting, diarrhea, etc.
・ Side effects: loss of appetite, nausea, vomiting, diarrhea

  In the fourth embodiment, the following configuration prevents the extraction of a plurality of instruction sentences having substantially the same contents but different notations.

  FIG. 12 is a diagram illustrating teacher data according to the fourth embodiment. The teacher data according to the fourth embodiment illustrated in FIG. 12 is provided with a “group number” in addition to the teacher data according to the third embodiment illustrated in FIG. The “group number” is a label assigned in advance to the instruction text having the same content (an example of the group label of the present invention).

  In the fourth embodiment, the learning model generation unit 17B generates a learning model in consideration of the group number. Specifically, the learning model generation unit 17B expands the random forest classifier described in the third embodiment to a multi-label type classifier using a multi-label classifier or the like, thereby generating a learning model. Generate. This learning model outputs a certainty factor and a group number in response to an input of a patient vector and an instruction sentence vector.

  In the fourth embodiment, the guidance sentence extracting unit 19B refers to the group number for the guidance sentence having a certainty factor equal to or greater than a predetermined threshold value, and when there are a plurality of guidance sentences with the same group number, the guidance content with the highest certainty factor. Extract only statements and discard others. As a result, it is possible to prevent a plurality of instruction sentences having substantially the same contents but different notations from being extracted.

  In addition, since the drug information document DB 13 includes an arbitrary document provided by a volunteer of a pharmacist, it is desirable to preferentially extract a document created by a veteran pharmacist or a document selected by a veteran pharmacist. There are cases. In this way, even when a specific document is to be preferentially extracted, it is possible to cope by expanding the fourth embodiment.

  Specifically, “priority” is assigned to the teacher data together with the “group number”. The “priority” is a value in which the higher the value, the higher the priority, and is a value that is set in a guidance sentence that should be preferentially extracted thereafter, such as selection by a veteran pharmacist. As an example of the priority, an arbitrary value such as the number of times selected by a specific pharmacist can be set. The priority of the instruction sentence that has not been selected before is set to “0”.

  The learning model generation unit 17B generates a learning model in which the group number and the priority are added, using the teacher data to which the group number and the priority are assigned. The learning model generated in this way outputs a certainty factor, a group number, and a priority for the input of the patient vector and the instruction sentence vector. The guidance sentence extracting unit 19B performs a priority process on the certainty with respect to the output result of the learning model for the guidance sentence for which the priority is set (non-zero) by using the value of the priority. It is possible to increase the possibility of extracting a guidance sentence with a priority set.

  As a specific example of the priority processing, there is a method of adding a value determined by the magnitude of the priority to the certainty factor. As a specific example, priority × fixed value (for example, 0.1) may be added to the certainty factor. Since the instruction sentence extracting unit 19B performs such priority processing after the output of the learning model, re-learning is not required even when the priority is changed (when a specific pharmacist newly selects an instruction sentence). It is. Therefore, the above extension can be easily realized.

  In order to assign a group number to an instruction text having the same content in the teacher data, it is necessary to extract an instruction text having the same content from the instruction text used as the teacher data. As a method of extracting instruction sentences having the same contents, for example, the following methods are available. That is, the instruction sentence vectors of all the instruction sentences are calculated in advance, and the similarity (cosine similarity or the like) between two specific instruction sentence vectors is calculated. Then, the similarities of all the combinations are calculated, and the combinations whose similarities are equal to or more than a predetermined threshold value are determined to be the same instruction text.

  At this time, dimension reduction and normalization of the instruction sentence vector may be performed in order to reduce the calculation amount of the similarity calculation.

  Weighting may be performed based on whether or not there is an instruction sentence vector having the same content. As described with reference to FIG. 7, when weighting is performed for each word included in the instruction sentence, the entire word included in the instruction sentence is determined based on the distance (scalar value) between two specific instruction instruction vectors. What is necessary is just to add the vector value according to distance. Further, as described with reference to FIG. 10, when weighting is performed for each instruction sentence, a value corresponding to the distance may be set as the importance instead of adding “0” or “1” as the importance. I just need.

  In addition, as an example of the value y according to the distance between the instruction sentence vectors, for example, when the distance is normalized and its variable is x (0 to 1), a value given by y = −x + 1 may be used. That is, y is a value closer to 1 as the degree of similarity between instruction sentence vectors is larger, and closer to 0 as the degree of similarity is smaller.

  With such weighting, a difference can be made between the weights of a guidance sentence having the same content and a guidance sentence that is not the same. For example, it is difficult to extract similar guidance sentences at the same time, or conversely, similar guidance sentences are simultaneously extracted. Or make it easier. In addition, when displaying the extracted instruction text on the display screen, the same instruction text is emphasized and displayed, or when one instruction text is selected, another instruction text with the same content is selected at the same time. Control, such as slipping.

(Fifth embodiment)
Further, a fifth embodiment of the present invention will be described. FIG. 13 is a block diagram showing a functional configuration of a medication instruction support device 10C according to the fifth embodiment. As shown in FIG. 13, a medication instruction support device 10C according to the fifth embodiment has a second learning model in addition to the functional configuration of the medication instruction support device 10B according to the third embodiment shown in FIG. 9. It has a generation unit 115, a second learning model DB 116, and a second instruction sentence extraction unit 117. Further, the medicine information document data acquisition unit 12C, the patient data acquisition unit 14C, and the vector generation unit 16C of the medication guidance support device 10C according to the fifth embodiment are configured to include the medicine information described in the third embodiment. The operations are different from those of the document data acquisition unit 12, the patient data acquisition unit 14, and the vector generation unit 16B. Hereinafter, these details will be described.

  In the fifth embodiment, the following case is assumed. For example, the fact that a patient who has prescribed a certain drug is using a different drug from the drug may be discovered by an interview with a pharmacist, a medicine notebook, or the like. Hereinafter, the prescribed drug is described as a main drug, and another drug is described as a concomitant drug. In such a case, it is desirable that the medication instruction document provided to the patient includes an instruction sentence including precautions when the main drug and the concomitant drug are used in combination.

  In order to extract a guidance sentence including precautions when the main drug and the concomitant drug are used together, for example, a method of performing a text search on the entire guidance text of the main drug by the drug name of the concomitant drug is considered. However, since there is frequent spelling in the drug name, there is a possibility that omission may occur by simply performing a text search. Specific examples of the swaying include, for example, "mumps vaccine" and "mumps vaccine" or "isoniazid" and "antituberculosis drug".

  In the fifth embodiment, when the medicine information document data acquisition unit 12C acquires all the instruction texts of the main medicine for the patient using the concomitant medicine in addition to the main medicine, The names of concomitant drugs that sometimes require attention and guidance texts containing the precautions are extracted. Of the counseling sentences of the main drug, the names of concomitant medications that require attention when concomitant are, for example, if the drug information document is an attached document, sub-items “Contraindications for concomitant use” and “Caution for concomitant use” In the “drug name” and the like. In addition, among all the instruction texts of the main drug, the precautions for concomitant drugs that require caution when concomitant are, for example, if the drug information document is an attached document, the sub-items It is described in "Clinical symptoms and measures" included in "Precautions for combined use".

  In the following description, the name of the concomitant drug that needs attention when used in combination with the main drug acquired by the drug information document data acquisition unit 12C is “Caution Concomitant Drug Name”, and the instruction sentence including the cautionary note is “Caution Instruction Text”. ". In the subsequent processing, the “name of cautionary concomitant drug” and the “cautionary instruction sentence” regarding the drug are used in association with each other.

  On the other hand, the patient data acquisition unit 14C acquires the name of the concomitant drug used by the patient, for example, from an interview or a medicine notebook in addition to the patient data. In the following description, the name of the concomitant drug used by the patient will be referred to as “concomitant drug name”.

  The patient data and the name of the concomitant drug used are input to the vector generation unit 16C from the patient data acquisition unit 14C. In addition, all instruction sentences, cautionary medicine names, and caution instruction sentences are input to the vector generation unit 16C from the medicine information document data acquisition unit 12C. Thereby, the vector generation unit 16C generates a patient vector, a combined use drug name vector, a guidance sentence vector of all the guidance sentences, and a caution combined use drug name vector. In the fifth embodiment, a method of generating each word vector (word2vec or the like) is used for these vector conversions.

  At this time, the vector generation unit 16C performs the vector conversion using the model trained in the medical term dictionary, so that not only the name of the drug but also the name of the concomitant drug containing the meaning of each word constituting the drug name is used. A vector and a cautionary medicine name vector are generated.

  Among the vectors generated by the vector generation unit 16C, the patient vector and the instruction sentence vector of all instruction sentences are input to the learning model generation unit 17B and the instruction sentence extraction unit 19B, as in the third embodiment. .

  On the other hand, the used combination drug name vector and the caution combined drug name vector are input to the second learning model generation unit 115. Then, the second learning model generation unit 115 generates a second learning model that has been learned to output a cautionary instruction sentence vector having a high degree of similarity to the used combination drug name vector. Here, since each medicine name vector includes the meaning of each word constituting the medicine name, omission of extraction of the same medicine due to fluctuation of notation can be minimized. The method of calculating the similarity is the same as in the first embodiment, and a description thereof will be omitted. The second learning model is registered in the second learning model DB 116.

  The second instruction sentence extracting unit 117 inputs the used concomitant drug name vector to the second learning model, obtains an output of a cautionary concomitant drug name vector having a high degree of similarity, and The attention instruction sentence associated with is extracted.

  The guidance text display unit 111C displays the guidance text extracted by the guidance text extraction unit 19B and the guidance text extracted by the second guidance text extraction unit 117. Subsequent processing is the same as that of the medication instruction support device of each embodiment described above.

  As described above, in the medication-taking support device 10C according to the fifth embodiment, the combination drug name vector based on the combination drug name used by the patient and the instruction text of the main drug are extracted in advance. Based on the attention-concomitant drug name vector based on the attention-concomitant drug name requiring attention, the second learning is performed to output the attention-concomitant drug name vector having a high degree of similarity to the input concomitant drug name vector. A learning model is generated, and by using the second learning model, a guidance sentence including precautionary notes on the concomitant drug used by the patient is extracted from the guidance text of the main drug. At this time, in the vector conversion of the drug name, since the vector conversion in consideration of the meaning of the word is performed, it is possible to avoid a situation in which a guidance sentence relating to the concomitant drug cannot be extracted from the guidance sentence of the main drug due to a sway of the notation. be able to.

  In the fifth embodiment, the certainty factor described in the third embodiment is not calculated for the instruction sentence extracted by the second instruction sentence extraction unit 117. For this reason, in the fifth embodiment, it is not possible to rank the extracted instruction sentences based on the certainty factor. However, by calculating the certainty factor using the learning model registered in the learning model DB 18B based on the instruction sentence extracted by the second instruction sentence extraction unit 117, the instruction extracted by the second instruction sentence extraction unit 117 is obtained. The ranking of sentences may be performed. Further, it is checked whether or not the instruction sentence extracted by the instruction sentence extraction unit 19B includes the same content as the instruction sentence extracted by the second instruction sentence extraction unit 117. Only the guidance text extracted by the second guidance text extraction unit 117 may be added to the guidance text extracted by the guidance text extraction unit 19B.

(Sixth embodiment)
Further, a sixth embodiment of the present invention will be described. FIG. 14 is a block diagram showing a functional configuration of a medication instruction support device 10D according to the sixth embodiment.

  In each of the above-described embodiments, a mode has been described in which a guidance sentence important to a patient is extracted for each drug based on information on the drug included in the prescription data. In the sixth embodiment, a mode will be described in which a guidance sentence is extracted not for each medicine but for each prescription.

  As illustrated in FIG. 14, a medication instruction support device 10D according to the sixth embodiment has a teacher data DB 118, in addition to the functional configuration of the medication instruction support device 10B according to the third embodiment (see FIG. 9). It has a third learning model generation unit 119, a third instruction sentence extraction unit 120, and a third learning model DB 121. Also, the medication instruction support device 10D of the sixth embodiment has a selection accepting unit 112D that operates differently from the selection accepting unit 112 of the medication instruction support device 10B according to the third embodiment.

  The teacher data DB 118 stores in advance teacher data based on the guidance sentence extracted by the guidance sentence extraction unit 19B with respect to past prescription data. The teacher data in the sixth embodiment is hereinafter referred to as prescription teacher data. The prescription teacher data is generated, for example, as follows.

  FIG. 15 is a diagram illustrating an example of collected data for generating prescription teacher data according to the sixth embodiment. The collected data includes past prescription data, data output from the instruction sentence extraction unit 19B (hereinafter, referred to as extraction unit output data), and a ranking assigned by a pharmacist (hereinafter, a pharmacist ranking). Pharmacist ranking data).

  As shown in FIG. 15, the prescription data includes data corresponding to a certain past prescription, for example, a patient name and a plurality of drug names prescribed for the patient. The extraction unit output data is data that the instruction sentence extraction unit 19B previously output for the prescription data. The extraction unit output data includes a predetermined number (three in FIG. 15) of guidance sentences extracted by the guidance sentence extraction unit 19B, and the certainty factor for each guidance sentence. The extraction unit output data may include the guidance sentence itself, or may include only identification information for identifying the guidance sentence. In addition, the pharmacist ranking data includes, among a predetermined number of instruction sentences extracted by the instruction sentence extracting unit 19B for the prescription, an instruction sentence actually selected by the pharmacist, and a pharmacist attaching each of the selected instruction sentences. It is the data which shows the order which was performed. The pharmacist ranking data is generated, for example, by the selection receiving unit 112D.

  FIG. 15 shows an example of collected data based on a prescription of patient 1 in which three drugs (drugs 1 to 3) are described. In the example shown in FIG. 15, for example, three types of instruction sentences 1-1 to 1-3 are extracted for the medicine 1, and the confidence levels of the instruction sentence vectors are 1, 0.9, and 0.8, respectively. In addition, three types of instruction sentences 2-1 to 2-3 are extracted for the medicine 2, and the confidence levels of the instruction sentence vectors are 0.9, 0.8, and 0.7, respectively. Also, three types of instruction sentences 3-1 to 3-3 are extracted for the medicine 3, and the confidence levels of the instruction sentence vectors are 1, 0.9, and 0.8, respectively. In this prescription, the pharmacist ranks the instruction sentence 1-1 first, the instruction sentence 2-2 second, and the instruction sentence 3-1 third. Is not selected. In the collected data, the instructional sentence not selected by the pharmacist is not ranked, and the pharmacist ranking data is indicated by "-".

  Prescription teacher data is generated based on such collected data. 16A to 16C are diagrams illustrating examples of prescription teacher data generated based on the collected data illustrated in FIG.

  As shown in FIG. 16A to FIG. 16C, the prescription teacher data includes a patient name, a guidance sentence, a certainty factor, a pharmacist rank, and a superior / decractive label corresponding to two guidance sentences. The superiority label indicates the superiority between the two instruction sentences.

  In the sixth embodiment, prescription teacher data corresponding to all combinations of two instruction sentences in all instruction sentences extracted from past prescription data is independently generated.

  The priority label is determined based on the pharmacist rank. As shown in FIG. 16A, when the pharmacist rank is given to both of the specific two instruction sentences, the superiority label “1” is assigned to the instruction sentence 1-1 having the higher pharmacist rank, and the instruction having the lower pharmacist rank is given. The sentence 2-2 is assigned the superiority / deterrence label “0”.

  As shown in FIGS. 16B and 16C, when one of the instruction sentences is given a pharmacist ranking and the other instruction sentence is not given a pharmacist order, the instruction sentence of the person without the pharmacist order is given. Is always “0”. Accordingly, as shown in FIG. 16B, when the instruction sentence 2-2 is given a ranking but the instruction sentence 3-2 is not given a ranking, the instruction sentence 2-2 is given a superiority / delete label "1". However, the instructor sentence 3-2 is assigned a superiority label “0”, respectively.

  In addition, as shown in FIG. 16C, when the instruction sentence 1-2 and the instruction sentence 1-3 have not been ranked (that is, have not been selected by the pharmacist in the past), any instruction sentence is not given. The priority label “0” is given.

  Alternatively, the certainty factor may be reflected on the superiority / inferior label. That is, as shown in FIG. 16A, FIG. 16B and FIG. 16C, for example, the value of the determined superiority label is multiplied by the certainty factor of the corresponding instruction sentence, so that the superiority / consciousness label considering the certainty factor is assigned to each instruction sentence. You may make it give. Specifically, in the example illustrated in FIG. 16A, the confidence level of the instruction sentence 1-1 is “1” and the priority label is “1”, so the priority label in consideration of the confidence level is “1”. . In the example shown in FIG. 16B, the confidence level of the instruction sentence 2-2 is “0.8” and the priority label is “1”, so the priority label in consideration of the confidence level is “0.8”. . Further, in the example shown in FIG. 16C, both the superiority and inferiority labels are “0”, and the superiority and inferiority label in consideration of the confidence is “0” regardless of the value of the confidence. By doing so, it is possible to control to relatively increase the inferiority label of the guidance sentence with high certainty.

  As described above, the prescription teacher data indicates the superiority (high or low of the pharmacist rank) between two specific instruction sentences among all the instruction sentences extracted from the past prescription data.

As described above, in the sixth embodiment, prescription teacher data corresponding to all combinations of two instruction sentences in all instruction sentences extracted from past prescription data is generated in advance. Need to be For example, when nine instruction sentences are extracted as in the collected data of FIG. 15, 36 prescription teacher data corresponding to ₉ C ₂ = 36 combinations are generated. However, in order to ensure the accuracy of the superiority / inferiority label output by the prediction model, it is necessary to distinguish the order in the data of the two instruction sentences, and prescription teacher data of ₉ P ₂ = 72 combinations is generated. You.

  The prescription teacher data generated in this way is stored in the teacher data DB 118.

  The third learning model generation unit 119 generates a learning model using the prescription teacher data stored in the teacher data DB 118. The learning model generated by the third learning model generation unit 119 is important for a patient when new prescription data is input, from a plurality of instruction sentences corresponding to a plurality of medicines described in the new prescription. This is a model that has been subjected to learning processing so as to output a guidance sentence.

  The generation of the learning model by the third learning model generation unit 119 is preferably performed, for example, before the operation of the medication instruction support device 10D in the pharmacy P, similarly to the generation of the learning model by the learning model generation unit 17B. However, if the extractor output data and the pharmacist ranking data for the new prescription data are newly generated in a state where the medication instruction support device 10D is operating at the pharmacy P, the data is used as new prescription teacher data at any time. A mode in which the learning model of the three learning model DB 121 is continuously updated (additional learning or re-learning at an arbitrary timing) may be adopted.

  The third instruction sentence extracting unit 120 extracts an instruction sentence to be provided to the patient using the learning model generated in this manner. Specifically, the third instruction sentence extracting unit 120 acquires an instruction sentence extracted for the prescription data based on the newly input prescription data. Using the learning model, the third instruction sentence extracting unit 120 determines the superiority or inferiority of the instructional sentence in the combination of all the two instructional sentences in the acquired instructional sentence, and extracts a predetermined number of upper instructional sentences. .

  With such a configuration, the medication instruction support device 10D according to the sixth embodiment can extract an important instruction sentence for the patient not for each drug but for each prescription. This makes it possible to provide the patient with a guidance sentence that takes into account the interrelationship between the drugs prescribed in the prescription. In addition, since the guidance sentence not selected by the pharmacist is not excluded from the prescription teacher data and is treated as an element having the priority label “0”, it is possible to select a guidance sentence more suited to the actual situation.

  With such a configuration, for example, the following effects can be obtained. Some drugs must not be used in combination with other drugs (contraindications for concomitant use), and some drugs cause an interaction when used in combination with other drugs. If the medicine described in the prescription has such contraindications or interactions, or if guidance text is extracted for each drug, guidance text on contraindications or interactions may not be extracted. However, it is highly probable that the pharmacist has selected (ranked) instructional sentences in consideration of contraindications for concomitant use and interaction with respect to past prescriptions in which drugs of the same combination are described. From this, the medication instruction support device 10D according to the sixth embodiment generates a learning model based on the result of ranking by the pharmacist for each past prescription, so that the instruction not extracted by the extraction for each drug is provided. The sentence is expected to be extracted.

(Modification of Sixth Embodiment)
In the sixth embodiment described above, based on the newly input prescription data, in all the instruction sentences extracted from the prescription data, the prescription teacher data corresponding to all combinations of the two instruction sentences are added to the prescription teacher data. Based on this, a learning model was generated. Therefore, at the time of extracting a guidance sentence by the third guidance sentence extraction unit 120, a superior or inferiority of a combination of all two guidance sentences is determined, and then a predetermined number of top guidance sentences are extracted.

  In the present modified example, an example will be described in which a learning model that can determine in advance the inferiority between a predetermined number of instruction sentences in order to reduce the amount of computation at the time of predicting the superiority and inferiority label using the generated learning model is described.

  In this modified example, teacher data indicating superiority between a predetermined number of instruction sentences is stored in the teacher data DB 118 as prescription teacher data. 17A and 17B are diagrams illustrating another example of the prescription teacher data. The prescription teacher data shown in FIGS. 17A and 17B is generated based on the collected data shown in FIG.

  In the example shown in FIG. 17A, when the predetermined number is set to 3, that is, prescription teacher data in which superiority and inferior labels based on pharmacist ranking data are assigned to each of the three instruction sentences is shown. In this modification, since it is necessary to indicate the superiority between a predetermined number of instruction sentences, an arbitrary number between 0 and 1 is used for the superiority label, for example. In the example illustrated in FIG. 17A, the superiority label “1” is assigned to the instruction sentence 1-1 having the first pharmacist rank. In addition, the superiority label “0.5” is assigned to the instruction sentence 2-2 in which the pharmacist ranks second. In addition, the superiority label “0” is assigned to the instruction sentence 3-1 in the third place of the pharmacist (the lowest rank because there are three in total). In this case, the prescription teacher data is 3! = 6 types are generated.

  On the other hand, in the example shown in FIG. 17B, when the predetermined number is set to 100, that is, for each of the 100 instruction sentences, prescription teacher data in which the superiority / inferior label based on the pharmacist ranking data is added is shown. In the example shown in FIG. 17B, the instructional sentence 3-1 having the third pharmacist rank is not the lowest, and all instructional sentences other than the instructional statements 1-1, 2-2, and 3-1 not selected by the pharmacist. Is lower. For this reason, in the example shown in FIG. 17A, the superiority label “1” is assigned to the instruction sentence 1-1 having the first pharmacist rank. Further, the superiority label “0.9” is assigned to the instruction sentence 2-2 having the second highest pharmacist rank. In addition, the instructor sentence 3-1 having the third pharmacist rank is assigned the superiority / deterrence label “0.8”. Furthermore, the superiority label “0” is assigned to all other instruction sentences. The value of the priority label is a relative value, and may be set to an arbitrary value reflecting the pharmacist rank. However, it is necessary to adopt the same setting method between a plurality of prescription teacher data for different prescriptions.

  Since the collected data illustrated in FIG. 15 includes only nine instruction sentences in total, in the example illustrated in FIG. 17B, 91 prescription teacher data including 100 instruction sentences are required to generate 91 instruction sentences. Instruction text is not enough. To make up for this, empty data is given to the missing instruction sentences. The empty data is data in which the instruction sentence and its identification information, the certainty factor, and the superiority / inferiority label are all 0, as exemplified in the lower part of FIG. In addition, since the prescription teacher data also uses data rearranged in consideration of the order of the guidance sentences, in the example shown in FIG. 17B, 9! = 362,880 types of data are generated.

  The third learning model generation unit 119 generates a learning model that can output the superiority between a plurality of instruction sentences based on the prescription teacher data generated in this way. Thus, in this modification, the amount of calculation can be reduced as compared with the above-described sixth embodiment.

(Seventh embodiment)
In the first embodiment described above, the training data is weighted based on the selection or non-selection of the guidance sentence by the pharmacist, and the learning model generation unit 17 performs re-learning using the weighted teacher data. I was In such a configuration, by assigning a weight of less than 1 to an instruction sentence not selected by the pharmacist, the ranking of the instruction sentence not selected by the pharmacist may be significantly reduced. When the ranking is lowered, the priority order at which the instruction is displayed is lowered and it is difficult for pharmacists to see it, so even if the instruction is actually important for a patient, It will be difficult to select.

  In the seventh embodiment, the following configuration is adopted to cope with such a situation. The functional configuration of the medication instruction support device of the seventh embodiment is the same as that of the medication instruction support device 10 of the first embodiment described with reference to FIG.

  The guidance sentence display unit 111 includes a guidance sentence display column 201 that displays a predetermined number of top guidance sentences for each item in accordance with the ranking performed by the ranking unit 110, and a check column 202 that receives a selection by a pharmacist for each guidance sentence. Is displayed. In addition, the guidance sentence display unit 111 displays a display element (such as a button) for displaying a guidance sentence on the display screen 200. FIG. 18 is a diagram showing a display example of a check box 202 and a button 210 for displaying a guidance sentence. When the pharmacist selects the button 210 via the input device 106, a guidance sentence slightly lower than the guidance sentence currently displayed in the guidance sentence display column 201 is displayed in the guidance sentence display column 201.

  The pharmacist can check the check box 202 via the input device 106 for a guidance sentence that does not make a selection for the current patient, but does not want to lower the rank because it is important for some patients. . When the pharmacist makes a check, processing is performed on the checked instruction sentence so as not to lower the ranking. Specifically, for example, the weighting unit 114 may weight the instruction sentence checked by the pharmacist by a factor of 1 or more.

  The pharmacist presses the button 210 when he / she wants to check a guidance sentence lower than the guidance sentence currently displayed in the guidance sentence display column 201. As a result, an instruction sentence slightly lower than the current instruction is displayed in the instruction sentence display column 201, and the pharmacist can check.

  Depending on the pharmacist, the check may not be performed, or the pharmacist may not know that the check can be performed. In order to cope with this, a sentence, a symbol, or the like may be displayed on the display screen 200 to notify that another instruction sentence can be viewed and that an important instruction sentence can be operated without lowering the ranking.

  Also, instead of checking for guidance texts that the pharmacist does not want to lower the ranking, the pharmacist may be able to directly edit the ranking of a plurality of guidance texts.

  With the above-described configuration, it is possible to avoid a situation in which the order of the instruction text not selected by the pharmacist is greatly reduced and the pharmacist does not easily see it.

  INDUSTRIAL APPLICABILITY The present disclosure is useful for a medication instruction support device that supports creating a medication instruction document in which an instruction sentence including information necessary for medication instruction is described.

1 Medication Guidance Support System 10, 10A, 10B, 10C, 10D Medication Guidance Support Device 11 Prescription Data Acquisition Unit 12, 12C Drug Information Document Data Acquisition Unit 13 Drug Information Document DB
14, 14C Patient data acquisition unit 15 Patient DB
16, 16B, 16C Vector generation unit 17, 17A, 17B Learning model generation unit 18, 18A, 18B Learning model DB
19, 19B Instruction sentence extraction unit 20 Drug administration support server 104 External storage device 105 Communication interface 106 Input device 107 Output device 110 Ranking unit 111, 111C Instruction sentence display unit 112, 112D Selection accepting unit 113 Drug administration instruction document data generation unit 114 , 114B weighting unit 115 second learning model generation unit 116 second learning model DB
117 Second instruction sentence extraction unit 118 Teacher data DB
119 third learning model generation unit 120 third instruction sentence extraction unit 121 third learning model DB
200 display screen 201 guidance sentence display field 202 check field 210 button N communication network

Claims (17)

A vector generation unit that generates a patient vector and a plurality of character string vectors based on a plurality of character strings for taking medication guidance corresponding to patient data and medicines regarding the patient,
A learning model generation unit that generates a learning model that has been subjected to learning processing so as to output a predetermined character string vector with respect to the input of the patient vector and the plurality of character string vectors,
A character string extraction unit that extracts a predetermined character string from the plurality of character strings using the learning model;
A medication instruction support device having The learning model generation unit calculates a similarity between the patient vector and the character string vector, and performs a learning process such that a character string vector is output in descending order of similarity with respect to an input of a predetermined patient vector. Generate a learned model,
The medication administration support device according to claim 1. When the character string has been classified into a plurality of items in advance, the character string extraction unit, based on the input of the new patient vector, a predetermined number of the predetermined number of the Extract a string,
The medication instruction support device according to claim 2. A ranking unit that ranks the extracted predetermined number of the character strings based on the learning model,
The medication compliance support device according to claim 3. A weighting unit that determines a weight value for the character string vector corresponding to the predetermined character string, based on whether the extracted predetermined character string is used in a medication instruction document created by a pharmacist,
Further having
The learning model generation unit updates the learning model based on the weight value.
The medication instruction support device according to any one of claims 1 to 4. Further comprising a sentence conversion unit for converting the sentence format of the extracted predetermined character string,
The medication administration support device according to claim 1. The learning model generation unit calculates a similarity between the plurality of patient vectors, and performs learning so that a character string vector previously associated with a patient vector having a high similarity to the input of the patient vector is output. Generate a processed learning model,
The medication administration support device according to claim 1. The learning model generation unit, for the input of the patient vector and the character string vector, for each of the character string vector, the character string corresponding to the character string vector, information necessary for the patient corresponding to the patient vector Generate a learning model that has been subjected to learning processing to output a certainty factor indicating the degree of inclusion,
The medication administration support device according to claim 1. A weighting unit that determines a weight value for the character string vector corresponding to the predetermined character string, based on the number of times the extracted predetermined character string is used in a medication instruction document created by a pharmacist;
Further having
The learning model generation unit updates the learning model based on the weight value.
A medication compliance support device according to claim 8. The character string extracting unit extracts a character string corresponding to a character string vector whose certainty factor output by the learning model with respect to the input of the patient vector and the character string vector is larger than a predetermined value.
A medication compliance support device according to claim 8. The learning model generation unit performs a learning process to output the group label in addition to the certainty factor, using teacher data to which a group label indicating a group indicating the content of the character string vector is assigned. Generated learning model,
The character string extraction unit extracts a character string corresponding to a predetermined number of character string vectors in order of the degree of certainty from among the character string vectors to which group labels indicating the same contents are assigned,
The medication instruction support device according to claim 10. The learning model generation unit uses the teacher data further assigned a priority label indicating a priority indicating the degree to which the character string is preferentially used, in addition to the certainty factor, the group label and the priority label. Generating the learning model subjected to the learning process to output
The character string extraction unit may include a character string corresponding to a predetermined number of character string vectors in the descending order of the certainty factor weighted based on the priority among the character string vectors to which a group label indicating the same content is assigned. Extract the
A medication compliance support device according to claim 11. A second learning model generator,
A second extraction unit;
Further having
When the patient uses a combination drug in addition to the main drug, the vector generation unit,
Generate a concomitant drug name vector based on the concomitant drug name that is the name of the concomitant drug used by the patient, and
Included in the character string of the main drug, to generate an attention combination medicine name vector based on the attention combination medicine name that is the name of the combination medicine requiring attention when used in combination with the main medicine,
The second learning model generation unit performs a learning process on the input of the concomitant drug name vector so as to output the cautionary concomitant drug name vector having a high degree of similarity to the input concomitant drug name vector. Generated second learning model,
The second extraction unit uses the second learning model to generate a pre-association with the cautionary drug name corresponding to the cautionary drug name vector output in response to the input of the concomitant drug name vector. Is extracted from the character string of the main drug as a cautionary combination medicine instruction sentence,
The medication administration support device according to claim 1. A third learning model generator,
A third extracting unit;
Further having
The third learning model generation unit includes a plurality of character strings extracted from the character strings corresponding to the medicines described by the character string extraction unit from among the character strings corresponding to the medicines described in a plurality of past prescriptions. Based on the prescription teacher data including the superiority / inferiority label indicating the superiority or inferiority between the plurality of character strings, the character strings corresponding to the medicines described in the new prescription are ranked, and a predetermined number of character strings are extracted. Generate a third learning model that has been subjected to learning processing to
The third extraction unit extracts, using the third learning model, a predetermined number of upper-order character strings from character strings corresponding to drugs described in a new prescription.
The medication instruction support device according to claim 10. The superiority / inferiority label is given based on a pharmacist rank given to at least a part of the plurality of character strings by a pharmacist,
The medication instruction support device according to claim 14. It further has a reception unit that receives the operation of the pharmacist,
The weighting unit sets a weight value so that the order of the character string selected by the operation of the pharmacist is not lowered among a predetermined number of the character strings.
A medication instruction support device according to claim 5. A vector generation unit that generates a patient vector and a plurality of character string vectors based on a plurality of character strings for taking medication guidance corresponding to patient data and medicines regarding a patient, and for inputting the patient vector and the plurality of character string vectors. A learning model generation unit that generates a learning model subjected to learning processing to output a predetermined character string vector, and a character that extracts a predetermined character string from the plurality of character strings using the learning model. A column extracting unit, and determining a weight value for the character string vector corresponding to the predetermined character string, based on whether the extracted predetermined character string is used in a medication instruction document created by a pharmacist. A plurality of medication guidance support devices having:
A server device,
A predetermined network that connects the plurality of medication support devices and the server device;
Has,
The server device generates or updates the learning model used in each of the medication instruction support devices based on the weight values transmitted from the plurality of medication instruction support devices.
Medication guidance support system.

Images