JP2011159078A - Information processing apparatus, determination program and determination method - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To automatically determine whether a reported side effect of a medicine subject to determination is known. <P>SOLUTION: A similar medicine recognition part 105 recognizes a similar medicine to a medicine subject to determination, and a side effect keyword extraction part 107 extracts a side effect phrase subject to determination from a safety information report document 205. A side effect determination/learning part 108 instructs a keyword similarity evaluation part 109 to evaluate the combination of the side effect phrase subject to determination with a phrase indicating a side effect in an attached document 202 to the similar medicine, and compares the evaluation with a threshold to determine whether a side effect indicated by the side effect phrase subject to determination is known to the similar medicine. The keyword similarity evaluation part 109 tabulates evaluations of similarity between character substrings divided from the two phrases to evaluate a similarity between the two phrases. A character substring similarity evaluation part 110 gives an evaluation higher than the sum of an evaluation for matching character substrings of a first length and an evaluation for matching character substrings of a second length when character substrings of a third length equal to the first and second lengths match. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a technique for determining whether a side effect of a pharmaceutical product is known or unknown.

  According to the Japanese Pharmaceutical Affairs Law, pharmaceutical companies report to the Minister of Health, Labor and Welfare if any new side effects are discovered after marketing for drugs manufactured and sold by the company (hereinafter also referred to as “in-house drugs”). There is an obligation. Specifically, when a pharmaceutical company receives a report of a side effect (or a symptom suspected of having a side effect) from a doctor in a hospital, for example, the pharmaceutical company determines whether the reported side effect is known or unknown. Report to the Minister of Health, Labor and Welfare.

  However, when a pharmaceutical company receives a report of a side effect from a doctor or the like, it is difficult to determine whether the reported side effect is known or unknown. Therefore, a pharmaceutical company manually checks whether a side effect reported by a doctor or the like is known or unknown in a similar drug by referring to a package insert of the similar drug (that is, the same-type synergistic drug).

  The package insert is a document created by a pharmaceutical company based on the Pharmaceutical Affairs Law and attached to each drug. The package insert includes, for example, items such as “medicinal effect classification name”, “market name code”, “efficacy or effect”, “side effect”, and “physical and chemical knowledge regarding active ingredients”.

  Based on the results of examining the package insert of a similar drug as described above, a pharmaceutical company determines that “a side effect reported by a doctor or the like is a known side effect or an unknown side effect”. .

  If a pharmaceutical company reports to the Minister of Health, Labor and Welfare, missing reports of unknown side effects or delays in reporting could have a significant impact on the health of the people. Therefore, if the pharmaceutical company can quickly and accurately determine whether or not the side effect reported from the clinical site to the pharmaceutical company is a known side effect, it is beneficial for the whole society.

  However, since the format and notation of the description of the attached document are not necessarily strictly unified, it is difficult to obtain an accurate determination result only by performing a simple character string search for the attached document of the similar medicine. Therefore, at present, the determination that “the side effect reported from the clinical site to the pharmaceutical company is a known side effect or an unknown side effect” largely depends on human power.

  Therefore, an object of the present invention is to automatically determine whether a side effect reported for a certain pharmaceutical is known or unknown.

  An information processing apparatus according to one aspect includes a specifying unit, a report document acquiring unit, an analog recognition unit, a partial character string similarity evaluating unit, a phrase similarity evaluating unit, a side effect phrase extracting unit, a comparison target set acquiring unit, a determining unit, and an output Means.

  The specifying unit receives information for specifying a drug, and specifies the drug indicated by the information as a determination target drug. The report document acquisition means acquires a report document describing the side effects of the determination target drug. The similar drug recognition means reads out the similar drug learning result information associating the identification information uniquely identifying the drug with the similar drug of the drug from the storage means, or for each of a plurality of drugs, the drug of the drug By reading the package insert including the identification information, the side effect of the medicinal product and the efficacy or effect of the medicinal product from the storage unit, the similar drug similar to the determination target drug is recognized among a plurality of other medicinal products.

  The partial character string similarity evaluation means evaluates the similarity between the partial character strings included in the phrase by adding a first length and a second length to a third length portion. If the character strings match, the evaluation is equal to or higher than the evaluation when the partial character strings of the first length match and the evaluation when the partial character strings of the second length match. Give a high rating. The phrase similarity evaluation means causes the character string similarity evaluation means to evaluate the similarity between the partial character strings obtained by dividing each of the two phrases, and the evaluation by the character string similarity evaluation means By summing up the results, a combination of division patterns that divide each of the two phrases into one or more partial character strings is evaluated, and evaluations for a plurality of combinations of the division patterns of the two phrases are used. Then, the similarity between the two words is evaluated.

  Further, the side effect phrase extracting unit extracts a phrase indicating the side effect of the determination target drug from the report document as a determination target side effect phrase. The comparison target set acquisition unit reads out a package insert of the medicine recognized as the analog by the analog drug recognition unit from the storage unit, and uses a phrase extraction process to extract a phrase from a side effect description part in the package insert. A set of phrases obtained by extracting a set of phrases or by a phrase extraction process from a side effect description part in a package insert of the medicine recognized as the similar drug by the similar drug recognition means from the storage means By reading side effect learning result information that associates the identification information of the drug recognized as the similar drug with the identification information of the drug, the word included in the description part of the side effect in the package insert of the drug recognized as the similar drug The set is acquired as a comparison target phrase set.

  The determination means includes, for at least a part of the analog, a combination of a word / phrase included in the comparison target phrase set acquired for the analog and the determination target side-effect word / phrase, and the phrase similarity evaluation means Then, using the evaluation result and the threshold value, it is determined whether or not the side effect indicated by the determination target side effect phrase is a known side effect in the related drug. And the said output means outputs the determination result by the said determination means.

  According to the information processing apparatus, it is automatically determined whether or not the side effect is a known side effect in the analog of the determination target drug from the report document describing the side effect of the determination target drug, and the determination result is output. . Therefore, the information processing apparatus described above can automate a determination that has been conventionally performed manually in a pharmaceutical company.

It is a block diagram of a determination apparatus. It is a block diagram of a computer. It is a flowchart of a side effect determination / learning process. It is a figure explaining the example of a learning result table. It is a figure explaining the example of a side effect determination result screen. It is a flowchart of a side effect similarity calculation process. It is a flowchart of a score calculation process. It is a figure explaining the constant value utilized for a process in the determination apparatus. It is a figure which illustrates the specific example of a score calculation process typically. It is a flowchart of a score normalization process. It is a flowchart of the pre-processing which a determination apparatus performs before the start of normal operation. It is a flowchart of an analog medicine learning process. It is a flowchart of an effect and effect similarity calculation process. It is a flowchart of an addition / update process.

Hereinafter, embodiments will be described in detail in the following order with reference to the drawings.
First, the configuration of a determination apparatus that determines whether a certain side effect is known or unknown will be described with reference to FIG. 1, and a specific example of hardware that realizes the determination apparatus will be described with reference to FIG. Thereafter, processing performed by the determination apparatus when a report of a side effect is received from a clinical site or the like will be described with reference to FIGS. In the embodiment shown in FIGS. 1 to 14, the above assumption can be established by the preprocessing performed by the determination device. Then, with reference to FIGS. 11-14, the learning of the analog by pre-processing is demonstrated. Finally, various modifications will be described.

  FIG. 1 is a configuration diagram of the determination apparatus. The determination apparatus 100 in FIG. 1 includes a storage unit 101 that stores various types of information, a side effect processing unit 102 that performs processing related to side effects, and an analog medicine processing unit 103 that performs processing related to analog drugs.

  The side effect processing unit 102 includes a determination target drug designating unit 104, a similar drug recognition unit 105, a report document acquisition unit 106, a side effect keyword extraction unit 107, and a side effect determination / learning unit 108. Further, the side effect processing unit 102 includes a keyword similarity evaluation unit 109, but in this embodiment, the keyword similarity evaluation unit 109 is also shared with the analog medicine processing unit 103. In addition to the keyword similarity evaluation unit 109 shared with the side effect processing unit 102, the similar drug processing unit 103 further includes an efficacy / effect keyword extraction unit 111 and an analog determination / learning unit 112.

  Further, the keyword similarity evaluation unit 109 includes a partial character string similarity evaluation unit 110, and the determination apparatus 100 further includes a preprocessing control unit 113. The preprocessing control unit 113 performs control for causing the side effect keyword extraction unit 107 in the side effect processing unit 102 and the similar drug determination / learning unit 112 in the similar drug processing unit 103 to perform preprocessing.

  FIG. 1 also illustrates information stored in the storage unit 101 and information input to the determination apparatus 100. The operation of each unit in the side effect processing unit 102 and the similar drug processing unit 103 will be described after various information has been described.

  As illustrated in FIG. 1, the storage unit 101 stores an attached document group 201, and the attached document group 201 includes an attached document 202 for each of a plurality of medicines. As described above, the package insert is, for example, a document including sections such as “medicinal effect classification name”, “market name code”, “efficacy or effect”, “side effects”, “physicochemical knowledge about active ingredients”, and the like.

  As the “medicinal effect classification name”, for example, names that are standardized to some extent, such as “antipyretic analgesic / anti-inflammatory agent” and “blood pressure lowering agent”, are used. When one pharmaceutical product is used for a plurality of uses, a plurality of drug effect classification names may be described.

The “sales name code” is identification information for uniquely identifying each medicine. In the present embodiment, the sales name code is used as an IDentification (ID) of a medicine.
“Effectiveness or effect” and “side effect” sections have a certain degree of freedom in description, and may be described in natural language sentences, or may be described in the form of lists or tables.

  For example, the “Efficacy or Effect” section is described in the form of a list of diseases or symptoms that the drug is effective, such as “rheumatoid arthritis, rheumatic fever, osteoarthritis, etc.”, separated by readings. Sometimes. Further, the side effect may be described in a natural language sentence such as “may cause an asthma attack” or may be described in a table format. For example, in a table having a heading column indicating what is a side effect, a column indicating the frequency of occurrence of the side effect, and a column describing a specific side effect, the frequency of “less than 0.1%” in the heading “blood” And a specific side effect of “anemia” may be associated.

  The “physicochemical knowledge about active ingredient” section may include a generic name such as “aspirin”, a chemical name such as “2-Acetoxybenzoic acid”, a molecular formula, a structural formula, and the like. Also, depending on the package insert, the general name of the active ingredient may be described in a section with another heading such as “general name” or “reference name”.

  A specific example of the package insert can be viewed, for example, on the “medical device information provision website” (http://www.info.pmda.go.jp/) of the Pharmaceuticals and Medical Devices Agency. The attached document includes various items other than those exemplified above, but a detailed description of the attached document is omitted.

  The storage unit 101 stores a synonym dictionary 203 and a learning result table 204 in addition to the attached document group 201 including the plurality of attached documents 202 as described above. The synonym dictionary 203 is a dictionary in which pairs of synonyms or synonyms are registered, and the learning result table 204 is a table that holds the results of learning by the side effect processing unit 102 and the analog medicine processing unit 103.

FIG. 1 illustrates the following two entries in the synonym dictionary 203 in a table format, but the contents and format of the synonym dictionary 203 are arbitrary depending on the embodiment.
・ Entry with “AST” (ASpartate aminoTransferase) and “GOT” (Glutamic Oxaloacetic Transaminase) as a pair ・ Entry with “whole body” and “body” as a pair

  The synonym dictionary 203 may be prepared and prepared in advance, or may be empty in the initial state. In this embodiment, learning of the synonym dictionary 203 is performed, and the number of entries in the synonym dictionary 203 increases as time passes. However, an embodiment in which the contents of the synonym dictionary 203 are fixed (that is, learning of the synonym dictionary 203 is performed). Embodiments that do not perform are also possible.

  The learning result table 204 is a table in which each entry corresponds to an individual medicine, as will be described in detail later with reference to FIG. Each entry includes fields of “ID”, “Efficacy / effect keyword group”, “Similar drug list”, “Side effect keyword group”, and “Known side effect list” as shown in FIG.

  In the learning result table 204, the sales name code is used as the ID as described above. The effect / effect keyword group is a keyword group extracted in advance from the “effect or effect” section of the attached document 202. The similar drug list is a list of similar drug IDs, and is prepared in advance by the processing described later with reference to FIGS. The side effect keyword group is a keyword group extracted in advance from the “side effect” section of the attached document 202. The known side effect list is a list for learning side effects determined to be known side effects in the similar drugs.

  In the present embodiment, a phrase extracted from a natural language sentence, a list, a table, or the like is referred to as a “keyword”. The keyword may be, for example, a single noun, a compound noun in which two or more nouns are connected, or a noun phrase including a combination of an adjective and one or more nouns.

As shown in FIG. 1, a safety information report document 205 and a determination target drug ID 206 are input to the determination apparatus 100.
The safety information report document 205 is a report document describing side effects. More specifically, the safety information report document 205 is a report document related to safety information that is brought from the clinical site to the pharmaceutical company after the sale of the drug. In the present specification, “safety information” refers to information regarding side effects caused by (or suspected of) the administration of pharmaceuticals by the pharmaceutical company.

  Note that both the attached document 202 and the safety information report document 205 include information on side effects, but they are completely different documents. That is, the attached document 202 is a document created by a pharmaceutical company in accordance with the Pharmaceutical Affairs Law, whereas the safety information report document 205 is a report document created by a clinical doctor or the like for reporting to the pharmaceutical company. It is.

  The format of the safety information report document 205 is arbitrary. For example, the safety information report document 205 may include a description in a natural language sentence regarding side effects, or may include a list or a table that lists side effects.

  When the safety information report document 205 is brought from the clinical site, the pharmaceutical company is obliged to report to the Minister of Health, Labor and Welfare whether the side effects reported in the safety information report document 205 are known or unknown. . This obligation is stipulated in the Pharmaceutical Affairs Law as described above.

  The determination apparatus 100 according to the present embodiment determines whether the side effects reported in the safety information report document 205 are known or unknown in order to support prompt and accurate reporting from the pharmaceutical company to the Minister of Health, Labor and Welfare. Judge automatically or semi-automatically. As will be described in detail later, the term “semi-automatic” is used here to mean that the determination apparatus 100 according to the present embodiment is an intermediate such that “the side effect cannot be determined as a known side effect, but there is a certain degree of known possibility” This refers to a case in which a final determination is made and left to the user input.

  Specifically, in the present embodiment, the determination target drug ID 206 specifies which drug the safety information reported in the safety information report document 205 relates to. That is, the determination target drug ID 206 is an ID of a drug for which a side effect is reported by the safety information report document 205. Hereinafter, the pharmaceutical specified by the determination target drug ID 206 is referred to as “determination target drug”.

  Now, the processing performed by the side effect processing unit 102 and the similar medicine processing unit 103 regarding the information stored in the storage unit 101 and the information input to the determination apparatus 100 as described above will be described.

The determination target drug specifying unit 104 receives the input of the determination target drug ID 206 and outputs the determination target drug ID 206 to the analogy drug recognition unit 105 and the side effect determination / learning unit 108.
The determination target drug specifying unit 104 is an example of a specifying unit that receives information for specifying a drug and specifies the drug indicated by the received information as a determination target drug. Depending on the embodiment, the information for specifying the drug may not be the determination target drug ID 206, and may be a combination of one or more items such as a drug name, a general name of the active ingredient, and a chemical name of the active ingredient. It may be a search condition. In this case, the specifying unit can be realized by a search unit that receives an input of a search condition and specifies a determination target drug by searching for a medicine that matches the search condition. The search unit may extract and index items used for search from the attached document 202 in advance.

  Further, the similar medicine recognition unit 105 searches the learning result table 204 using the input determination target drug ID 206 as a search key, and reads out the similar drug list from an entry whose ID matches the determination target drug ID 206, thereby determining the determination target. Recognize the similar drugs. Then, the similar drug recognition unit 105 outputs the read similar drug list to the side effect determination / learning unit 108.

  On the other hand, the report document acquisition unit 106 acquires the safety information report document 205 for the determination target drug and outputs it to the side effect keyword extraction unit 107. Then, the side effect keyword extraction unit 107 extracts one or more keywords indicating the side effect of the determination target drug from the safety information report document 205 and outputs the extracted keywords to the side effect determination / learning unit 108. The side effect keyword extracting unit 107 is an example of a side effect phrase extracting unit that extracts a phrase indicating a side effect of a determination target drug as a determination target side effect phrase.

  Then, the side effect determination / learning unit 108 determines, for each keyword extracted by the side effect keyword extraction unit 107 from the safety information report document 205, whether the side effect indicated by the keyword is known or unknown, and learns the determination result.

  Specifically, the side effect determination / learning unit 108 searches the learning result table 204 using each ID included in the similar drug list input from the similar drug recognition unit 105 as a search key, and sets a side effect keyword group for each similar drug. Get. Then, the side effect determination / learning unit 108 causes the keyword similarity evaluation unit 109 to evaluate the similarity between each keyword in the side effect keyword group and the keyword extracted from the safety information report document 205 by the side effect keyword extraction unit 107.

  The keyword similarity evaluation unit 109 causes the partial character string similarity evaluation unit 110 to evaluate the similarity between partial character strings obtained by dividing each of the two keywords specified as targets for evaluating the similarity. Then, the keyword similarity evaluation unit 109 evaluates combinations of division patterns that divide each of the two keywords into one or more partial character strings by counting the evaluation results by the partial character string similarity evaluation unit 110. To do. The keyword similarity evaluation unit 109 evaluates the degree of similarity between two keywords by using the evaluation for a plurality of combinations of the division patterns of the two keywords.

  Then, the side effect determination / learning unit 108 uses the result of the evaluation by the keyword similarity evaluation unit 109 to determine which side effect indicated by the keyword extracted from the safety information report document 205 is the side effect of the determination target drug analog Determine if they are similar to some extent.

  If the degree of similarity is clearly high, the side effect determination / learning unit 108 determines that the side effect indicated by the keyword extracted from the safety information report document 205 is “a known side effect already known in a similar drug”, and the determination Output the result. If the degree of similarity is clearly low, the side effect determination / learning unit 108 determines that the side effect indicated by the keyword extracted from the safety information report document 205 is “an unknown side effect that is not known even in similar drugs”. And the determination result is output.

  If the degree of similarity is medium, the side effect determination / learning unit 108 determines that the side effect indicated by the keyword extracted from the safety information report document 205 is “a side effect that may be known”, and outputs the determination result. To do. In this case, the side effect determination / learning unit 108 prompts the user to input whether the side effect determined to be “may be known” is known or unknown, and receives an input from the user.

  Then, the side effect determination / learning unit 108 learns as a known side effect for the keyword that is determined to be “known” by itself and the keyword for which the determination of “known” is input by the user. That is, the side effect determination / learning unit 108 updates the known side effect list in the entry corresponding to the determination target drug in the learning result table 204.

  Further, the side effect determination / learning unit 108, based on the determination result, is a pair of keywords extracted from the safety information report document 205 and keywords included in the side effect keyword group of similar drugs, which can be regarded as synonyms, It is added to the synonym dictionary 203.

  The side effect keyword group in the learning result table 204 referred to by the side effect determination / learning unit 108 is obtained in advance by preprocessing performed by the side effect keyword extracting unit 107 in accordance with a command from the preprocessing control unit 113. That is, the side effect keyword extraction unit 107 performs the following preprocessing for each attached document 202 in the attached document group 201 as preprocessing in addition to the operation described above.

  The side effect keyword extraction unit 107 extracts a keyword indicating a side effect from the “side effect” section of the attached document 202. Then, the side effect keyword extraction unit 107 searches the learning result table 204 for an entry corresponding to the attached document 202, and additionally registers the keyword extracted from the attached document 202 in the side effect keyword group of the searched entry.

  In the present embodiment, each attached document 202 is associated with the ID by including the ID of the medicine to which the attached document 202 is attached in the file name. Therefore, the side effect keyword extraction unit 107 can recognize to which entry of the learning result table 204 the keyword extracted from which attached document 202 should be registered.

  However, the method for associating the attached document 202 with the ID is arbitrary depending on the embodiment. For example, the learning result table 204 may include a field indicating the file name of the attached document 202, and the ID and the attached document 202 may be associated with each other by the field. That is, the side effect keyword extraction unit 107 may recognize the association between the attached document 202 and the ID by referring to the field. In addition, on the homepage for providing information on medical devices for pharmaceuticals, attached documents are disclosed in Standard Generalized Markup Language (SGML) format and Portable Document Format (PDF) format, but in this embodiment, the file format of the attached document 202 is arbitrary. .

  As described above, the side effect processing unit 102 can automatically or semi-automatically determine whether the side effect reported in the safety information report document 205 is known or unknown. Therefore, pharmaceutical companies can make prompt reports to the Minister of Health, Labor and Welfare.

  Further, as will be described in detail later, in the present embodiment, while matching long character strings is evaluated to be higher than matching short character strings, even if long character strings do not necessarily match completely, partial matching may occur. A policy that gives a certain degree of evaluation if it agrees with is adopted. Specifically, the keyword similarity evaluation unit 109, the side effect determination / learning unit 108, and the similar drug determination / learning unit 112 operate according to the above policy.

  According to the above policy, the accuracy of determination by the side effect determination / learning unit 108 can be increased. Further, according to the above policy, it is possible to perform learning of an analog medicine list in the learning result table 204 described below with high accuracy and as little leakage as possible.

  Because, firstly, even if long character strings do not necessarily match completely, if they are partially matched, a certain degree of evaluation is given to absorb fluctuations in notation and inconsistency of terms. This is because the similarity between keywords representing similar concepts can be evaluated.

  For example, the two keywords "general anesthetic" and "general inhalation anesthetic" do not match completely, but they match in the substrings "systemic" and "anesthetic" and have strong semantic relevance. have. Therefore, instead of setting the evaluation to zero for partial matching of character strings between keywords, giving a certain degree of evaluation, evaluates the semantic match or similarity between keywords well, and evaluates Can improve the accuracy. Also, even if there is a difference in notation between similar keywords, giving a certain degree of evaluation to partial matching of character strings reduces the risk of overlooking similarities, so learning without omission is possible It becomes.

  The second reason is that, as a general tendency, long keywords often represent semantically limited contents, and most of noise in keyword extraction is short keywords. Therefore, it is possible to reduce the influence of noise by placing importance on matching long character strings.

  For example, a short keyword “symptom” may be extracted from a natural sentence such as “please discontinue administration if such a symptom appears” in the “side effect” section of the package insert 202. However, the keyword “symptom” thus extracted does not indicate a specific side effect. On the other hand, long keywords such as “anaphylactoid symptoms” often show specific side effects. Therefore, the influence of noise can be reduced by placing importance on matching long character strings.

  And the 3rd reason is because the bad influence of the excessive evaluation resulting from partial matching can be suppressed by attaching importance to matching of long character strings. In some cases, by giving an evaluation to partial matching of character strings, there is a possibility that the similarity between keywords may be evaluated excessively high, but in this embodiment, there is an adverse effect caused by excessive evaluation. Can be suppressed.

  In other words, in this embodiment, based on a policy that places importance on matching long character strings, the similarity between keywords that match several short partial character strings is the total length of those partial character strings. It is evaluated below the degree of similarity between keywords matching the corresponding long character string. For example, even if three characters in a certain place are accidentally shared by two keywords, the similarity between the two keywords is the similarity between two keywords that share three consecutive characters. Only evaluated below. Therefore, by placing importance on matching long character strings, noise due to accidental matching of a plurality of short partial character strings can be reduced.

  For the reasons as described above, according to the present embodiment, the side effect determination / learning unit 108 can perform high-precision determination, the analogy drug determination / learning unit 112 can accurately learn the analogy drug list, and It can also be carried out so as not to leak as much as possible. The side effect determination / learning unit 108 can eventually improve the determination accuracy by using the analogy medicine list learned with high accuracy and as little leakage as possible.

  Because, if the accuracy of the similar drug list is poor, there is a risk that the side effect determination / learning unit 108 may erroneously determine “known side effects” for side effects that are known only for drugs that are not the target drug analogs. Because there is. In addition, if there is a leak in the similar drug list, there is a possibility that the side effect determination / learning unit 108 may erroneously determine “unknown side effect” for a known side effect in the similar drug of the determination target drug. Therefore, the side effect determination / learning unit 108 can improve the determination accuracy by using a similar medicine list learned with high accuracy and as little leakage as possible.

  Therefore, by using the determination apparatus 100 of the present embodiment, the pharmaceutical company can make an accurate report on side effects to the Minister of Health, Labor and Welfare more quickly than before.

  Now, learning of the analog medicine list in the learning result table 204 as described above is performed in advance by the analog medicine processing unit 103 in accordance with an instruction from the preprocessing control unit 113 in this embodiment. In other words, in the present embodiment, the similar medicine determination / learning unit 112 determines that two medicines corresponding to the two attached documents 202 are similar to each other with respect to a set of arbitrary two attached documents 202 in the attached document group 201. It is determined whether or not each other, and the determined result is learned.

  Specifically, the efficacy / effect keyword extraction unit 111 first extracts a keyword from the “efficacy or effect” section for each attached document 202. The efficacy / effect keyword extraction unit 111 searches the learning result table 204 for an entry corresponding to the attached document 202, and additionally registers the extracted keyword in the efficacy / effect keyword group of the searched entry. The effect / effect keyword extraction unit 111 can recognize the correspondence between the attached document 202 and the ID, for example, by the file name of the attached document 202.

  In addition, in the similar drug determination / learning unit 112, the two package inserts 202 that are to be determined whether or not they are similar to each other are identical in the “medicinal effect classification name” or “physical and chemical knowledge regarding active ingredients” section. The two medicines are regarded as similar drugs.

  On the other hand, if no match is detected in the “medicinal effect classification name” or “physical and chemical knowledge regarding active ingredient” section, the analogy drug determination / learning unit 112 determines the similarity between the “efficacy or effect” sections of the two package inserts 202. Find the degree. That is, the similar medicine determination / learning unit 112 obtains the similarity between the learned effect / effect keyword groups in the entries in the learning result table 204 corresponding to the two attached documents 202, respectively.

  Specifically, the analogy drug determination / learning unit 112 determines the keyword in the efficacy / effect keyword group of the entry corresponding to one attached document 202 and the keyword in the efficacy / effect keyword group of the entry corresponding to the other 202. Is evaluated by the keyword similarity evaluation unit 109. Then, the similar medicine determination / learning unit 112 determines how similar the “efficacy or effect” sections of the two attached documents 202 are based on the evaluation result by the keyword similarity evaluation unit 109.

  If the degree of similarity is clearly high, the similar medicine determination / learning unit 112 determines that the two medicines are “similar medicines” and outputs a determination result. If the degree of similarity is clearly low, the similar drug determination / learning unit 112 determines that the two medicines are not similar drugs, and outputs a determination result.

  Then, if the degree of similarity is moderate, the similar drug determination / learning unit 112 determines that the two drugs are “a possible combination of similar drugs”, and outputs a determination result. In this case, the similar medicine determination / learning unit 112 urges the user to input whether or not the two medicines determined as “possibility of similar medicines” are similar medicines, and receives input from the user .

  Then, the similar drug determination / learning unit 112 determines the relationship between similar drugs with respect to the combination of drugs that have been determined as “analogous drugs” and the combination of drugs that have been input by the user as “drugs”. learn. That is, the similar drug determination / learning unit 112 updates the similar drug list in the entry in the learning result table 204 corresponding to each of two drugs that are similar drugs.

  Furthermore, when the two similar drugs are determined to be similar according to the similarity between the “efficacy or effect” sections of the two package inserts 202, the similar drug determination / learning unit 112 uses a pair of keywords indicating the efficacy or effect. What can be regarded as synonyms are added to the synonym dictionary 203.

  Incidentally, the determination apparatus 100 of FIG. 1 can be realized by a dedicated hardware circuit, a general-purpose information processing apparatus that executes a program, or a combination thereof, but in the present embodiment, the determination apparatus 100 includes an information processing apparatus that executes a program. Is realized. Therefore, subsequently, what kind of hardware is used to implement each unit of the determination apparatus 100 will be described.

FIG. 2 is a configuration diagram of the computer. A computer 300 in FIG. 2 is an example of a general-purpose information processing apparatus that executes a program.
The computer 300 includes a central processing unit (CPU) 301, a read only memory (ROM) 302, a random access memory (RAM) 303, and a communication interface 304. The computer 300 also includes an input device 305, an output device 306, a storage device 307, and a drive device 308 for a portable storage medium 310. The CPU 301, ROM 302, RAM 303, communication interface 304, input device 305, output device 306, storage device 307, and drive device 308 are connected to each other via a bus 309.

  The input device 305 may be, for example, a keyboard, a pointing device device such as a mouse, or a combination thereof. The output device 306 is, for example, a display such as a liquid crystal display, a speaker, a printer, or any combination thereof.

  The storage device 307 is a hard disk device, a nonvolatile semiconductor memory device such as a flash memory, or a combination thereof. As the portable storage medium 310, an optical disk such as a Compact Disc (CD) or a Digital Versatile Disk (DVD), a non-volatile semiconductor memory card such as a magneto-optical disk, a magnetic disk, or a flash memory can be used.

  The computer 300 may be connected to another computer 312 via a communication interface 304 and a network 311 as shown in FIG. The network 311 may be an arbitrary network such as a local area network (LAN) or the Internet.

  The CPU 301 loads the program into the RAM 303 and executes the program while using the RAM 303 as a work area, thereby realizing the functions of the side effect processing unit 102, the similar medicine processing unit 103, and the preprocessing control unit 113 in FIG. it can. The program executed by the CPU 301 may be installed in the ROM 302 or the storage device 307 in advance, provided by being stored in the portable storage medium 310, read by the drive device 308, and copied to the storage device 307. May be. Alternatively, the program may be downloaded from the other computer 312 to the storage device 307 via the network 311.

More specifically, the correspondence between FIG. 1 and FIG. 2 in the present embodiment will be described as follows.
The storage unit 101 in FIG. 1 is realized by the storage device 307 in FIG. 1 is realized by the input device 305 and the CPU 301 in FIG. That is, the determination target drug ID 206 is input from the input device 305 and recognized by the CPU 301. 1 is realized by the CPU 301.

  Further, the safety information report document 205 of FIG. 1 may be given from another computer 312, and in that case, the report document acquisition unit 106 of FIG. 1 may be realized by the communication interface 304 and the CPU 301. Alternatively, the safety information report document 205 may be read from the portable storage medium 310, and in that case, the report document acquisition unit 106 may be realized by the drive device 308 and the CPU 301 of the portable storage medium 310. Alternatively, the safety information report document 205 may be input from the input device 305, and in that case, the report document acquisition unit 106 may be realized by the input device 305 and the CPU 301.

  Moreover, the safety information report document 205 is stored in the storage device 307 or the RAM 303 regardless of where it is given. That is, the side effect keyword extraction unit 107 to which the report document acquisition unit 106 outputs the safety information report document 205 further includes the storage device 307 or the RAM 303.

  The side effect keyword extraction unit 107 in FIG. Further, the side effect determination / learning unit 108 of FIG. 1 is realized by a CPU 301 that performs processing according to a program, and an output device 306 and an input device 305 for realizing a user interface.

  The keyword similarity evaluation unit 109 and the efficacy / effect keyword extraction unit 111 including the partial character string similarity evaluation unit 110 of FIG. 1 can also be realized by the CPU 301. Similarly to the side effect determination / learning unit 108, the analogy drug determination / learning unit 112 is realized by the CPU 301, the output device 306, and the input device 305. The preprocessing control unit 113 can also be realized by the CPU 301 and the input device 305 that receives an input for instructing the start of the preprocessing.

  The outline of the configuration and operation of the determination apparatus 100 of FIG. 1 realized by the computer 300 of FIG. 2 has been described above, and the outline of information used by the determination apparatus 100 has also been described. Therefore, hereinafter, the operation of the side effect processing unit 102 will be described in more detail with reference to FIGS.

FIG. 3 is a flowchart of the side effect determination / learning process.
In step S <b> 101, the determination target drug specifying unit 104 receives the input of the determination target drug ID 206, identifies the determination target drug, and outputs the determination target drug ID 206 to the similar drug recognition unit 105 and the side effect determination / learning unit 108.

  Then, in the next step S102, the similar drug recognition unit 105 searches the learning result table 204 using the determination target drug ID 206 input from the determination target drug specifying unit 104 as a search key, and the ID matches the determination target drug ID 206. Get a list of similar drugs from an entry.

Here, an example of the learning result table 204 will be described in more detail with reference to the example of FIG.
As described above, the learning result table 204 includes fields of “ID”, “efficacy / effect keyword group”, “similar medicine list”, “side effect keyword group”, and “known side effect list”. As will be described in detail later, the known side effect list is set in step S116 or S117 of FIG. 3, and the other fields are set in advance by the processing described later with reference to FIGS.

  FIG. 4 illustrates a learning result table 204 including entries described in the following (a1) to (a5). In FIG. 4, some keywords omitted for convenience of paper width are indicated as “...

(A1) Entry with ID “111222A3333” The efficacy / effect keyword group in this entry includes the keyword “bronchitis”. In addition, the similar medicine list in this entry includes an ID “396396B7777” indicating an entry not shown in FIG. 4 and an ID “998877F5050” indicating the entry (a5) below. In this entry, the side effect keyword group includes the keywords “headache” and “pulmonary hemorrhage”, and the known side effect list includes the keyword “headache”.

  For example, when a side effect of a headache is newly reported from a clinical site regarding a drug with an ID of “111222A3333”, the headache is known as a side effect in the similar drug indicated by the ID of “396396B7777”. After that, it is assumed that the package insert 202 of the medicine with the ID “111222A3333” is revised and a headache is added to the “side effect” section. For example, in the above case, the entry (a1) is in the state shown in FIG.

(A2) Entry with ID “234567F0909” The efficacy / effect keyword group in this entry includes the keywords “urticaria” and “eczema”. Further, the similar medicine list in this entry includes an ID “567567A1212” indicating an entry not shown in FIG. In this entry, the side effect keyword group includes the keywords “diabetes” and “anemia”, and the known side effect list is empty.

(A3) Entry with ID “444555A7777” The efficacy / effect keyword group in this entry includes the keywords “arthritis” and “acute upper respiratory tract inflammation”. The similar medicine list in this entry includes an ID “777788C9090” indicating the entry (a4) below and an ID “998877F5050” indicating the entry (a5) below. In this entry, the side effect keyword group includes the keywords “anemia” and “pulmonary hemorrhage”, and the known side effect list is empty.

(A4) Entry with ID “777788C9090” The efficacy / effect keyword group in this entry includes the keywords “acute upper respiratory tract inflammation” and “antipyretic”. The analogy drug list in this entry includes an ID “444555A7777” indicating the entry (a3) and an ID “998877F5050” indicating the entry (a5) below. In this entry, the side effect keyword group includes the keywords “heart failure” and “gastric ulcer”, and the known side effect list is empty.

(A5) Entry with ID “998877F5050” The efficacy / effect keyword group in this entry includes the keywords “acute upper respiratory tract inflammation”, “antipyretic fever”, and “bronchitis”. Further, the analogy medicine list in this entry includes an ID “111222A3333” indicating the entry (a1), an ID “444555A7777” indicating the entry (a3), and “777888C9090 indicating the entry (a4). "Is included. In this entry, the side effect keyword group includes the keywords “blood pressure reduction” and “anemia”, and the known side effect list includes the keyword “heart failure”.

  For example, it is assumed that heart failure has been reported from the clinical site as a side effect for a drug with an ID of “998877F5050” corresponding to the entry of (a5). At that time, in the similar drug corresponding to the entry (a4), the side effect of heart failure is described in the package insert 202, and as shown in FIG. 4, the side effect keyword group of the entry (a4) is “heart failure”. ”Is included. As a result, in the situation where heart failure is learned as “a known side effect in a similar drug”, while the package insert 202 of the drug corresponding to the entry in (a5) has not yet been revised, the entry in (a5) The state shown in FIG. 4 is obtained.

  Returning to the description of FIG. For convenience of explanation, for example, it is assumed that the determination target drug ID 206 input in step S101 is an ID “998877F5050” indicating the entry (a5). Then, in step S102, the similar drug recognition unit 105 acquires the similar drug list (111222A3333, 444555A7777, 777888C9090).

  In step S 103, the report document acquisition unit 106 receives an input of the safety information report document 205 related to the determination target drug, and outputs the safety information report document 205 to the side effect keyword extraction unit 107.

  Note that step S103 may be performed before steps S101 to S102. In some embodiments, the safety information report document 205 may include the ID of a target drug for reporting side effects. In this case, the determination target drug specifying unit 104 extracts the determination target drug ID 206 from the safety information report document 205 received in step S103 instead of receiving an explicit determination target drug ID 206 input in step S101. The determination target drug may be specified.

  After execution of step S103, in step S104, the side effect keyword extraction unit 107 extracts a keyword indicating the side effect of the determination target drug from the safety information report document 205, and outputs the extracted keyword to the side effect determination / learning unit 108. Note that the algorithm used by the side effect keyword extraction unit 107 for keyword extraction may vary depending on the embodiment.

  For example, the side effect keyword extracting unit 107 may perform morphological analysis on the safety information report document 205 and extract one or more series of nouns as keywords. The side effect keyword extraction unit 107 may further perform syntax analysis on the result of morphological analysis, and extract a keyword using the result of syntax analysis.

  Alternatively, depending on the format of the safety information report document 205, the side effect keyword extraction unit 107 may be able to acquire a keyword simply by cutting out a predetermined item in the safety information report document 205.

  Of course, the side effect keyword extraction unit 107 performs morphological analysis on only a part of the safety information report document 205 according to the format of the safety information report document 205 even when processing such as morphological analysis is performed. Can do. For example, when the safety information report document 205 includes a column of information such as a medical history regarding a patient who has had side effects, a column describing side effects, and a column indicating the name of a doctor who is a reporter, the side effect keyword extracting unit 107 For morphological analysis, only the data in the column describing the side effects should be extracted.

  Alternatively, the side effect keyword extraction unit 107 may perform simple keyword extraction processing based on the character type. For example, the side effect keyword extraction unit 107 may extract a sequence of kanji, katakana, or English as a keyword in view of the characteristics of medical terms.

  Further, for example, if dictionary data of various side effects is available, the side effect keyword extracting unit 107 may extract a character string that matches the dictionary entry from the safety information report document 205 as a keyword.

  In any case, the side effect keyword extraction unit 107 extracts keywords indicating side effects from the safety information report document 205 acquired by the report document acquisition unit 106 according to an appropriate algorithm according to the embodiment. Then, the side effect keyword extraction unit 107 notifies the extracted keyword to the side effect determination / learning unit 108.

  Note that there may be one or more keywords extracted in step S104. Hereinafter, for convenience of explanation, it is assumed that three keywords “heart failure”, “headache”, and “cerebral hemorrhage” are extracted in step S104.

  In subsequent step S <b> 105, the side effect determination / learning unit 108 determines whether or not an unprocessed side effect remains among those reported in the safety information report document 205. That is, the side effect determination / learning unit 108 determines whether or not the processing after step S106 has been performed for all the keywords notified from the side effect keyword extraction unit 107 in step S104.

  If unprocessed side effects remain, the process proceeds to step S106. Conversely, if all the side effects have been processed, the process proceeds to step S117. If there is no possibility of misunderstanding, the “keyword indicating a side effect” may be simply referred to as “side effect” for the sake of simplification.

  In step S106, the side effect determination / learning unit 108 selects one unprocessed side effect (that is, an unprocessed keyword). Hereinafter, the side effect selected in step S106 is referred to as “selected side effect”. For example, when the three keywords “heart failure”, “headache”, and “cerebral hemorrhage” are extracted in step S104 as described above, the side effect determination / learning unit 108 selects “heart failure” in the first execution of step S106. But you can.

  In next step S107, the side effect determination / learning unit 108 determines whether or not the selected side effect has been learned. That is, the side effect determination / learning unit 108 searches the learning result table 204 for an entry whose ID matches the determination target drug ID 206, and acquires a known side effect list of the found entry.

  Then, if there is an element in the acquired known side effect list that matches the selected side effect or a synonym of the selected side effect, the side effect determination / learning unit 108 states that “the selected side effect has been learned”. judge. Conversely, if there is no element that matches the selected side effect in the acquired list of known side effects and no element that matches the synonym of the selected side effect, the side effect determination / learning unit 108 determines that “the selected side effect has not been learned. Is determined.

  The side effect determination / learning unit 108 can recognize whether or not a synonym exists in the selected side effect by searching the synonym dictionary 203 using the selected side effect as a search key. Can be recognized as well.

  If it is determined in step S107 that the selected side effect has been learned, the process proceeds to step S108. Conversely, if the selected side effect has not been learned, the process proceeds to step S109.

  For example, when “heart failure” is selected in step S106 as in the above example, the known side effect list of the entry (a5) corresponding to the determination target drug in the learning result table 204 of FIG. Since the element is included, the selective side effects have been learned. Therefore, the process proceeds to step S108.

  Conversely, if “headache” or “cerebral hemorrhage” is selected in step S106, it is determined in step S107 that “the selected side effect has not been learned”. This is because there is no element of “headache” or its synonym and no element of “cerebral hemorrhage” or its synonym in the known side effect list of the entry (a5).

In step S108, the side effect determination / learning unit 108 outputs that the selected side effect is known. Then, the process returns to step S105.
Here, the output in step S108 will be described in more detail with reference to FIG. FIG. 5 is a diagram illustrating an example of the side effect determination result screen.

  The side effect determination result screen 400 of FIG. 5 includes a determination target drug display column 401 indicating the determination target drug and a determination result list 402 indicating the result of the processing of FIG. Depending on the processing result, the determination result list 402 may include a radio button 403 for allowing the user to determine whether a side effect is known or unknown, and the side effect determination result screen 400 may further include a learning button 404. Since the radio button 403 and the learning button 404 will be described later together with step S118, description thereof will be omitted here.

The side effect determination / learning unit 108 of the present embodiment outputs the side effect determination result screen 400 of FIG. 5 on a display corresponding to the output device 306 of FIG.
Although omitted in the description regarding FIG. 3 above, the side effect determination / learning unit 108 receives the keyword notification from the side effect keyword extraction unit 107 in step S104, for example, the determination target medicine display column 401, and the determination result list The header line 402 may be output. Alternatively, the side effect determination / learning unit 108 may output the determination target medicine display column 401 and the header line when step S108 or step S116 is executed for the first time.

  If the selected side effect is “heart failure” as in the above example and it is found that the selected side effect has been learned in step S107, the side effect determination / learning unit 108 displays the determination result table of FIG. 5 in step S108. A line of 402 “heart failure” may be output on the display.

  In the example of FIG. 5, the determination target drug display field 401 indicates the determination target drug by displaying the determination target drug ID. Of course, depending on the embodiment, the side effect determination / learning unit 108 may read, for example, the sales name or general name of the determination target drug from the attached document 202 of the determination target drug and output it to the determination target drug display field 401.

  Note that the side effect determination / learning unit 108 of the present embodiment embeds a link to the attached document 202 of the determination target drug in the determination target drug ID in the determination target drug display field 401. Therefore, the user can easily identify and refer to the attached document 202 of the determination target drug by clicking the link, and can confirm the validity of the determination result by the determination apparatus 100.

  Further, the determination result list 402 illustrated in FIG. 5 includes a header line and a line corresponding to each side effect keyword extracted from the safety information report document 205. The example of FIG. 5 is an example in the case where three keywords “heart failure”, “headache”, and “cerebral hemorrhage” are extracted in step S104 of FIG. 3, so that three keywords corresponding to these three keywords are displayed below the header line. There is a line.

  The header row of the determination result list 402 includes the headings “side effects”, “determination based on learning results”, “determination based on comparison with similar drugs”, and “determination input field”. Then, as illustrated in FIG. 4, the similar drug list of the medicine whose ID is “998877F5050” is (111222A3333,444555A7777,777788C9090). Therefore, under the heading “Judgment by comparison with similar drugs”, there are three columns corresponding to these three similar drugs, and the ID of each of these three similar drugs is displayed in the header row. .

  Note that the side effect determination / learning unit 108 of the present embodiment embeds a link to the attached document 202 of each similar drug in each of the three similar drug IDs. Therefore, the user can easily identify and refer to the attached document 202 of the similar medicine by clicking the link, and can confirm the validity of the determination result by the determination apparatus 100.

  The first line of the determination result list 402 in FIG. 5 is a line that is output when “heart failure” is selected as the selected side effect in step S106 in FIG. In this case, the side effect determination / learning unit 108 determines in step S107 that the selected side effect has been learned as a known side effect. Therefore, in step S108, the side effect determination / learning unit 108 outputs the keyword “heart failure” indicating the selected side effect in the column “side effect” of the determination result list 402, and “known” in the column “determination based on learning result”. Is output.

  In this case, it is not necessary to make a comparison with similar drugs. Therefore, in step S108, the side effect determination / learning unit 108 of this embodiment does not perform comparison with similar drugs in the three columns under the heading “Determination by comparison with similar drugs”. The character “-” is output.

  Similarly, in this case, there is no need to ask the user for judgment. Therefore, the side effect determination / learning unit 108 of the present embodiment outputs the character “-” in the column of “determination input field” in step S108.

  Returning to the description of FIG. If the side effect determination / learning unit 108 determines in step S107 that “the selected side effect has not been learned”, it initializes the “known side effect learning list” and “known side effect candidate list” to empty in step S109. The known side effect learning list is a list including, as an element, the ID of a similar drug whose side effect having a high degree of similarity with the currently selected selected side effect is described in the package insert 202. In addition, the known side effect candidate list is a list including, as an element, the ID of the similar drug in which the side effect having a medium similarity to the currently selected selected side effect is described in the package insert 202. After the initialization of the known side effect learning list and the known side effect candidate list, the process proceeds to step S110.

  In step S <b> 110, the side effect determination / learning unit 108 determines whether there are any unprocessed selected side effects that are currently focused on among the similar drugs to be determined. That is, the side effect determination / learning unit 108 does not perform the processing after step S111 regarding the selected side effect currently focused on among the IDs included in the analogy drug list notified from the analogy drug recognition unit 105 in step S102. Whether or not remains.

  If there are any unprocessed analogs for the currently selected selective side effect, the process proceeds to step S111. On the other hand, regarding the selected side effect that is currently focused on, if all the IDs included in the similar medicine list notified in step S102 have been processed, the process proceeds to step S116.

  In step S111, the side effect determination / learning unit 108 selects one unprocessed analog for the selected side effect that is currently focused on from among the analogs of the determination target drug. Hereinafter, the similar drug selected in step S111 is referred to as “selected drug”, and the ID of the selected drug is referred to as “selected drug ID”.

  Subsequently, in step S112, the side effect determination / learning unit 108 obtains the similarity between the selected side effect selected in step S106 and the keyword group in the “side effect” section of the attached drug 202 of the selected similar drug selected in step S111. Details of the processing in step S112 will be described later with reference to FIG. 6 as “side effect similarity calculation processing”, but the outline is as follows.

  That is, the side effect determination / learning unit 108 acquires from the learning result table 204 a side effect keyword group extracted from the “side effect” section of the attached document 202 of the selected analogy drug. Then, the side effect determination / learning unit 108 causes the keyword similarity evaluation unit 109 to evaluate the similarity to the selected side effect for each keyword included in the acquired side effect keyword group.

  Then, the side effect determination / learning unit 108 calculates the similarity by totaling the evaluation results. In the present embodiment, the higher the similarity, the larger the calculated value. After calculating the similarity, the process proceeds to step S113.

Then, in step S113, the side effect determination / learning unit 108 corresponds to which range the similarity obtained in step S112 falls within “α ₁ or more”, “α ₂ or more and less than α ₁ ”, or “less than α ₂ ”. Judging. In the present embodiment, α ₁ and α ₂ are predetermined threshold values, and α ₁ > α ₂ .

Although the threshold values α ₁ and α ₂ in the present embodiment are fixed values, the threshold values α ₁ and α ₂ are values determined so as to change according to the length of the selected side effect depending on the embodiment. Also good. Further, the comparison with the threshold value is not limited to the example of step S113, and the comparison with the threshold value may be a comparison “whether it exceeds the threshold value or less than the threshold value” or a comparison “whether it is more than the threshold value or less than the threshold value”. Well, it is possible to establish a policy as appropriate.

The threshold value α ₁ is a reference value indicating that it is appropriate to determine that the selected side effect is a “known side effect”. The threshold α ₂ is a reference value indicating a boundary between determining that the selected side effect is “unknown side effect” or “side effect that may be known”.

When the similarity is _one or more α obtained in step S112, the processing proceeds from step S113 to step S114. Also, similarity obtained in step S112 is a case of _two or more alpha less than ₁ alpha, the process proceeds from step S113 to step S115. When the similarity obtained in step S112 is less than alpha _2, the process returns from step S113 to step S110.

  Note that, as described in the outline of step S112, the side effect determination / learning unit 108 obtains a set of words / phrases included in the side effect description part in the attached document of the drug recognized as a similar drug as a comparison target word / phrase set. It is an example of object set acquisition means. Specifically, the side effect determination / learning unit 108 as the comparison target set acquisition unit reads the side effect learning result information on the medicine recognized as a similar drug from the learning result table 204 of the storage unit 101, thereby comparing the target phrase Get a set. The side effect learning result information related to a drug recognized as a similar drug is a set of words and phrases obtained in advance by word extraction processing from the side effect description part in the package insert 202 of the drug recognized as a similar drug. This corresponds to the entry associated with the identification information of the medicine.

  As is clear from the above description regarding steps S112 and S113, the side effect determination / learning unit 108 is also an example of a determination unit that determines whether or not the side effect indicated by the determination target side effect phrase is a known side effect in a similar drug.

That is, the side effect determination / learning unit 108 determines a combination of a phrase and a determination target side effect phrase included in the set of comparison target phrases acquired for the analog for at least a part of the similar drug, as a keyword similarity evaluation unit. 109. In other words, the side effect determination / learning unit 108 sets, for at least a part of the similar drug, the keyword similarity evaluation unit 109 for each keyword combination in the side effect keyword group acquired for the similar drug and the selected side effect. Let me evaluate. Then, the side effect determination / learning unit 108 uses the evaluation result by the keyword similarity evaluation unit 109 and the threshold value α ₁ to determine whether the side effect indicated by the determination target side effect phrase is a known side effect in the related drug. .

  In step S114, the side effect determination / learning unit 108 adds the selected drug ID to the known side effect learning list. That is, the side effect determination / learning unit 108 stores that “the selected side effect is a known side effect in the selected drug”. Then, the process returns to step S110.

  In step S115, the side effect determination / learning unit 108 adds the selected drug ID to the known side effect candidate list. That is, the side effect determination / learning unit 108 stores that “the selected side effect is somewhat similar to the known side effect in the selected drug, and may be a known side effect”. Then, the process returns to step S110.

  As described above, the side effect determination / learning unit 108 executes the processes after step S111 for each similar drug with respect to the currently selected selective side effect, so that the known side effect learning list regarding the currently selected selected side effect is listed. The list of known side effect candidates is completed. When the known side effect learning list and the known side effect candidate list are completed, the process proceeds from step S110 to step S116 as described above.

  In step S116, the side effect determination / learning unit 108 outputs a result regarding the selected side effect, and learns whether the selected side effect is known according to the known side effect learning list. With reference to FIG. 5 again, the operation of step S116 will be specifically described with reference to two examples.

  As a first example, a case where “headache” is selected as the selected side effect in step S106 will be described. Further, as described above, it is assumed that the determination target drug ID 206 is “998877F5050” and the learning result table 204 is as shown in FIG.

Then, when the selected drug ID is “111222A3333”, there is a keyword that matches the selected side effect (ie, “headache”) in the side effect keyword group of the entry of FIG. 4 described in (a1) above. Therefore, if the threshold value α ₁ is if properly defined, selected such drugs ID of "111222A3333" is added to the known side effects learning list.

On the other hand, there is no keyword similar to “headache” in the side effect keyword group of the entry of FIG. 4 described in (a3) and (a4) above. Therefore, if the thresholds α ₁ and α ₂ are appropriately determined, when the selected drug ID is “444555A7777” or “777888C9090”, the selected drug ID is added to both the known side effect learning list and the known side effect candidate list. Not.

  Therefore, in this first example, in step S116, the side effect determination / learning unit 108 determines the second side of the determination result list 402 in FIG. 5 as the determination result for the selected side effect “headache” as follows. Output.

  That is, the side effect determination / learning unit 108 outputs the keyword “headache” indicating the selected side effect in the “side effect” column, and outputs “unknown” in the “determination based on learning result” column. Since the ID “111222A3333” is included in the known side effect learning list, the side effect determination / learning unit 108 outputs “known” to the ID column “111222A3333” as “determination based on comparison with similar drugs”. To do.

  On the other hand, the IDs “444555A7777” and “777888C9090” are not included in the known side effect learning list or the known side effect candidate list. Therefore, the side effect determination / learning unit 108 outputs “unknown” to these two ID columns as “determination based on comparison with similar drugs”.

  In the first example, since the known side effect learning list is not empty, the selected side effect is determined as a known side effect. Therefore, no instruction from the user is required. Therefore, the side effect determination / learning unit 108 outputs “-” indicating that the determination is unnecessary to the “determination input column” column.

  Subsequently, a case where “cerebral hemorrhage” is selected as a selected side effect in step S106 will be described as a second example of step S116 in FIG. As described above, the determination target drug ID 206 is “998877F5050”, and the learning result table 204 is as shown in FIG.

Then, when the selected drug ID is “111222A3333”, the side effect keyword group of the entry of FIG. The keyword “pulmonary hemorrhage” is found. Similarly, when the selected analog ID is “444555A7777”, the keyword “pulmonary hemorrhage” is found in the side effect keyword group of the selected analog. Therefore, if the threshold values α ₁ and α ₂ are appropriately determined, IDs “111222A3333” and “444555A7777” are added to the known side effect candidate list.

On the other hand, when the selected drug ID is “777888C9090”, the selected drug ID is not added to the known side effect learning list or the known side effect candidate list. This is because the side effects keyword group entry 4 corresponding to the selected class drugs (i.e. entry described above (a4)) is contained only keywords similarity to the "cerebral hemorrhage" is less than the threshold value alpha ₂ This is because there is not.

  Therefore, in this second example, in step S116 of FIG. 3, the side effect determination / learning unit 108 sets 3 of the determination result list 402 of FIG. 5 as the determination result for the selected side effect “cerebral hemorrhage” as follows. Output the line.

  That is, the side effect determination / learning unit 108 outputs the keyword “cerebral hemorrhage” indicating the selected side effect in the “side effect” column, and outputs “unknown” in the “determination based on learning result” column. Since the ID “111222A3333” is included in the known side effect candidate list, the side effect determination / learning unit 108 sets “known candidate” in the ID column “111222A3333” as “determination by comparison with similar drugs”. Output. Similarly, the side effect determination / learning unit 108 outputs “known candidate” to the ID column “444555A7777”.

  In the second example, since the known side effect learning list is empty, the side effect determination / learning unit 108 cannot determine that “the selected side effect is a known side effect”. On the other hand, since the known side effect candidate list is not empty, the side effect determination / learning unit 108 cannot determine that “the selected side effect is an unknown side effect”.

  Therefore, the side effect determination / learning unit 108 sets “known” in the “determination input field” column in the row of the selected side effect to which attention is currently given in order to make the user determine whether the selected side effect is a known side effect. A radio button 403 for selecting “Unknown” is displayed. Of course, depending on the embodiment, a user interface other than a radio button (for example, a check box or a pull-down list) may be used as a user interface for receiving input from the user.

  As described above, as described with reference to the first and second examples, the side effect determination / learning unit 108 determines in step S116 that, for each analog, the ID of the analog is included in the known side effect learning list. “Known” is output, and if it is included in the known side effect candidate list, “Known candidate” is output. If the ID of the related drug is not included in the known side effect learning list or the known side effect candidate list, the side effect determination / learning unit 108 outputs “unknown”.

  Then, when the known side effect candidate list is not empty and the known side effect learning list is empty, the side effect determination / learning unit 108 displays a radio button 403 in the “determination input field” column. In other cases, the side effect determination / learning unit 108 can determine that “the selected side effect is a known side effect” or “the selected side effect is an unknown side effect”. 108 does not display the radio button 403.

  In step S116, the side effect determination / learning unit 108 further learns the determination result of whether or not the selected side effect is known. Specifically, the side effect determination / learning unit 108 determines whether or not the known side effect learning list is empty, and if the known side effect learning list is not empty, the entry having the determination target drug ID 206 as an ID in the learning result table 204 Add the selected side effect to the list of known side effects. When the known side effect learning list is empty, the side effect determination / learning unit 108 does not update the known side effect list in the learning result table 204.

  As described above, when the output of the selected side effect row in the determination result list 402 and the learning result table 204 are learned as necessary in step S116, the process returns to step S105.

  If it is determined in step S105 that all the side effects reported in the safety information report document 205 have been processed, then the side effect determination / learning unit 108 executes the process of step S117. That is, in step S117, the side effect determination / learning unit 108 determines whether determination by the user is necessary.

  For example, the side effect determination / learning unit 108 may store whether or not the radio button 403 is displayed once in step S116. If the radio button 403 has been displayed, the side effect determination / learning unit 108 may determine in step S117 that “the user needs to make a determination”.

  The case where determination by the user is unnecessary is a case where the side effect determination / learning unit 108 can determine that all side effects extracted from the safety information report document 205 are known or unknown. In this case, the process of FIG. 3 is also terminated. On the other hand, if the side effect determination / learning unit 108 determines that “the user needs to make a determination”, the process proceeds to step S118.

  In step S118, the side effect determination / learning unit 108 receives input from the user regarding the known side effect candidate, and learns the input content.

  Specifically, referring again to FIG. 5, the side effect determination / learning unit 108 displays the learning button 404 on the side effect determination result screen 400 in step S118, and waits until the learning button 404 is pressed. . In the example of FIG. 5, the learning button 404 is written as “learn input content about known candidate side effects”. When the learning button 404 is pressed in step S118, the side effect determination / learning unit 108 learns the content input from the user via the radio button 403.

  As described above, the side effect on which the radio button 403 is displayed is a side effect determined by the side effect determination / learning unit 108 as “a known side effect cannot be determined, but a known side effect is possible”. Therefore, when the user determines whether the side effect is known or unknown, inputs the determination result via the radio button 403, and presses the learning button 404 to confirm the input, the side effect determination / learning unit 108 In step S118, the following operation is performed.

  That is, for each side effect for which the radio button 403 is displayed, the side effect determination / learning unit 108 registers the side effect in the learning result table 204 when the radio button 403 in the side of the side effect indicates “known”. That is, the side effect determination / learning unit 108 adds the side effect indicated as “known” by the radio button 403 to the known side effect list of the entry whose ID matches the determination target drug ID 206 in the learning result table 204. When the radio button 403 instructs “unknown”, the side effect determination / learning unit 108 does nothing particularly about the side effect.

  For example, in the case of the example in FIG. 5, a radio button 403 is displayed for the keyword “cerebral hemorrhage” extracted from the safety information report document 205. Here, it is assumed that the radio button 403 indicates “known” and the learning button 404 is pressed. Then, the side effect determination / learning unit 108 adds the keyword “cerebral hemorrhage” to the known side effect list of the entry whose ID is “998877F5050” in the learning result table 204 of FIG. On the other hand, if the radio button 403 indicates “unknown” and the learning button 404 is pressed, the side effect determination / learning unit 108 knows the known side effect of the entry whose ID is “998877F5050” in the learning result table 204 of FIG. Do not update the list.

When step S118 ends as described above, the processing in FIG. 3 also ends.
Next, details of the side effect similarity calculation process executed in step S112 of FIG. 3 will be described with reference to FIGS.

  FIG. 6 is a flowchart of the side effect similarity calculation process. As described with reference to step S112 in FIG. 3, the side effect similarity calculation process is a process for obtaining a similarity for a combination of a certain selected side effect and a selected drug ID.

In step S201, the side effect determination / learning unit 108 initializes the _three variables P ₁ , P ₂ , and P ₃ to NULL. The three variables P ₁ , P ₂ , and P ₃ indicate the score of the similarity among the keywords that satisfy the certain criteria for the similarity to the selected side effect among the side effects keyword group of the selected class of drugs. It is a variable for memorizing.

In some embodiments, a specific value that is not used as a score (for example, −1) can be used as an initial value of the variables P ₁ , P ₂ , and P ₃ . The three variables P ₁ , P ₂ , and P ₃ correspond to the first, second, and third rank points, respectively.

  Then, in the next step S202, the side effect determination / learning unit 108 acquires a side effect keyword group of the selected drug. That is, the side effect determination / learning unit 108 searches the learning result table 204 for an entry whose ID matches the selected analogy drug ID, and acquires a side effect keyword group of the found entry. Then, the process proceeds to step S203.

  In step S203, the side effect determination / learning unit 108 determines whether or not an unprocessed one that has not been processed in step S204 and subsequent steps remains in the side effect keyword group acquired in step S202. If unprocessed keywords remain, the process proceeds to step S204. If all keywords have been processed, the process proceeds to step S210.

  In step S204, the side effect determination / learning unit 108 selects any one of unprocessed keywords that have not been processed in step S204 and subsequent steps from the side effect keyword group acquired in step S202. Hereinafter, the keyword selected in step S204 is referred to as “selected keyword”.

  In the next step S205, the side effect determination / learning unit 108 causes the keyword similarity evaluation unit 109 to evaluate the similarity between the selected side effect and the selected keyword, and obtains a score of the evaluation result. Details of step S205 will be described later with reference to FIGS.

  Subsequently, in step S206, the side effect determination / learning unit 108 determines whether or not the score obtained in step S205 satisfies a predetermined criterion indicating that “the selected side effect matches the selected keyword”. . The reference will be described later with reference to FIG. 8 together with the details of the score calculation process.

  When the score obtained in step S205 satisfies the criterion, the process proceeds to step S207. On the other hand, when the score obtained in step S205 does not satisfy the criterion, the process returns to step S203.

  Step S207 is executed when the selected side effect and the selected keyword may be considered to match as described above. In step S207, the side effect determination / learning unit 108 learns the selected side effect and the selected keyword as synonyms. That is, the side effect determination / learning unit 108 adds an entry in which the selected side effect and the selected keyword are paired to the synonym dictionary 203.

  Then, in the next step S208, the side effect determination / learning unit 108 determines whether the score obtained in step S205 falls within the top three points among the scores obtained so far for keywords that can be considered to match the selected side effect. Determine whether.

Specifically, if any one of the variables P _{1 to} P ₃ remains NULL in the initial state, the side effect determination / learning unit 108 determines that “the score obtained in step S205 is within the top three”. to decide. If specific values have already been set for the variables P _{1 to} P ₃ , the side effect determination / learning unit 108 determines that the score obtained in step S205 is larger than the value of the variable P ₃ (that is, the third-ranked score). , “The score obtained in step S205 is within the top three” is determined.

Conversely, if specific values have already been set for the variables P _{1 to} P ₃ and the score obtained in step S205 is equal to or smaller than the value of the variable P ₃ , the side effect determination / learning unit 108 determines that “in step S205 The score obtained is not within the top 3 ”.

  If the side effect determination / learning unit 108 determines that “the score obtained in step S205 is within the top three”, the process proceeds to step S209. Otherwise, the process returns to step S203.

In step S209, the side effect determination / learning unit 108 appropriately updates the variables P _{1 to} P ₃ in accordance with the score obtained in step S205.
Specifically, when the variable P ₁ is NULL, the side effect determination / learning unit 108 substitutes the score obtained in step S 205 for the variable P ₁ . When the variable P ₁ is not NULL and the variable P ₂ is NULL, the side effect determination / learning unit 108 substitutes the score obtained in step S205 for the variable P ₂ . When the variables P ₁ and P ₂ are not NULL and the variable P ₃ is NULL, the side effect determination / learning unit 108 substitutes the score obtained in step S205 for the variable P ₃ .

On the other hand, when specific values are set for all of the variables P _{1 to} P ₃ , the side effect determination / learning unit 108 updates the variables as follows.
That is, if the number obtained in step S205 is greater than the value of the variable _{P 1,} side effects judgment and learning unit 108 substitutes the current value of the variable _{P 2} into the variable _{P 3,} the current in the variable _{P 2} variable _{P 1} substituting values, it substitutes the number obtained in step S205 to the variable P _1. Alternatively, a number of points is equal to or less than the value of the variable _{P 1} in step S205, and if the value is greater than the variable _{P 2,} side effects judgment and learning unit 108 substitutes the current value of the variable _{P 2} into the variable _{P 3,} substituting score obtained in step S205 to the variable _{P 2.} Alternatively, a number of points is equal to or less than the value of the variable _{P 2} in step S205, and if greater than the value variable _{P 3,} side effects judgment and learning unit 108 assigns a score obtained in the variable _{P 3} at step S205.

When the updating of the variables P _{1 to} P ₃ is completed as described above, the process returns to step S203.
Furthermore, side effects judgment and learning unit 108 in step S210, by using the variable P ₁ and P ₂ and P ₃ as arguments, perform score normalization process to be described later in conjunction with FIG. 10, and calculates a score normalized. In step S211, the side effect determination / learning unit 108 returns the normalized score as a return value of the process in FIG. 6, and the process in FIG. 6 ends. That is, in step S112 of FIG. 3 corresponding to the process of FIG. 6, the side effect determination / learning unit 108 acquires the normalized score as a similarity.

  FIG. 7 is a flowchart of the score calculation process. As described with reference to step S205 in FIG. 6, the score calculation process is a process in which the keyword similarity evaluation unit 109 calculates a score indicating the similarity between two keywords. Hereinafter, in the description of FIG. 7, the two keywords designated by the keyword similarity evaluation unit 109 are referred to as “keyword A” and “keyword B”.

In step S301, the keyword similarity evaluation unit 109 assigns an initial value of 0 to a variable max indicating the similarity between the keywords A and B.
In subsequent step S302, the keyword similarity evaluation unit 109 initializes a variable a for counting the division pattern for dividing the keyword A into one or more partial character strings to 1. For example, if the keyword A is “cerebral hemorrhage” and the division part is indicated by “/”, “cerebral hemorrhage”, “brain / bleeding”, “brain / blood”, “brain / out / blood” 4 Various division patterns are possible, and the variable a is a variable for counting from 1 to 4 in order.

Then, the keyword similarity degree evaluation unit 109 at the next step S303 generates a second division pattern _{Q a} keyword A.
Subsequently, in step S304, the keyword similarity evaluation unit 109 initializes a variable b for counting the division pattern for dividing the keyword B into one or more partial character strings to 1.

Then, the keyword similarity degree evaluation unit 109 at the next step S305 generates a b-th division pattern _{Q b} keyword B.
Also, the keyword similarity degree evaluation unit 109 at the next step S306 initializes to zero a variable score for keyword indicating the similarity of A and B when using division pattern _{Q a} and _{Q b.}

Further, the keyword similarity degree evaluation unit 109 at the next step S307 initializes to 1 a variable j for sub-string of interest in the division pattern Q _a counts what number of things.
The partial string similarity evaluation unit 110 of the keyword similarity within the evaluation unit 109 in step S308 obtains the j th substring in the divided pattern _{Q a,} into a variable substr. For example, when the division pattern Q _a is “brain / bleeding”, the first partial character string in the division pattern Q _a is “brain” and the second partial character string is “bleeding”. Therefore, for example, when j = 1, the variable substr indicates a partial character string “brain”.

Subsequently, the partial character string similarity evaluation unit 110 at step S309, the in division pattern Q _b, it is determined whether there is a substring that matches as synonyms or exact match and the partial string substr.

That is, the partial character string similarity evaluation unit 110, in division pattern Q _b, investigate whether there is a substring substr exactly match subexpression. If a partial character string that completely matches the partial character string substr is found, the process proceeds to step S310.

Also, in division pattern _{Q b,} if there is no substring substr exactly match subexpression, substring similarity evaluation unit 110 refers to the synonym dictionary 203, synonymous substring substr Check if the word is registered. And if part if it is a synonym for the string substr registered partial string similarity evaluation unit 110, in division pattern Q _b, the partial character string that completely matches synonymous substring substr Find out if there is. As a result, if in the division pattern Q _b, if found substring that perfectly matches synonymous substring substr, the process proceeds to step S310.

On the other hand, when the synonym of the partial character string substr is not registered in the synonym dictionary 203, the process proceeds to step S311. Although synonym substring substr is registered in the synonym dictionary 203, even when the partial string that exactly match the corresponding synonym is not found in the division pattern Q _b, the process steps The process proceeds to S311.

  In step S310, the partial character string similarity evaluation unit 110 obtains a score f (| substr |) corresponding to the length of the partial character string substr (hereinafter, expressed as | substr |). That is, the partial character string similarity evaluation unit 110 gives an evaluation of a scoring point f (| substr |) for matching of the partial character strings substr.

  Then, the keyword similarity evaluation unit 109 adds the score f (| substr |) obtained by the partial character string similarity evaluation unit 110 to the variable score. That is, the keyword similarity evaluation unit 109 totals the evaluations by the partial character string similarity evaluation unit 110.

In this embodiment, the length of the partial character string is counted by the number of bytes. However, depending on the embodiment, the length of the partial character string may be counted by the number of characters.
Here, FIG. 8 will be referred to in order to explain the points allocated in step S310. FIG. 8 is a diagram for explaining constant values used for processing in the determination apparatus 100.

  For convenience of explanation, the scoring information 501 and the reference value information 502 are illustrated in a table format in FIG. 8, but the scoring information 501 and the reference value information 502 may be defined as constants in the program executed by the CPU 301. . Alternatively, the scoring information 501 and the reference value information 502 may be defined by a file on the storage device 307 and read by the CPU 301.

  According to the scoring information 501 in FIG. 8, for a length of 1 to 10 bytes, 1 point, 2 points, 3 points, 4 points, 5 points, 7 points, 8 points, 10 points, 11 points, A score of 12 points is defined. Scoring for a length of 11 bytes or more is omitted in FIG.

Here, as described above, the length of the character string s is expressed as | s |, and the number of points assigned to the length | s | in the scoring information 501 is expressed as f (| s |). . To explain with reference to this notation, Scoring information 501 of the present embodiment is defined to satisfy equation (1) follows for any string s ₁ and s _2.
f (| s ₁ |) + f (| s ₂ |) ≦ f (| s ₁ | + | s ₂ |) (1)

For example, according to the scoring information 501, the scoring points for the 1-byte and 7-byte character strings are 1 point and 8 points, respectively, and the scoring points for the 8 (= 1 + 7) -byte character string are 10 points. And 10 points is more than the sum of 1 point and 8 points. Therefore, when | s ₁ | = 1 and | s ₂ | = 7, Expression (1) is satisfied. In other cases, according to the stipulation information 501 in FIG. 8, the expression (1) is satisfied.

  According to the scoring information 501 that satisfies Expression (1), the partial character string similarity evaluation unit 110 evaluates the partial character string as follows. That is, the partial character string similarity evaluation unit 110 determines that the partial character strings having the first length are equal to each other when the partial character strings having the third length obtained by adding the first and second lengths match each other. A higher evaluation than the evaluation obtained by adding the evaluation when the character strings match and the evaluation when the partial character strings of the second length match is given.

Note that the significance of equation (1) and the reference value information 502 in FIG. 8 will be described again after the description of the score calculation processing in FIG. 7, and here, the description returns to FIG. 7.
Keyword similarity evaluation unit 109 in step S311 determines whether or not finished examined until the end of the division pattern Q _a. That is, the keyword similarity evaluation unit 109 for the last substring in the divided pattern Q _a is determined whether finished the determination in step S309.

And, if not finished examining still until the end of the division pattern Q _a, the process proceeds to step S312. On the other hand, if finished it examined until the end of the divided pattern Q _a, the process proceeds to step S313.

Step keyword similarity evaluation unit 109 in S312, in order to focus within the division pattern Q _a in the next substring, the value of the variable j is incremented by one. Then, the process returns to step S308.

In step S313, the keyword similarity evaluation unit 109 determines whether the value of the variable score exceeds the value of the variable max.
If the value of the variable score is greater than the value of the variable max, the degree of similarity calculated according to the combination of the current division pattern Q _a and Q _b, than the similarity calculated in accordance with a combination of any division pattern ever high. Therefore, in this case, to store the highest similarity obtained from the combination of the current division pattern Q _a and Q _b, the process proceeds to step S314. On the other hand, if the value of the variable score is less than or equal to the value of the variable max, the process proceeds to step S315.

In step S314, the keyword similarity evaluation unit 109 substitutes the value of the variable score for the variable max. Then, the process proceeds to step S315.
In step S315, the keyword similarity evaluation unit 109 increases the value of the variable b by one. Then, the process proceeds to step S316.

In step S < _b > 316, the keyword similarity evaluation unit 109 determines whether the similarity calculation for all the divided patterns of the keyword B has been completed for the divided pattern Q _a currently focused on the keyword A. Specifically, if the number of characters of a certain character string s is expressed as c (s), there are 2 ^{c (B) -1} division patterns of the keyword B, so the keyword similarity evaluation unit 109 It is determined whether or not the value of b exceeds ^{2c (B) -1} .

If the value of the variable b exceeds ^{2c (B) -1} , the process proceeds to step S317 in order to examine the next division pattern for the keyword A. On the other hand, if the value of the variable b is 2 ^{c (B) -1} or less, the degree of similarity is calculated for the divided pattern Q _a currently focused on the keyword A by the next divided pattern of the keyword B. Therefore, the process returns to step S305.

In step S317, the keyword similarity evaluation unit 109 increases the value of the variable a by 1. Then, the process proceeds to step S318.
In step S <b> 318, the keyword similarity evaluation unit 109 determines whether the similarity calculation for all the divided patterns of the keyword A has been completed. Specifically, since there are ^{2c (A) -1} division patterns of keyword A, the keyword similarity evaluation unit 109 determines whether the value of the variable a exceeds ^{2c (A) -1.} To do.

If the value of the variable a exceeds ^{2c (A) -1} , the process proceeds to step S319. On the other hand, if the value of the variable a is 2 ^{c (A) −1} or less, the process returns to step S303 in order to calculate the similarity according to the next division pattern of the keyword A.

In step S319, the keyword similarity evaluation unit 109 returns the value of the variable max as a score indicating the similarity between the keywords A and B.
In the above description regarding FIG. 7, it is not limited which number of ^{2c (A) −1} divided patterns for keyword A is counted. That is, the keyword similarity evaluation unit 109 can order ^{2c (A) -1} divided patterns according to an arbitrary order. Similarly, the ^{2c (B) -1} divided patterns related to the keyword B can be ordered according to an arbitrary order by the keyword similarity evaluation unit 109.

  As described above, the score calculated from the keywords A and B in the score calculation process of FIG. 7 is the highest score obtained from an arbitrary combination of the divided patterns of the keywords A and B. And the score obtained by the process of FIG. 7 is obtained by the cumulative addition of the score defined by the score information 501 so that the said Formula (1) may be satisfy | filled.

  From the above, “the evaluation by the keyword similarity evaluation unit 109 gives a reasonable score if keywords A and B do not completely match but there is a match in a short partial character string”. Can be said. At the same time, the evaluation by the keyword similarity evaluation unit 109 also evaluates higher as the partial character string that matches between the keywords A and B becomes longer.

Therefore, in order to explain the evaluation by the keyword similarity evaluation unit 109 in more detail, the above formula (1) is taken up again.
The left side of the expression (1) indicates that when a partial character string s ₁ matches between two keywords, and a partial character string s ₂ also matches between two keywords, the partial character strings s ₁ and s ₂ respectively Indicates the number of points added due to the match. Further, the right side of the expression (1) is a character string (when the character string obtained by concatenating the partial character strings s ₁ and s ₂ (referred to as “s ₁ · s ₂ ” for convenience) matches between two keywords ( The number of points added due to the coincidence of s ₁ · s ₂ is shown. Thus, equation (1), the partial strings s ₁ and s ₂ is the score or scores in the case of matching between each away keyword substrings s ₁ and s ₂ the connecting string (s ₁ S ₂ ) Indicates that the keyword is assigned when the entire keyword matches.

Further, when the following expression (2) is satisfied, expression (3) is satisfied from expression (1). Therefore, f (| s ₁ |) is added to both sides of expression (3) to obtain expression (4). It is done. Since the right side of Expression (4) is less than f (| s ₁ | + | s ₂ |) from Expression (1), Expression (5) is obtained from Expression (2).
| S ₂ | = | s ₃ | + | s ₄ | (2)
f (| s ₃ |) + f (| s ₄ |) ≦ f (| s ₂ |) (3)
f (| s ₁ |) + f (| s ₃ |) + f (| s ₄ |)
≦ f (| s ₁ |) + f (| s ₂ |) (4)
f (| s ₁ |) + f (| s ₃ |) + f (| s ₄ |)
≦ f (| s ₁ | + | s ₃ | + | s ₄ |) (5)
The expression (5) indicates that the partial character strings s ₁ , s _3, and s ₄ are score points equal to or higher than the points when the partial character strings s ₁ , s _3, and s ₄ match between the keywords at three separate locations. This indicates that the entire continuous character string (s ₁ s ₃ s ₄ ) is assigned when the keywords match.

  Similarly, by deriving the equation, the following holds for any two or more Rs. That is, according to the scoring information 501, a score that is equal to or higher than the score when the R partial character strings are matched between the keywords at the distant R locations is that the entire character string obtained by connecting the R partial character strings is between the keywords. It is given when they match.

Therefore, according to the scoring information 501 defined so as to satisfy the formula (1), the highest score that can be obtained by the score calculation process of FIG. Any of (b1) to (b3).
(B1) When the lengths of the keywords A and B are the same, the score assigned to the length | A | (that is, the length | B |) in the score information 501.
(B2) When the keyword A is shorter than the keyword B, the score assigned to the length | A | in the score information 501.
(B3) When the keyword B is shorter than the keyword A, the score associated with the length | B | in the score information 501.

  Therefore, whether or not the score calculated by the score calculation process of FIG. 7 for a certain keyword A and B satisfies a criterion that can be regarded as “keywords A and B match” is keyword A, B or its It is reasonable to judge according to both lengths. Therefore, as shown in FIG. 8, in the determination apparatus 100 according to the present embodiment, a value obtained by multiplying the maximum score that can be taken according to the length (the number of bytes in the present embodiment) by a constant β less than 1 is used as the length. Reference value information 502 defined as a reference value corresponding to is used.

  Specifically, in step S206 of FIG. 6, a reference value associated with the length of the selected side effect in the reference value information 502 is used as a reference that may be considered that the selected side effect matches the selected keyword. That is, in step S206, the side effect determination / learning unit 108 determines whether or not the score calculated by the keyword similarity evaluation unit 109 in step S205 is greater than or equal to the reference value associated with the length of the selected side effect in the reference value information 502. Determine whether. If the score is equal to or greater than the reference value, the side effect determination / learning unit 108 determines that the criterion is satisfied, and the process proceeds to step S207.

  When the selected keyword is shorter than the selected side effect in step S206 in FIG. 6, the score calculated in step S205 is “the highest possible value” associated with the length of the selected side effect in the reference value information 502 in FIG. It is clear that it is less than "score". However, in the process of FIG. 6, even if the selected keyword is shorter than the selected side effect, in step S206, the reference value associated with the length of the selected side effect is used in step S206 as described above. This is because the process shown in FIG. 6 has a viewpoint on the side of the selected side effect, and the process shown in FIG. 6 determines whether or not a side effect having a high similarity to the selected side effect is included in the side effect keyword group of the selected drug. Because it is aimed.

Note that the scoring information 501 and the reference value information 502 in FIG. 8 are also used by the similar medicine processing unit 103, and the usage by the similar medicine processing unit 103 will be described later with reference to FIGS.
Further, the score satisfying the formula (1) may be selected and set based on human knowledge. For example, in the present embodiment, “for lengths of 1 to 5 bytes, a value obtained by multiplying the length by weight 1 is used as a score, and for lengths of 6 to 10 bytes, the length is weighted 1.2. Scoring information 501 is set in accordance with the policy of “multiplying and rounding off the decimal point to be a score”.

Alternatively, using an appropriate positive constant C, for example, by defining a score f (| s |) as shown in Equation (6) for an arbitrary length | s |, Equation (1) is satisfied. It can also be made.
f (| s |) = C | s | ² (6)
The expression (6) is obtained by setting the score per unit length to a value proportional to the character string length | s | (that is, C | s |). s | ² points are given. Note that, not limited to the example of Expression (6), generally, a score defined such that the score per unit length monotonously increases with respect to the character string length satisfies Expression (1).

Next, a specific example will be given with reference to FIG. 9 in order to facilitate understanding of the score calculation processing described with reference to FIGS.
FIG. 9 is a diagram schematically illustrating a specific example of the score calculation process. The keyword A in FIG. 7 corresponds to the keyword 601 “general anesthetic” in FIG. 9, and the keyword B in FIG. 7 corresponds to the keyword 602 “general inhalation anesthetic” in FIG.

Since the keyword 601 is five characters, there are 16 (= 2 ^5-1 ) division patterns of the keyword 601. In FIG. 9, as an example, a division pattern 603a “general anesthesia / agent”, a division pattern 603b “general anesthetic”, and a division pattern 603c “general anesthesia / agent” are illustrated. The other 13 division patterns are omitted in FIG.

Since the keyword 602 has seven characters, the keyword 602 has 64 (= 2 ^7-1 ) division patterns. For example, FIG. 9 shows a division pattern 604a of “whole body / inhalation / anesthetic / agent”, a division pattern 604b of “whole body inhalation / anesthetic / agent”, and a division pattern 604c of “whole body / inhalation anesthesia / agent”. It is shown. The other 61 division patterns are omitted in FIG.

  Further, in FIG. 9, below the rectangle indicating each partial character string in each divided pattern, the score corresponding to the length of the partial character string in the score information 501 in FIG. It is illustrated.

  According to the score calculation process of FIG. 8, the score is calculated for all combinations of the 16 division patterns of the keyword 601 and the 64 division patterns of the keyword 602, and the highest score is obtained from the keywords 601 and 602. It is obtained as a score indicating the degree of similarity.

For example, regarding the combination of the division patterns 603a and 604a, the score is calculated as follows in FIG.
That is, in step S309 when j = 1, a partial character string that matches the first partial character string “whole body” in the divided pattern 603a is found in the divided pattern 604a. Therefore, in step S310, 4 points corresponding to the length of the “whole body” (that is, 4 bytes) are added in the scoring information 501 of FIG.

  In step S309 when j = 2, a partial character string that matches the second partial character string “anesthetic” in the divided pattern 603a is found in the divided pattern 604a. Therefore, in step S310, 4 points corresponding to the length of “anesthetic” (that is, 4 bytes) are added.

  In step S309 when j = 3, a partial character string that matches the third partial character string “agent” in the divided pattern 603a is found in the divided pattern 604a. In step S310, two points corresponding to the length of the “agent” (that is, 2 bytes) are added.

  As a result, the number of points obtained for the combination of the divided patterns 603a and 604a (that is, the value of the variable score when executing step S313) is 10 (= 4 + 4 + 2).

  Similarly, the number of points obtained for the combination of the division patterns 603a and 604b is a total of 6 (= 4 + 2) points added due to the match between the partial character strings “anesthetic” and “agent”. Further, the number of points obtained for the combination of the divided patterns 603a and 604c is a total of 6 (= 4 + 2) points added due to the match between the partial character strings “whole body” and “agent”.

  In the case of the keywords 601 and 602 in FIG. 9, 11 (= 4 + 7) obtained from a combination of the division pattern “whole body / anesthetic agent” and the division pattern “whole body / inhalation / anesthetic agent” not shown. The point is the highest score. Therefore, the score returned by the keyword similarity evaluation unit 109 as a result of the score calculation process of FIG. 7 for the keywords 601 and 602 is 11 points.

  The score calculation process in step S205 of FIG. 6 has been described in detail above with reference to FIGS. Next, details of the point normalization process in step S210 of FIG. 6 will be described.

FIG. 10 is a flowchart of the score normalization process.
Side effects judgment and learning unit 108 in step S401, either specific value of P ₁ is given as the first argument, or is the first argument to determine whether it is NULL. If the first argument is NULL, the process proceeds to step S402. On the other hand, if it be a specific value of P ₁ is given as the first argument, the process proceeds to step S403.

  In step S402, the side effect determination / learning unit 108 returns 0. Then, the score normalization process ends. Note that step S402 is a step when there is no argument that is specifically given a value.

  For example, in the process of FIG. 6, if it is determined in step S206 that no selected keyword satisfies the matching criteria, step S402 of FIG. 10 is executed. Therefore, the normalized score is 0 points.

Side effects judgment and learning unit 108 in step S403, either specific value of P ₂ is given as the second argument, or is a second argument to determine whether it is NULL. If the second argument is NULL, the process proceeds to step S404. On the other hand, if it be a specific value of P ₂ is given as the second argument, the process proceeds to step S405.

Step S404 is a step for a case where a value is given as one argument. In this case, side effects judgment and learning unit 108 returns the one value (i.e. the value of P ₁₎ itself, as a normalized value. Then, the score normalization process ends.

  For example, in the process of FIG. 6, if it is determined that only one selected keyword satisfies the matching criterion in step S206, step S404 of FIG. 10 is executed. Therefore, the normalized score is the score itself obtained for the one selected keyword.

Furthermore, side effects judgment and learning unit 108 in step S405, either a specific value of P ₃ is given as the third argument, or a third argument to determine whether it is NULL. If the third argument is NULL, the process proceeds to step S406. On the other hand, if it be a specific value of P ₃ is given as the third argument, the process proceeds to step S407.

Step S406 is a step for the case where only two values are given as arguments. In this case, side effects judgment and learning unit 108, the value of the two values (i.e. values of P ₁ and P ₂₎ and normalized calculated according to the following formula (7), and returns the calculated value. Then, the score normalization process ends. Incidentally, the left side of the P _R of formula (7) and below (8) shows the normalized scores.

On the other hand, step S407 is a step for the case where values are specifically specified for all three arguments. In this case, side effects judgment and learning unit 108, the three values (i.e. values of P ₁ and P ₂ and P ₃₎ a value obtained by normalizing the calculated according to the following equation (8), and returns the calculated value. Then, the score normalization process ends.

  In the processing by the side effect processing unit 102 described with reference to FIGS. 1 to 10 as described above, the learning result table 204 is used. In addition, as described above, some fields of the learning result table 204 are learned in advance by preprocessing in the present embodiment. Therefore, the preprocessing will be described below.

  Prior to the start of normal operation of the determination apparatus 100 in FIG. 1, for example, a series of preprocessing such as the following (c1) to (c4) is performed. In addition, when the attached document 202 is updated or added after the normal operation of the determination apparatus 100 is started, the following pre-processing (c5) is performed.

(C1) Processing for storing the attached document group 201 in the storage unit 101.
(C2) If there is data in the synonym dictionary 203 that can be used, the synonym dictionary 203 having the data is stored in the storage unit 101. If there is no data in the synonym dictionary 203 that can be used, there is no entry. A process of creating a synonym dictionary 203 in an initial state on the storage unit 101.
(C3) Processing for creating an initial learning result table 204 having no entry on the storage unit 101.
(C4) Processing for adding an entry to the learning result table 204 and learning for fields other than the known side effect list according to the flowchart of FIG.
(C5) A process of updating or adding the attached document 202 according to the flowchart of FIG. 14 and learning the learning result table 204 in accordance with the update or addition.

  Note that the preprocessing (c1) to (c3) may be performed by, for example, a system administrator, or may be performed by the preprocessing control unit 113. Hereinafter, the processes (c4) and (c5) performed in accordance with the control by the preprocessing control unit 113 will be described in detail with reference to FIGS.

FIG. 11 is a flowchart of the preprocessing (c4) performed by the determination apparatus 100 before the start of normal operation.
In step S501, the preprocessing control unit 113 obtains the number of medicines in which the attached document 202 is registered by counting the number of files of the attached document 202, for example, and stores the obtained number in a variable N.

  In the next step S502, the preprocessing control unit 113 initializes the value of the variable i for paying attention in order to the medicines for which the attached document 202 is registered to 1. In the following, for simplification of description, the i-th medicine package 202 is simply referred to as “i-th package 202”.

Subsequently, in step S <b> 503, the preprocessing control unit 113 adds an entry for the i-th medicine to the learning result table 204.
As described with reference to FIG. 1, in the present embodiment, the file name of the attached document 202 includes the ID of the medicine, so that the attached document 202 is associated with the ID of the medicine. Therefore, the preprocessing control unit 113 selects the i-th attached document 202 from the attached document group 201 and recognizes the ID corresponding to the attached document 202 from the file name of the selected attached document 202. Then, the preprocessing control unit 113 sets the recognized ID in the “ID” field and adds an entry in which other fields are initialized to be empty to the learning result table 204.

  In the next step S504, the preprocessing control unit 113 extracts the keyword from the “efficacy or effect” section of the i-th attached document 202 and registers it in the learning result table 204 with respect to the effect / effect keyword extraction unit 111. Order to do. Then, the efficacy / effect keyword extraction unit 111 performs keyword extraction and registration in the learning result table 204 according to the command.

  Similar to the keyword extraction by the side effect keyword extraction unit 107 in step S104 of FIG. 3, the algorithm used by the efficacy / effect keyword extraction unit 111 for keyword extraction may vary depending on the embodiment.

  For example, the efficacy / effect keyword extraction unit 111 may perform morphological analysis on the “efficacy or effect” section of the i-th attached document 202 and extract one or more series of nouns as keywords. The effect / effect keyword extraction unit 111 may further perform a syntax analysis on the result of the morphological analysis and extract a keyword using the result of the syntax analysis.

  Alternatively, the efficacy / effect keyword extraction unit 111 may perform a simple keyword extraction process based on the character type, such as extracting kanji, katakana or a sequence of English letters as a keyword. In addition, if medical dictionary data that can be described as an effect or effect is available, the effect / effect keyword extraction unit 111 reads the dictionary entry from the “effect or effect” section of the i-th attached document 202. A matching character string may be extracted as a keyword.

  In any case, in step S504, the effect / effect keyword extraction unit 111 extracts keywords from the “effect or effect” section of the i-th attached document 202 according to an appropriate algorithm according to the embodiment. Then, the efficacy / effect keyword extraction unit 111 registers the extracted keyword in the field of the efficacy / effect keyword group of the entry of the learning result table 204 added in step S503 corresponding to the i-th medicine. The pre-processing control unit 113 notifies the efficacy / effect keyword extraction unit 111 of the ID of the i-th medicine, so that the efficacy / effect keyword extraction unit 111 can add the attached document 202 and the keyword registration destination of the keyword extraction target. Can be recognized.

  In the next step S505, the preprocessing control unit 113 instructs the side effect keyword extraction unit 107 to extract a keyword from the “side effect” section of the i-th attached document 202 and register it in the learning result table 204. . Then, the side effect keyword extraction unit 107 performs keyword extraction and registration in the learning result table 204 according to the command.

  Note that the keyword extraction algorithm in step S505 may be varied depending on the embodiment, as in step S104 in FIG. Further, the pre-processing control unit 113 notifies the side effect keyword extraction unit 107 of the ID of the i-th medicine, so that the side effect keyword extraction unit 107 recognizes the attached document 202 to be subjected to keyword extraction and the entry of the keyword registration destination. can do.

  In the next step S506, the preprocessing control unit 113 increases the value of the variable i by one. Then, the process proceeds to step S507. Note that the order of steps S504 and S505 may be reversed.

  In step S507, the preprocessing control unit 113 determines whether the value of the variable i exceeds the value of the variable N. If the value of the variable i exceeds the value of the variable N, the processing of steps S503 to S505 has been completed for all the attached documents 202 included in the attached document group 201, and the processing moves to step S508. On the contrary, if the value of the variable i is equal to or less than the value of the variable N, since the attached document 202 that has not been subjected to the processes of steps S503 to S505 remains, the process returns to step S503.

In step S508, the preprocessing control unit 113 initializes the value of the variable i to 1 again. Then, the process proceeds to step S509.
In step S509, the preprocessing control unit 113 instructs the similar drug determination / learning unit 112 to learn the similar drug of the i-th drug, and the similar drug determination / learning unit 112 determines the similar drug of the i-th drug. To learn.

  In the present embodiment, the relationship between the analogs is symmetric. That is, for any i and j, if the j-th drug is an i-th drug analog, the i-th drug is a j-th drug analog. Therefore, in step S509, the preprocessing control unit 113 designates the ID of the i-th drug that is the target of learning the similar drug to the similar drug determination / learning unit 112, and the (i + 1) th to N-th drug Is designated as a comparison range for examining whether or not it is an analog of the i-th drug.

Details of the operation of the analogy drug determination / learning unit 112 in step S509 will be described later with reference to FIG. 12, but the outline is as follows.
In other words, the similar drug determination / learning unit 112 determines whether each drug included in the comparison range designated by the preprocessing control unit 113 is similar to the i-th drug. Then, if the i-th drug similar drug is found, the similar drug determination / learning unit 112 adds the ID of the found similar drug to the similar drug list of the entry of the learning result table 204 corresponding to the i-th drug. . Further, the similar drug determination / learning unit 112 adds the ID of the i-th drug to the similar drug list of the entry of the learning result table 204 corresponding to the found similar drug.

Thereafter, in step S510, the preprocessing control unit 113 increases the value of the variable i by one. Then, the process proceeds to step S511.
In step S511, the preprocessing control unit 113 determines whether the value of the variable i is greater than or equal to the value of the variable N. If the value of the variable i is equal to or greater than the value of the variable N, it means that learning of similar drugs has been completed for all the drugs for which the attached document 202 is registered, so the preprocessing of FIG. On the other hand, if the value of variable i is less than the value of variable N, the process returns to step S509.

  FIG. 12 is a flowchart of the similar drug learning process performed by the similar drug determination / learning unit 112 in step S509 in FIG. 11 and step S808 in FIG. In the present embodiment, the pre-processing control unit 113 tells the similar drug determination / learning unit 112 the ID (hereinafter referred to as “learning target drug ID”) of the target drug for learning similar drugs (hereinafter referred to as “learning target drug”) ) Is specified. In addition, the preprocessing control unit 113 also designates the lower limit and upper limit of the comparison range for comparing whether or not the learning target drug is similar to the similar drug determination / learning unit 112.

  In step S <b> 601, the similar drug determination / learning unit 112 reads the learned content of the learning target drug and stores it as a “similar drug learning list”. That is, the similar drug determination / learning unit 112 searches the learning result table 204 for an entry having the learning target drug ID designated by the preprocessing control unit 113 as an ID, and reads the similar drug list of the entries found as a result of the search. , Remember as a similar medicine learning list.

In step S602, the similar drug determination / learning unit 112 initializes the “drug candidate list” to be empty.
In step S <b> 603, the similar medicine determination / learning unit 112 determines whether or not an unprocessed medicine remains within the comparison range designated by the preprocessing control unit 113. If there remains a medicine that has not yet been determined whether it is a similar drug to be learned within the specified range, the process proceeds to step S604. On the other hand, if it is already determined whether or not all the medicines in the designated range are similar medicines to be learned, the process proceeds to step S617.

  In step S <b> 604, the similar medicine determination / learning unit 112 selects one unprocessed medicine within the comparison range designated by the preprocessing control unit 113. Hereinafter, the drug selected in step S604 is referred to as “selected drug”, and its ID is referred to as “selected drug ID”.

  In step S605, the similar drug determination / learning unit 112 determines whether the selected drug ID and the learning target drug ID are equal. When the selected drug ID and the learning target drug ID are equal, there is no need to check whether or not the learning target drug itself is an analog of the learning target drug, and the process returns to step S603. Conversely, if the selected drug ID and the learning target drug ID are different, the process proceeds to step S606.

In step S606, the similar medicine determination / learning unit 112 determines whether or not the selected medicine ID is included in the similar medicine learning list.
When the selected drug ID is included in the similar drug learning list, the selected drug has already been learned as the similar drug of the learning target drug and has already been registered in the learning result table 204, so it is necessary to perform further processing on the selected drug There is no. Therefore, the process returns to step S603. Conversely, if the selected drug ID is not included in the analogy medicine learning list, the process proceeds to step S607.

  In step S <b> 607, the similar drug determination / learning unit 112 refers to the “medicinal effect classification name” section in the attached document 202 of the learning target drug and the selected drug, and compares them. As described above, since one or more names such as “antipyretic analgesic / anti-inflammatory agent” are described in the “medicinal effect classification name” section, the similar drug determination / learning unit 112 matches the learning target drug with the selected drug. Check if there is a medicinal classification name.

  In step S608, the analogy drug determination / learning unit 112 determines whether there is a medicinal effect classification name that matches the learning target drug and the selected drug as a result of the comparison in step S607. When there is a name of a medicinal effect classification that matches the learning target drug and the selected drug, the learning target drug and the selected drug are similar drugs. Therefore, the process proceeds to step S609 in order to learn the selected drug as an analog of the learning target drug. Conversely, if there is no matching drug class name for the target drug and the selected drug, the process is performed to further investigate whether the selected drug is an analog of the target drug based on the descriptions in other sections. Proceeds to step S610.

In step S609, the similar drug determination / learning unit 112 adds the selected drug ID to the similar drug learning list. Then, the process returns to step S603.
In step S610, the similar drug determination / learning unit 112 refers to and compares the “physical and chemical knowledge regarding active ingredients” section in the package insert 202 of each of the learning target drug and the selected drug.

  As mentioned above, the “physicochemical findings regarding active ingredients” section includes generic names, chemical names, molecular formulas and structural formulas. Moreover, in the case of a pharmaceutical containing a plurality of active ingredients, the general name, chemical name, molecular formula and structural formula of each of the plurality of active ingredients are described.

  Therefore, in this embodiment, the similar drug determination / learning unit 112 extracts a common name and a chemical name from the “physical and chemical knowledge about active ingredients” section in the package insert 202 of each of the learning target drug and the selected drug. Compare chemical names with each other.

  In step S611, the similar drug determination / learning unit 112 determines whether there is a common name or chemical name that matches between the learning target drug and the selected drug. If there is a common name or chemical name between the learning target drug and the selected drug, the learning target drug and the selected drug are similar drugs. Therefore, the process proceeds to step S612 in order to learn the selected drug as an analog of the learning target drug. On the other hand, if no match is found between the learning target drug and the selected drug, the generic name and the chemical name are not found. In order to investigate, the process proceeds to step S613.

  In the embodiment in which the structural formula is described in a predetermined markup language in the attached document 202, the analogy medicine determination / learning unit 112 may further compare the structural formulas in step S610. Then, in step S611, the similar drug determination / learning unit 112 determines that the selected drug is the similar drug of the learning target drug if the structural formulas match between the learning target drug and the selected drug, and then performs step S612. May be executed.

In step S612, the similar drug determination / learning unit 112 adds the selected drug ID to the similar drug learning list. Then, the process returns to step S603.
In step S613, the analogy drug determination / learning unit 112 performs a process shown in FIG. 13 as “efficacy / effect similarity calculation process”. That is, the similar drug determination / learning unit 112 has already learned from the “efficacy or effect” section of the attached document 202 for each of the learning target drug and the selected drug, and the similarity between the effect / effect keyword groups registered in the learning result table 204. Ask for.

  The details of step S613 will be described later with reference to FIG. 13, but in step S613, an algorithm different from general document similarity calculation using a vector space model using term frequency-inverse document frequency (TF / IDF) values is used. Is used. That is, in the process of FIG. 13, an algorithm is adopted in accordance with the purpose of determining whether two medicines are similar or not based on efficacy or effect. According to the processing of FIG. 13, in medical terms, there are a large number of compound words including common words, and even if the efficacy or overall effect is not similar between the two medicines, a part of them may match. For example, it is considered that two drugs are highly likely to be similar drugs. As a result, according to the process of FIG. 13, a similarity suitable for determination of an analog is obtained.

Then, in the next step S614, the analogy drug determination / learning unit 112 has the similarity calculated in step S613 in any range of “γ ₁ or more”, “γ ₂ or more and less than γ ₁ ”, and “less than γ ₂ ”. Determine if it applies. In the present embodiment, γ ₁ and γ ₂ are predetermined threshold values, and γ ₁ > γ ₂ .

Note that the threshold values γ ₁ and γ ₂ in the present embodiment are fixed values. However, in some embodiments, the thresholds γ ₁ and γ ₂ change according to the number or length of keywords included in the efficacy / effect keyword group learned in the learning result table 204 for the learning target drug and the selected drug. It may be a value determined in this way.

The threshold value γ ₁ is a reference value indicating that it is appropriate to determine that “the learning target drug and the selected drug are similar drugs”. In addition, the threshold γ ₂ indicates a boundary for determining that “the learning target drug and the selected drug are not similar drugs” or “the learning target drug and the selected drug may be similar drugs”. This is the reference value.

When the similarity is _one or more γ obtained in step S613, the process proceeds from step S614 to step S615. Also, similarity obtained in step S613 is a case of _two or more gamma less than ₁ gamma, the process proceeds from step S614 to step S616. When the similarity obtained in step S613 is less than gamma _2, the process returns from step S614 to step S603.

  In step S615, the similar drug determination / learning unit 112 adds the selected drug ID to the similar drug learning list. That is, the similar drug determination / learning unit 112 stores the selected drug as the similar drug of the learning target drug. Then, the process returns to step S603.

  In step S616, the similar drug determination / learning unit 112 adds the selected drug ID to the similar drug candidate list. That is, the similar drug determination / learning unit 112 stores the selected drug as a drug that may be a similar drug to be learned. Then, the process returns to step S603.

  As described above, the similar drug determination / learning unit 112 executes the process from step S604 onward for each drug included in the comparison range specified by the preprocessing control unit 113, whereby the similar drug related to the learning target drug. The learning list and candidate list are completed. When the similar drug learning list and the similar drug candidate list are completed, the process proceeds from step S603 to step S617 as described above.

  In step S617, the similar drug determination / learning unit 112 records the contents of the similar drug learning list in the learning result table 204. Specifically, the similar drug determination / learning unit 112 adds each ID included in the similar drug learning list to the similar drug list of the entry having the learning target drug ID as an ID in the learning result table 204. Furthermore, the similar medicine determination / learning unit 112 adds, for each ID included in the similar medicine learning list, a learning target drug ID to the similar medicine list of an entry having the ID as an ID in the learning result table 204.

In step S618, the similar drug determination / learning unit 112 determines whether the similar drug candidate list is empty.
When the similar drug candidate list is empty, there is no drug for which it is unknown whether it is a similar drug to be learned. Therefore, in this case, the similar medicine learning process of FIG. 12 is also terminated.

  On the other hand, if the similar drug candidate list is not empty, it is a case where a drug that cannot be determined as a similar drug as a learning target drug but that may be a similar drug as a learning target drug is found. Therefore, in this case, in order to accept a determination input by the user, the process proceeds to step S619.

  In step S619, the similar drug determination / learning unit 112 receives an input from the user as to whether or not each drug whose ID is included in the similar drug candidate list is a similar drug to be learned. For example, the similar medicine determination / learning unit 112 may display the following (d1) to (d4) on a display corresponding to the output device 306 of FIG.

  (D1) Learning target drug ID. Note that the similar medicine determination / learning unit 112 may embed a link to the attached document 202 of the learning target drug in the learning target drug ID. Further, the similar medicine determination / learning unit 112 may display the sales name of the learning target drug on the display instead of the learning target drug ID (or together with the learning target drug ID) as information for specifying the learning target drug. .

  (D2) Each ID included in the drug candidate list. The similar medicine determination / learning unit 112 may embed a link to the attached document 202 corresponding to the ID in each ID. Further, the similar medicine determination / learning unit 112 may display the sales name of the medicine on the display instead of the ID (or together with the ID) as information specifying each medicine whose ID is included in the similar medicine candidate list. Good.

  (D3) A user interface for designating, for each ID included in the similar drug candidate list, whether the drug of the ID is a similar drug to be learned. More specifically, radio buttons, check boxes, pull-down lists, etc. can be used.

(D4) A user interface such as a button for confirming the input content.
For example, when the user interfaces such as (d1) to (d4) described above are employed, the analogy medicine determination / learning unit 112 waits until the button (d4) is pressed. When the button (d4) is pressed, the similar drug determination / learning unit 112 captures the determination result input via the user interface (d3) for each ID included in the similar drug candidate list.

  In step S620, the similar drug determination / learning unit 112 records the ID of the drug determined as the similar drug to be learned in the learning result table 204 according to the input result received in step S619.

  That is, the similar drug determination / learning unit 112 indicates that the ID included in the similar drug candidate list indicates that the input content acquired through the user interface (d3) is “similar drug”. Recognized as “drug ID”. It should be noted that any ID included in the similar drug candidate list may not be recognized as the similar drug ID, only one ID may be recognized as the similar drug ID, or a plurality of IDs may be recognized as the similar drug ID. unknown.

  When one or more similar drug IDs are recognized, the similar drug determination / learning unit 112 adds the similar drug ID to the similar drug list of the entry having the learning target drug ID as an ID in the learning result table 204 for each similar drug ID. Add Furthermore, the similar drug determination / learning unit 112 adds the learning target drug ID to the similar drug list of the entry having the similar drug ID as an ID in the learning result table 204 for each similar drug ID.

  Note that the analogy medicine determination / learning unit 112 particularly processes the IDs included in the analogy drug candidate list whose input content captured through the user interface (d3) indicates “not an analogy”. Do not do. Then, when the similar drug determination / learning unit 112 finishes the process of step S620 as described above, the similar drug learning process of FIG. 12 is also ended.

  FIG. 13 is a flowchart of the efficacy / effect similarity calculation process performed by the similar drug determination / learning unit 112 in step S613 of FIG. As described above, the efficacy / effect similarity calculation process is performed for each combination of the learning target drug specified by the preprocessing control unit 113 and the selected drug selected by the similar drug determination / learning unit 112 in step S604 of FIG. Done.

In step S701, the analogy drug determination / learning unit 112 initializes the _three variables P ₁ , P ₂ , and P ₃ to NULL. The three variables P ₁ , P ₂ , and P ₃ are similar to each other in terms of the similarity between the keywords among the combinations of the keywords in the efficacy / effect keyword group of the learning target drug and the selected drug. It is a variable for storing the score of the similarity of the one whose degree is in the top three.

In some embodiments, a specific value that is not used as a score (for example, −1) can be used as an initial value of the variables P ₁ , P ₂ , and P ₃ . The three variables P ₁ , P ₂ , and P ₃ correspond to the first, second, and third rank points, respectively.

Then, similar drugs determination and learning unit 112 in step S702 acquires Indications keywords _{K 1} of the learning control drug. That is, the similar medicine determination / learning unit 112 searches the learning result table 204 for an entry whose ID matches the learning target drug ID, and acquires the efficacy / effect keyword group of the found entry. Then, the process proceeds to step S703.

Similar drugs determination and learning unit 112 in step S703 acquires Indications keywords _{K 2} selective drugs. That is, the similar drug determination / learning unit 112 searches the learning result table 204 for an entry whose ID matches the selected drug ID, and acquires the efficacy / effect keyword group of the found entry. Then, the process proceeds to step S704. Note that the execution order of steps S702 and S703 may be reversed.

Step S704 In similar drugs determination and learning unit 112, in combination with the keywords of the keyword and Indications keywords K ₂ of Indications keyword group K _1, untreated not subjected to the step S705 and subsequent steps Determine if there are any leftovers. If an unprocessed combination remains, the process proceeds to step S705. If all combinations have been processed, the process proceeds to step S711.

Step S705 In similar drugs determination and learning unit 112, of the combination of the keyword of the keyword and Indications keywords K ₂ of Indications keyword group K _1, follow the one of the combinations of raw, to select a keyword _{W 1} and _{W 2.} That is, the similar medicine determination / learning unit 112 selects the keyword W ₁ from the efficacy / effect keyword group K ₁ and selects the keyword W ₂ from the efficacy / effect keyword group K ₂ .

Then, similar drugs determination and learning unit 112 at the next step S706 causes the evaluated similarity keyword W ₁ and W ₂ in the keyword similarity evaluation unit 109, to obtain a score of evaluation results. The score calculation processing performed by the keyword similarity evaluation unit 109 in step S706 is score calculation performed by the keyword similarity evaluation unit 109 according to the flowchart of FIG. 7 with respect to step S205 of the side effect similarity calculation processing of FIG. Same as processing. Therefore, a detailed description of the score calculation process is omitted here, but one example is as follows.

For example, when the learning target drug ID is “111222A3333” and the selected drug ID is “998877F5050”, “bronchitis” is selected as the keyword W ₁ from the learning result table 204 of FIG. 4, and “acute upper respiratory tract inflammation” is selected as the keyword W _2. May be selected. For the combination of “bronchitis” and “acute upper respiratory tract inflammation”, the keyword similarity evaluation unit 109 uses the score calculation process shown in FIG. As a result, 4 (= 2 + 2) points are obtained, and a score of 4 points is returned to the analogy drug determination / learning unit 112.

After causing the keyword similarity evaluation unit 109 to calculate the score corresponding to the combination of the keywords W ₁ and W ₂ in step S706, the analog similarity determination / learning unit 112 calculates the score in the keyword similarity evaluation unit 109 in step S707. It is determined whether or not the score satisfies a predetermined standard. Specifically, in the reference value information 502 of FIG. 8, if the shorter reference value or more points that are associated with the length of the keyword W ₁ and W ₂ are obtained in step S706, the similar drugs determination The learning unit 112 determines that “predetermined criteria are satisfied”. For example, when | W ₁ | = 10 and | W ₂ | = 8, the analogy drug determination / learning unit 112 has a reference value (10 × β) associated with a length of 8 bytes in the reference value information 502 It is determined whether or not the above score has been obtained.

If the score is equal to or greater than the reference value, the similarity between the keywords W ₁ and W ₂ has reached a criterion that can be regarded as “match”, and the process moves to step S708. On the other hand, if the score is less than the reference value, the keywords W ₁ and W ₂ have not reached a criterion that can be regarded as “match”, and the process returns to step S704.

Similar drugs determination and learning unit 112 in step S708, the learning pairs keyword _{W 1} and _{W 2} as synonyms. That is, the analogy drug determination / learning unit 112 adds an entry in which the keywords W ₁ and W ₂ are paired to the synonym dictionary 203.

  Then, in the next step S709, the analogy drug determination / learning unit 112 determines whether the score obtained in step S706 falls within the top three of the scores obtained so far for the keyword combinations that can be regarded as matching each other. Judge whether or not.

Specifically, if any one of the variables P _{1 to} P ₃ remains as NULL in the initial state, the analogy medicine determination / learning unit 112 “the score obtained in step S706 is within the top three” Judge. If specific values have already been set for all of the variables P _{1 to} P ₃ , the analogy drug determination / learning unit 112 has a larger score in step S706 than the value of the variable P ₃ (that is, the third-ranked score). For example, it is determined that “the score obtained in step S706 is within the top three”.

Conversely, if specific values have already been set for all of the variables P _{1 to} P ₃ and the score obtained in step S706 is equal to or less than the value of the variable P ₃ , the analogy medicine determination / learning unit 112 determines that “step S706 The score obtained in is not within the top three. "

  If the similar drug determination / learning unit 112 determines that “the score obtained in step S706 is within the top three”, the process proceeds to step S710, and otherwise, the process returns to step S704.

In step S710, the similar drug determination / learning unit 112 appropriately updates the variables P _{1 to} P ₃ according to the score obtained in step S706.
Specifically, if the variable _{P 1} is NULL, similar drugs determination and learning unit 112 assigns a score obtained in step S706 to the variable _{P 1.} Further, when the variable P ₁ is not NULL and the variable P ₂ is NULL, the analog medicine determination / learning unit 112 substitutes the score obtained in step S706 for the variable P ₂ . Then, when the variables P ₁ and P ₂ are not NULL and the variable P ₃ is NULL, the analog medicine determination / learning unit 112 substitutes the score obtained in step S706 for the variable P ₃ .

On the other hand, when specific values are set for all of the variables P _{1 to} P _3, the analogy drug determination / learning unit 112 updates the variables as follows.
That is, if the number obtained in step S706 is greater than the value of the variable _{P 1,} similar drugs determination and learning unit 112 substitutes the current value of the variable _{P 2} into the variable _{P 3,} the variable _{P 2} of the current variable P substituting a value of _1, assigning the score obtained in step S706 to the variable _{P 1.} Alternatively, a number of points is equal to or less than the value of the variable P ₁ in step S706, the and the value is greater than the variable P _2, similar drugs determination and learning unit 112 substitutes the current value of the variable P ₂ into the variable P ₃ substitutes score obtained in step S706 to the variable _{P 2.} Alternatively, a number of points is equal to or less than the value of the variable _{P 2} in step S706, and if greater than the value variable _{P 3,} similar drugs determination and learning unit 112 assigns a score obtained in step S706 to the variable _{P 3.}

When the updating of the variables P _{1 to} P ₃ is completed as described above, the process returns to step S704.
Also, similar drugs determination and learning unit 112 in step S711 performs the score normalization process using the variables _{P 1} and _{P 2} and _{P 3} as arguments. The point normalization process in step S711 is the point normalization performed by the side effect determination / learning unit 108 according to the flowchart of FIG. 10 in step S210 of FIG. This is the same as the conversion process. Therefore, detailed description is omitted here.

  In step S712, the analogy drug determination / learning unit 112 returns the score normalized in step S711 as a return value of the process in FIG. 13, and the process in FIG. 13 ends. That is, in step S613 in FIG. 12 corresponding to the process in FIG. 13, the analogy drug determination / learning unit 112 acquires the normalized score as the similarity.

  As described above, according to the preprocessing (c4) described above with reference to FIGS. 11 to 13, fields other than the “known side effect list” field of the learning result table 204 are learned in advance. Therefore, even if safety information report documents 205 are received from a plurality of medical institutions for a specific in-house drug, the similar drug processing unit 103 only needs to learn the similar drug for the specific in-house drug once in the pre-processing. Good.

Further, the effect / effect similarity calculation process of FIG. 13 has the following characteristics (e1) and (e2).
(E1) The algorithm illustrated in FIG. 13 is different from a general algorithm that calculates similarity between documents according to a vector space model using TF / IDF values, and whether or not it is a similar drug based on efficacy or effect. It is devised so that it is suitable for judging. That is, the algorithm shown in FIG. 13 indicates that it is not uncommon for a medicine to have a plurality of effects or effects, and that medicines showing a high degree of similarity with some effects or effects can be regarded as similar drugs. It has been devised to be suitable for the determination of similar drugs.

  For example, a first drug may be effective for diseases X and Y, and a second drug may be effective for diseases Y and Z. In this case, the effects of the first and second drugs are not necessarily highly similar as a whole. Therefore, according to a general algorithm that calculates similarity between documents according to a vector space model using TF / IDF values, a result that “the first and second drugs are similar” is not always obtained. Absent.

  On the other hand, the first and second medicines can be regarded as similar drugs in that they are effective against disease Y. However, according to the processing of FIG. The medicine determination / learning unit 112 can obtain a conclusion that “the first and second medicines are similar medicines”. The reason is as follows.

The efficacy / effect keyword groups corresponding to the first and second pharmaceutical products each include a keyword related to the disease Y. Therefore, according to this embodiment, regarding the combination of keywords related to the disease Y, it is determined in step S707 in FIG. Therefore, the score normalization process using the score indicating the similarity between the keywords related to the disease Y is performed. Therefore, if the thresholded gamma ₁ is appropriately set as shown in FIG. 12, similar drugs determination and learning unit 112 can determine that "the first and second pharmaceutical is between similar drugs."

  Depending on the embodiment, the analogy drug determination / learning unit 112 may use the TF / IDF value as an auxiliary. That is, the similar drug determination / learning unit 112 obtains the feature vector of the “efficacy or effect” section of the attached document 202 of each of the learning target drug and the selected drug using the TF / IDF value, and the proximity of the feature vectors (for example, The angle between feature vectors may be calculated. Then, based on both the proximity of the feature vectors and the similarity obtained by the processing of FIG. 13, the similar medicine determination / learning unit 112 determines whether or not the learning target medicine and the selected medicine are similar medicines in step S614 of FIG. 12. It may be judged.

  (E2) In the process shown in FIG. That is, as apparent from the score calculation process of FIG. 7, the longer the match between the keywords, the higher the score naturally. However, in step S709 of FIG. Only is used as a criterion for judgment. Therefore, when there is a matching combination between long keywords, the matching between short keywords may not affect the processing after step S711 at all.

  For example, it is assumed that there are three sets of 10-byte keywords that completely match between the learning target drug and the selected drug. Then, no matter how many other 4-byte keyword pairs are completely matched, the score obtained by matching these 4-byte keywords is not considered in step S711 and subsequent steps. Therefore, even if the similarity determination obtained by the processing of FIG. 13 uses the similarity determination / learning unit 112 in step S614 of FIG. Does not reach.

  As described above, in this embodiment, matching between long keywords is more important than matching between short keywords, and is considered with priority. This point can be said to be a feature suitable for determination of analogs for the following reasons.

  As a general tendency, longer keywords often show specific details. Therefore, if the long keywords match between the learning target drug and the selected drug, there is a high probability that the learning target drug and the selected drug have the same kind of effect on limited individual specific diseases or symptoms. . That is, if long keywords match, the learning target drug and the selected drug are highly likely to be similar drugs. Therefore, in the present embodiment, in order to emphasize the matching of long keywords reflecting the high probability, in step S709 of FIG. Only is used as a criterion for judgment.

Next, the preprocessing (c5) performed by the determination apparatus 100 in accordance with the addition or update of the attached document 202 will be described with reference to FIG.
FIG. 14 is a flowchart of the addition / update process. In the present embodiment, the preprocessing control unit 113 starts the addition / update process in response to an input instructing the start of the addition / update process. The input also includes designation of the following information (f1) and (f2).

(F1) ID (hereinafter referred to as “target drug ID”) of a pharmaceutical product (hereinafter referred to as “target drug”) to be added or updated in the attached document 202.
(F2) New attached document specifying information for specifying a new attached document to be added or updated. For example, when a file of a new attached document has already been created on the storage device 307, the new attached document specifying information may be the path of the file. Alternatively, when there is a file of a new attached document on another computer 312, the new attached document specifying information may be a Uniform Resource Identifier (URI).

  When the target drug ID and the new attached document specifying information are input via the input device 305 and an instruction to start the addition / update process is given, the preprocessing control unit 113 starts the addition / update process. In step S801, the preprocessing control unit 113 searches the learning result table 204 using the input target drug ID as a search key, and whether or not there is an entry having the input target drug ID as an ID in the learning result table 204. To check.

  If there is an entry having the input target drug ID as an ID in the learning result table 204, it means that the start of the addition / update process has been instructed to update the registered attached document 202, so the process is step S802. Migrate to On the other hand, if there is no entry having the inputted target drug ID as an ID in the learning result table 204, it means that the start of the addition / update process has been instructed to add the attached document 202 for the new medicine. The processing moves to step S803.

  In step S802, the preprocessing control unit 113 replaces the attached document 202 of the target drug with a new attached document specified by the new attached document specifying information. Then, the process proceeds to step S805.

  In step S803, the preprocessing control unit 113 adds the new attached document specified by the new attached document specifying information to the attached document group 201 as the attached document 202 of the target drug. Then, the process proceeds to step S804. In the present embodiment, as described above, since the file name of the attached document 202 includes the medicine ID, the preprocessing control unit 113 appropriately changes the file name in steps S802 and S803.

  In step S804, the preprocessing control unit 113 adds an entry for the target drug to the learning result table 204. That is, the preprocessing control unit 113 sets the target drug ID designated via the input device 305 in the “ID” field and adds an entry in which the other fields are initialized to empty to the learning result table 204. . Then, the process proceeds to step S805.

  In step S805, the preprocessing control unit 113 instructs the efficacy / effect keyword extraction unit 111 to extract a keyword from the “efficacy or effect” section of the attached document 202 of the target drug and register it in the learning result table 204. . Then, the efficacy / effect keyword extraction unit 111 performs keyword extraction and registration in the learning result table 204 according to the command. Note that step S805 is similar to step S504 in FIG. 11, and the only difference is in which medicine attached document 202 is the target of keyword extraction, and detailed description thereof is omitted.

  In step S806, the preprocessing control unit 113 instructs the side effect keyword extraction unit 107 to extract a keyword from the “side effect” section of the attached document 202 of the target drug and register it in the learning result table 204. . Then, the side effect keyword extraction unit 107 performs keyword extraction and registration in the learning result table 204 according to the command. Note that step S806 is similar to step S505 in FIG. 11, and the only difference is in which medicine attached document 202 is the target of keyword extraction, and detailed description thereof is omitted. Further, the execution order of steps S805 and S806 may be reversed.

Further, in the next step S807, the preprocessing control unit 113 calculates the number of medicines for which the attached document 202 is registered, and stores the calculated number in a variable N.
In the next step S808, the preprocessing control unit 113 instructs the analogy drug determination / learning unit 112 to learn the analogy of the target drug, and the analogy drug determination / learning unit 112 selects the analogy of the target drug. learn. That is, the process performed by the similar drug determination / learning unit 112 in step S808 is the similar drug learning process shown in FIG. In step S808, the preprocessing control unit 113 notifies the target drug ID designated at the start of the addition / update process to the analogy drug determination / learning unit 112 as the learning target drug ID and compares it with the learning target drug. Designate the range from 1st to Nth.

  Note that, as shown in step S601 of FIG. 12, the similar drug determination / learning unit 112 initializes the similar drug learning list using the learned drug list in the learning result table 204. Therefore, when the addition / update process of FIG. 14 is performed to update the attached document 202, the contents of the target drug analogy drug list already learned in the learning result table 204 regarding the target drug are not deleted.

  As described above, according to the present embodiment, the analog medicine processing unit 103 learns the analog medicine list in advance regarding the determination target drug specified by the determination target drug ID 206. It is not determined which drug is the determination target drug at the pre-processing stage, but the pre-processing will be described again from the viewpoint of paying attention to the determination target drug.

  That is, the efficacy / effect keyword extraction unit 111 performs a phrase extraction process on the description part of the efficacy or effect in the attached document 202 of each of the determination target drug and other drugs other than the determination target drug to generate an effect / effect phrase set. It is an example of the effect effect phrase extraction means to acquire. In the present embodiment, the acquired effect / effect phrase set is stored in the “effect / effect keyword group” field of the learning result table 204.

  The similar drug determination / learning unit 112 is an example of a first similar drug determination unit that determines whether or not the determination target drug is similar to at least a part of the drug other than the determination target drug. is there. The analogy drug determination / learning unit 112 as the first analogy drug determination unit includes a phrase included in the efficacy effect phrase set acquired for the other drug and a phrase included in the efficacy effect phrase set acquired for the determination target drug And the keyword similarity evaluation unit 109 evaluates the combination.

  Furthermore, the similar drug determination / learning unit 112 as the first similar drug determination unit aggregates evaluations regarding a plurality of combinations, so that the determination target drug and the other drug focused on as the selected drug A value indicating the efficacy or similarity between effects is calculated. In other words, the analogy medicine determination / learning unit 112 selects and normalizes the top three rankings of words that satisfy the matching criteria, specifically, as a summary of evaluation by the keyword similarity evaluation unit 109. I do.

Then, the analog drug determination / learning unit 112 as the first analog drug determination unit compares the value indicating the similarity calculated by aggregation as described above with the threshold γ ₁ . Then, similar drugs determination and learning unit 112, when the degree of similarity indicated calculated value is higher than the similarity degree indicating the threshold value gamma _1, determines the other drugs and determination target drug similar drugs. Therefore, the analog drug recognition unit 105 that recognizes the analog of the determination target drug with reference to the analog drug list recognizes the analog according to the determination result of the analog drug determination / learning unit 112 as the first analog drug determination unit. I can say that.

  In addition, it has been described that the determination by the similar drug determination / learning unit 112 as the first similar drug determination unit is performed for at least a part of the pharmaceutical other than the determination target drug. However, the meaning of “at least a part” means It is as follows.

  In other words, the determination as the first analog determination unit is performed in steps S613 and S614 in FIG. And step S613 is not necessarily performed about all the pharmaceutical products.

  That is, the medicinal product determination / learning unit 112 determines whether the medicinal property classification name, the reference name, the common name, the chemical name, or the structural formula described in the attached document 202 among the plurality of other drugs other than the determination target drug is to be determined. A function as a second similar drug determination unit that determines a drug matching the drug as a similar drug is also realized. Then, the similar drug determination / learning unit 112 performs the processing of steps S607 to S612 in FIG. 12 as the second similar drug determination unit, and only the drug that has not been determined as the similar drug as the second similar drug determination unit. Then, the processing of steps S613 to S614 is performed as a first analogy drug determination unit.

  Therefore, the target for determination by the similar drug determination / learning unit 112 as the first similar drug determination means is not necessarily all of the pharmaceuticals other than the determination target drug. Therefore, in the above description, “at least part” is described. As described above, by limiting the target for performing the processing of steps S613 to S614, which is more complicated than the processing of steps S607 to S612, to some analogs, it is possible to reduce a wasteful processing load.

  By the way, the present invention is not limited to the above embodiment. Although some modifications have been described in the above description, the above embodiment can be further modified variously from the viewpoints of (g1) to (g8) below, for example, as long as these modifications do not contradict each other. Any combination is possible.

(G1) Modification Regarding Score Calculation Processing in FIG. 7 Depending on the keyword extraction algorithm, the score calculation processing in FIG. The data format of the side effect keyword group may be modified.

  Specifically, when the side effect keyword extraction unit 107 performs keyword extraction using morphological analysis, the score calculation process of FIG. 7 performed by the keyword similarity evaluation unit 109 in step S205 of FIG. 6 is modified as follows. May be. Similarly, when the effect / effect keyword extraction unit 111 performs keyword extraction using morphological analysis, the score calculation process of FIG. 7 performed by the keyword similarity evaluation unit 109 in step S706 of FIG. 13 is modified as follows. May be.

That is, the score calculation process of FIG. 7 may be modified so that the division position for dividing the keyword into partial character strings is limited to the morpheme division position.
The score calculation algorithm in FIG. 7 follows a policy of examining all the possible division patterns for both keywords A and B. Therefore, in the score calculation process of FIG. 7, an arbitrary position in the keyword can be a division position for dividing the partial character strings.

  That is, the keyword similarity evaluation unit 109 obtains the similarity between two words, and the two words can be divided as any divided pattern that is divided between any two adjacent characters in each word. Aggregate evaluations for all combinations of patterns. Note that “aggregation” here specifically corresponds to selecting the highest evaluation in the example of FIG. 7.

  Therefore, in the embodiment shown in FIG. 7, the number of combinations of division patterns considered by the keyword similarity evaluation unit 109 is large, and the processing load on the keyword similarity evaluation unit 109 is high.

  On the other hand, in the score calculation process modified so as to limit the division position for dividing the keyword into partial character strings to the morpheme division position, the number of division patterns for each of the keywords A and B is limited. The number can be kept small. Therefore, the processing load of the keyword similarity evaluation unit 109 is also reduced.

For example, since the keyword 601 “general anesthetic” illustrated in FIG. 9 is five characters, 16 (= 2 ^5-1 ) division patterns are considered for the keyword 601 in the score calculation process of FIG.

  On the other hand, if the keyword 601 is extracted using the result of morphological analysis, in the score calculation process modified as described above, the number of possible division patterns for the keyword 601 is very small. is there.

For example, it is assumed that the keyword 601 “general anesthetic” is obtained from the result of morphological analysis as a series of a noun “whole body”, a noun “anesthetic”, and a noun “agent”. Then, since there are only two morpheme division positions in the keyword 601, in the score calculation process modified as described above, the number of possible division patterns for the keyword 601 is only 4 (= 2 ² ). Absent. In other words, in this example, there are only four patterns in total, that is, the three division patterns 603a to 603c shown in FIG. 9 and the division pattern “whole body / anesthetic”.

  Therefore, the score calculation process modified as described above has an effect of reducing the calculation load of the keyword similarity evaluation unit 109 as compared with the score calculation process of FIG. On the contrary, the point calculation process of FIG. 7 is superior to the modified score calculation process described above in terms of robustness against unknown words that are not registered in the morphological analysis dictionary. It is done.

  In addition, the learning result table 204 may be modified so that the keyword similarity evaluation unit 109 can perform the score calculation process modified as described above. For example, the learning result table 204 may be modified so as to include information indicating the position of the morpheme segmentation in each keyword extracted by the side effect keyword extraction unit 107 using the result of the morpheme analysis. Information indicating the position of the morpheme break may be stored in a new field of the learning result table 204 or may be included in the “side effect keyword group” field itself, for example.

  For example, the side effect keyword extraction unit 107 uses the appropriate delimiter character or marks up according to a predetermined grammar to indicate the keyword indicating the position of the morpheme separation in the learning result table 204 as each element of the side effect keyword group. It may be recorded. Alternatively, the side effect keyword extraction unit 107 may record information indicating the position of the morpheme break for each keyword in the side effect keyword group in a new field of the learning result table 204.

Of course, the effect / effect keyword extraction unit 111 and the “effect / effect keyword group” field can be modified in the same manner as described above.
If the learning result table 204 holds information indicating the position of the morpheme break, the side effect determination / learning unit 108 or the analogy drug determination / learning unit 112 receives the information indicating the position of the morpheme break from the learning result table 204. It can be read easily.

  Therefore, the side effect determination / learning unit 108 can notify the keyword similarity evaluation unit 109 of information indicating the position of the morpheme break in the selected keyword in step S205 of FIG.

  Further, if the selected side effect is extracted from the safety information report document 205 by morpheme analysis in step S104 of FIG. 3, the side effect keyword extraction unit 107 uses information indicating the position of the morpheme break in the selected side effect to determine the side effect. The learning unit 108 can also be notified. Therefore, the side effect determination / learning unit 108 can notify the keyword similarity evaluation unit 109 of the information notified from the side effect keyword extraction unit 107, and the keyword similarity evaluation unit 109 recognizes the position of the morpheme break in the selected side effect. can do.

Likewise, similar drugs determination and learning unit 112 in step S706 of FIG. 13, it is possible to notify the information indicating the position of the morpheme separated in the keyword W ₁ and W ₂ respectively in the keyword similarity evaluation unit 109.

  Therefore, in the score calculation process modified as described above, the keyword similarity evaluation unit 109 can recognize information indicating the positions of morpheme breaks in each of the two keywords to be processed. Then, as described above, the keyword similarity evaluation unit 109 calculates a relatively small number of points indicating the degree of similarity between two keywords by limiting the division position at which the keyword is divided into partial character strings to morpheme division positions. Can be calculated in quantity.

  Further, the keyword similarity evaluation unit 109 may perform the processes of steps S306 to S314 in FIG. 7 only for combinations of divided patterns that do not include a partial character string shorter than a predetermined lower limit. For example, the lower limit may be determined as “2 characters” or “4 bytes”. Then, the effect of reducing the amount of calculation is also obtained. For example, a keyword including “heart” and a completely different keyword including “kidney” are added in noise due to a match of one character “vibration”. The influence of the number of points can be eliminated.

(G2) Modification Regarding Timing of Learning Similar Medicine In the above embodiment, the similar medicine learning process of FIG. 12 is performed in step S509 of FIG. 11 and step S808 of FIG. That is, the similar medicine is learned before the processing of FIG. 3 is executed. That is, in the above-described embodiment, the similar drug recognition unit 105 reads out the similar drug learning result information that associates the identification information for uniquely identifying the drug with the similar drug of the drug from the learning result table 204 of the storage unit 101. , Recognize the similar drugs to be judged.

  However, depending on the embodiment, the similar drug recognition unit 105 may include the similar drug processing unit 103. In this case, when the side effect processing unit 102 performs the processing of FIG. 3, the similar drug processing in the similar drug recognition unit 105 is performed. The unit 103 may perform the similar medicine learning process of FIG.

  That is, the similar medicine processing unit 103 performs the similar medicine learning process on the spot when the side effect processing part 102 performs the process of FIG. 3 instead of performing the similar medicine learning process in advance as in the above embodiment. Also good. Specifically, the similar medicine learning process of FIG. 12 may be performed immediately before step S102 of FIG.

  In this case, for example, the similar drug recognition unit 105 may designate the determination target drug ID 206 as the learning target drug ID to the similar drug determination / learning unit 112. Further, the analogy drug recognition unit 105 compares all the attached documents 202 from the first to the Nth in the same manner as the preprocessing control unit 113 designates to the analogy drug determination / learning unit 112 in step S808 of FIG. It is specified to the analogy drug determination / learning unit 112 as a target range.

  When the similar drug determination / learning unit 112 finishes the similar drug learning process of FIG. 12, the similar drug recognition unit 105 can acquire the learned drug list in step S102. Alternatively, the similar medicine determination / learning unit 112 may directly notify the similar drug learning unit 105 of the result of the similar drug learning process instead of recording the result of the similar drug learning process in the learning result table 204. That is, depending on the embodiment, the field of the analogy drug list in the learning result table 204 may be omitted.

  In summary, when the analog drug recognition unit 105 that realizes the analog drug recognition unit includes the analog drug processing unit 103 and the analog drug processing unit 103 performs the analog drug learning process immediately before step S102, the analog drug recognition unit 105 Recognizes similar drugs by reading the package insert 202 for each of a plurality of drugs. That is, the analog drug processing unit 103 as a part of the analog drug recognition unit reads the attached document 202 from the storage unit 101 for each of a plurality of drugs, and performs an analog drug learning process using the read attached document 202. Then, as a result of the analog drug learning process, the analog drug recognition unit 105 including the analog drug processing unit 103 can recognize an analog similar to the determination target drug among a plurality of other drugs.

  In addition, the addition of an entry to the synonym dictionary 203 by the side effect determination / learning unit 108 or the similar drug determination / learning unit 112 may affect the determination of whether or not a certain two drugs are similar drugs. This is because the processing of FIG. 13 is performed in step S613 of the similar medicine learning processing of FIG. 12, and the synonym dictionary 203 is performed in step S309 of the score calculation processing of FIG. 7 performed by the keyword similarity evaluation unit 109 in step S706 of FIG. Is referred to. Therefore, the preprocessing control unit 113 orders the analogy medicine determination / learning part 112 to re-execute the analogy drug learning process irregularly triggered by the addition of an entry to the synonym dictionary 203 or periodically at an appropriate interval. May be.

(G3) Modification Regarding Device Configuration In the above embodiment, the determination device 100 includes both the side effect processing unit 102 and the similar drug processing unit 103, but the first device including the similar drug processing unit 103 and the side effect processing unit 102 The 2nd apparatus containing may be provided separately. Then, the first device learns fields other than the known side effect list in the learning result table 204 by preprocessing, and the second device receives the data of the learning result table 204 obtained as a result of learning from the first device. Also good.

  The second device can store the data of the received learning result table 204 in a storage device accessible by the second device and refer to it. Therefore, the second device can execute the process of FIG. 3 without performing the pre-processing by itself.

  Further, in the determination apparatus 100 of FIG. 1, the keyword similarity evaluation unit 109 is shared between the side effect processing unit 102 and the similar medicine processing unit 103. However, depending on the embodiment, separate keyword similarity evaluation units 109 may be provided for the side effect processing unit 102 and the similar drug processing unit 103, respectively.

  Further, in FIG. 1, the storage unit 101 is inside the determination apparatus 100, but the storage unit 101 may be outside the determination apparatus 100. For example, components other than the storage unit 101 of the determination apparatus 100 may be realized by the computer 300 in FIG. 2, and the storage unit 101 may be realized by a storage device of the other computer 312 in FIG.

(G4) Modifications Related to User Interface The user interface illustrated with respect to the above embodiment is merely an example.
For example, the side effect determination / learning unit 108 may cause the output device 306 to display a screen other than the side effect determination result screen 400 shown in FIG. Alternatively, the side effect determination / learning unit 108 may output the determination result for each selected side effect via a command-line interface (CLI) instead of a graphical user interface (GUI). Further, the side effect determination / learning unit 108 receives an input from the user regarding whether the side effect is known or unknown regarding the side effect determined to be known via a GUI such as the side effect determination result screen 400 of FIG. Can also be received via the CLI.

  Similarly, the user interface for the analogy medicine determination / learning unit 112 to accept the determination result of the user regarding whether or not two medicines are similar is also optional depending on the embodiment, and may be GUI or CLI. Good.

  In the example of FIG. 3, the side effect determination / learning unit 108 obtains the similarity between the side effect keyword group of the related drug and the selected side effect for all similar drugs as the determination target drugs in step S112. However, depending on the embodiment, the side effect determination / learning unit 108 may perform the process of step S112 only for some of the determination target drugs. Hereinafter, a case where there are three similar drugs in the determination target drug will be described as a specific example.

The side effect determination / learning unit 108 first performs the process of step S112 on the first drug, and determines that the selected side effect is a known side effect if the resulting similarity is α ₁ or more. can do. Therefore, the side effect determination / learning unit 108 may perform the process of step S116 immediately after performing the processes of steps S113 and S114 without returning to step S110. That is, in this case, the side effect determination / learning unit 108 may omit the comparison between the second and third similar drugs and the determination target drug regarding the selected side effect.

In some cases, resulting similarity of processing first step S112 of performing with respect to similar drugs are the α less than _1, as a result of the processing of step S112 was carried out with respect to the second similar drugs The obtained similarity may be α ₁ or more. In this case, the side effect determination / learning unit 108 may perform the process of step S114 after the determination of step S113 for the second similar drug, and immediately execute the process of step S116.

That is, the side effects judgment and learning unit 108 until alpha ₁ or more similarity is obtained consider one after the other class drugs, as long obtained alpha ₁ or more similarity with respect to a class of drugs, the remaining similar drugs The processing of steps S111 to S115 may be omitted.

(G5) Data-related modification FIG. 1 shows the synonym dictionary 203 in a table format, but the data format of the synonym dictionary 203 is arbitrary. The data represented by the learning result table 204 in FIG. 4 can also be represented in any data format other than the table. For example, the synonym dictionary 203 and the learning result table 204 may be realized by an eXtensible Markup Language (XML) database.

  The synonym dictionary 203 can be omitted. In that case, the process of adding an entry to the synonym dictionary 203 can be omitted, and the partial character string similarity evaluation unit 110 performs only the partial character string that completely matches the partial character string substr in step S309 of FIG. Search for.

  In the above embodiment, the case where each attached document 202 is one file has been described as an example. However, the entire attached document group 201 may be one file. Alternatively, the entire learning result table 204 and the attached document group 201 may be realized by a single XML database file.

  In the above-described embodiment, the sales name code is used as the ID of each medicine, but data other than the sales name code (for example, a serial number automatically assigned to each medicine by the preprocessing control unit 113) is used as the ID. Also good.

  Moreover, the specific numerical values in the scoring information 501 and the reference value information 502 shown in FIG. 8 are examples, and the specific numerical values can be appropriately changed according to the embodiment. Further, for example, according to the reference value information 502 in FIG. 8, the reference value is defined as a constant multiple (β times) of the maximum possible score, but if the reference value is less than the maximum possible score, It may not be a constant multiple of the maximum possible score. In some embodiments, β may be a variable parameter that can be specified by the user.

  In the learning result table 204, some of the fields illustrated in FIG. 4 may be omitted. For example, as described in the above (g2), the “similar medicine list” field may be omitted depending on the embodiment.

  Similarly, the “efficacy / effect keyword group” field and the “side effect keyword group” field may be omitted depending on the embodiment. That is, instead of eliminating these two fields, the efficacy / effect keyword extraction unit 111 or the side effect keyword extraction unit 107 can perform keyword extraction each time a process using a keyword group is performed.

  For example, when the side effect keyword extraction unit 107 performs keyword extraction each time a process using a keyword group is performed, the side effect keyword extraction unit 107 can realize a comparison target set acquisition unit. That is, the side effect keyword extraction unit 107 serving as a comparison target set acquisition unit acquires a set of words and phrases included in the side effect description part in the attached document 202 of the medicine recognized as a similar drug as a comparison target word set.

  Specifically, the side effect keyword extraction unit 107 serving as a comparison target set acquisition unit reads out from the storage unit 101 the attached document 202 of a medicine recognized as an analog. Then, the side effect keyword extraction unit 107 acquires a set of comparison target phrases by extracting a set of phrases from the side effect description part in the read attached document 202 by a phrase extraction process.

  In some embodiments, the “known side effect list” field can also be omitted. However, in the above embodiment, a “known side effect list” field is provided in the learning result table 204 on the assumption that the same side effect may be reported multiple times for the same in-house drug from various medical institutions.

  In other words, since the side effect is not known during the period from the time when the occurrence of a side effect is first reported to the pharmaceutical company by the medical institution until the revision of the package insert 202, the different medical institution will be assigned to the pharmaceutical company. May report side effects. In this case, in the above embodiment, the learning result table 204 includes the “known side effect list” field so that the determination apparatus 100 or the user does not have to repeat the same determination many times. Study the “known side effects list” field.

  Contrary to the omission of fields as exemplified above, the learning result table 204 may further include fields not shown in FIG. For example, when the determination apparatus 100 is operated in a pharmaceutical company, the learning result table 204 may have a flag field indicating whether it is an in-house drug or a competitor drug. Then, it becomes possible to limit learning target drugs for learning similar drugs to in-house drugs.

  The processing of FIGS. 11 and 14 in the above-described embodiment is processing that does not depend on which pharmaceutical company each pharmaceutical product is, and can be applied when the determination apparatus 100 is operated by a third party organization other than the pharmaceutical company. It is processing. On the other hand, when the determination apparatus 100 is operated by a pharmaceutical company, the safety information report document 205 is only related to its own drug. Therefore, information on whether or not the other company's drugs are similar is not necessary, and it is sufficient if there is only information on whether or not the own company's drugs are similar or whether the other company's drugs and other companies' drugs are similar.

  Therefore, for example, the preprocessing control unit 113 sets the learning result table 204 in the learning result table 204 corresponding to the ID of the i-th medicine before causing the similar drug determination / learning unit 112 to perform the similar drug learning process in step S509 of FIG. Refer to the value of the flag in the entry. Then, the preprocessing control unit 113 can determine from the value of the flag whether or not the i-th drug is an in-house drug.

  If the i-th drug is an in-house drug, the preprocessing control unit 113 causes the similar drug determination / learning unit 112 to perform similar drug learning processing as in step S509. On the other hand, if the i-th drug is another company's drug, step S509 is omitted.

  Further, the processing of FIG. 14 may be modified as follows. That is, after step S806, the preprocessing control unit 113 refers to the value of the flag in the entry of the learning result table 204 corresponding to the target drug ID, and determines whether the target drug is an in-house drug. If the target drug is an in-house drug, the pre-processing control unit 113 performs the process of step S807 as in FIG. 14, and the analog drug determination / learning unit 112 performs the analog drug learning process in step S808. .

  On the other hand, when the target drug is another company's drug, the preprocessing control unit 113 instructs the similar drug determination / learning unit 112 to perform the similar drug learning process of FIG. . At that time, the preprocessing control unit 113 limits the comparison range in which the similar drug learning process is performed to only the target drug in FIG. 14 (that is, the other drug for which the attached document 202 has been added or updated). What is necessary is just to specify to the learning part 112.

By modifying the pre-processing as described above, it becomes possible for the similar drug processing unit 103 to learn the similar drug efficiently only for the own drug.
In the example of FIG. 4, the elements of the known side effect list in the learning result table 204 are side effects determined to be known. However, depending on the embodiment, a pair of the similar drug ID indicating which side effect is determined to be known based on which similar drug may be recorded as each element of the known side effect list. That is, during the learning in step S116 or S117 in FIG. 3, the side effect determination / learning unit 108 sets a pair in which the ID in the known side effect learning list or the known side effect candidate list is associated with the side effect in the learning result table 204. It may be added to the side effect list.

(G6) Modification related to point normalization processing The point normalization processing in FIG. 10 uses the upper three points as arguments. However, the score normalization process may be modified so that the score up to the upper T rank is used as an argument, where T is an arbitrary integer of 2 or more. Even in that case, the side effect determination / learning unit 108 or the analogy drug determination / learning unit 112 that performs the point normalization process uses t (t ≦ T) arguments that are specifically given values instead of NULL. , The root mean square of the t argument values may be calculated as a normalized value.

  Alternatively, the side effect determination / learning unit 108 or the analogy drug determination / learning unit 112 may calculate an arithmetic average, a geometric average, a weighted average, and the like of a plurality of points instead of a root mean square of a plurality of points. You may obtain | require as a normalized score.

(G7) Modification Regarding Scoring In the above embodiment, the score calculation process that the analog medicine determination / learning unit 112 causes the keyword similarity evaluation unit 109 to perform, and the score that the side effect determination / learning unit 108 performs to the keyword similarity evaluation unit 109 The same scoring information 501 in FIG. 8 is used in the calculation process. However, in some embodiments, the side effect determination / learning unit 108 evaluates the keyword similarity by using different score information from the point calculation processing score information that the analog similarity determination / learning unit 112 causes the keyword similarity evaluation unit 109 to perform. It may be used in the score calculation process to be performed by the unit 109.

  In that case, two types of reference value information may be used properly in accordance with the two types of scoring information being used properly. In other words, the matching criterion in step S206 in FIG. 6 and the matching criterion in step S707 in FIG. 13 may be different criteria.

  Further, in step S309 in FIG. 7, the points added in step 310 may be different depending on whether the partial character strings are completely matched with each other as synonyms. That is, the scoring information 501 in FIG. 8 is information that defines scoring for each length. However, even in the scoring information 501, a scoring for perfect match and a scoring for synonym matching are defined for each length. Good. In that case, the complete matching score corresponding to the same length is defined to be equal to or higher than the synonym matching score.

(G8) Modification Regarding Keyword Extraction In the above embodiment, the side effect keyword extraction unit 107 extracts a keyword from the “side effect” section of the attached document 202. However, depending on the embodiment, the side effect keyword extraction unit 107 may further extract keywords from other sections such as the “interaction” section of the attached document 202 and add them to the “side effect keyword group” of the learning result table 204. Good.

  Similarly, the efficacy / effect keyword extraction unit 111 extracts keywords from not only the “efficacy or effect” section of the attached document 202 but also other sections such as the “medicinal pharmacology” section, and the “effect / effect” in the learning result table 204. It may be added to “keyword group”.

  Further, regarding the above-described embodiment, several examples of keyword extraction algorithms respectively performed by the side effect keyword extraction unit 107 and the efficacy / effect keyword extraction unit 111 are illustrated. However, the algorithm for keyword extraction is not limited to that exemplified above. For example, the side effect keyword extraction unit 107 or the efficacy / effect keyword extraction unit 111 may have a stop word list, and stop words in the stop word list may not be extracted as keywords. For example, in terms used in the package insert 202, words such as “patient” and “administration” may be included in the stop word.

  In the above embodiment, as an example of keyword extraction using the result of morphological analysis, a technique of extracting a series of nouns as a keyword has been illustrated. However, the keyword may not be a series of nouns. For example, the side effect keyword extraction unit 107 or the effect / effect keyword extraction unit 111 may extract, as a keyword, the entire series of adjectives and nouns following the adjective from the result of morphological analysis.

  Further, when a plurality of types of parts of speech are included in the keyword as described above, the synonym dictionary 203 may register synonym pairs exceeding the difference in parts of speech. For example, an entry in which “rising” and “high” are paired may be registered in the synonym dictionary 203 in advance. Then, the score calculated by the score calculation process of FIG. 7 for the two keywords “blood pressure increase” and “high blood pressure” also increases to some extent, and it is possible to better reflect the meaning of the keyword in the score.

Finally, the following additional notes are disclosed regarding the various embodiments described above.
(Appendix 1)
A means for receiving information for identifying a drug, and identifying the drug indicated by the information as a determination target drug;
Report document acquisition means for acquiring a report document describing the side effects of the determination target drug;
By reading from the storage means the identification information that uniquely identifies the drug and the similar drug learning result information from the storage means, or for each of a plurality of drugs, the identification information of the drug and the side effects of the drug And a medicinal product recognition means for recognizing a similar drug to the determination target drug among a plurality of other medicinal products by reading out a package insert including the efficacy or effect of the medicinal product from the storage unit,
To evaluate the similarity between partial character strings included in a phrase, the first length and the second length are added to each other when the first and second partial character strings match. A partial character string similarity evaluation unit that gives a higher evaluation than the evaluation obtained by adding the evaluation when the partial character strings having the same length match each other and the evaluation when the partial character strings having the second length match. ,
By allowing the character string similarity evaluation means to evaluate the similarity between partial character strings obtained by dividing each of the two phrases, and summing up the evaluation results by the character string similarity evaluation means, the 2 Evaluate a combination of division patterns that divide each of the two phrases into one or more partial character strings, and use the evaluation of a plurality of combinations of the division patterns of each of the two phrases to determine the similarity between the two phrases A word similarity evaluation means for evaluating
From the report document, a side effect phrase extracting means for extracting a phrase indicating the side effect of the determination target drug as a determination target side effect phrase;
By reading out a package insert of a medicine recognized as the drug by the drug recognizing means from the storage means and extracting a set of words by a word extraction process from a side effect description part in the package attached, or The medicinal product recognized from the storage means as the medicinal product by a set of words and phrases obtained by the word extraction process from the side effect description part in the package insert of the medicinal product recognized by the medicinal product recognition unit as the medicinal product By reading out the side effect learning result information associated with the identification information, the set of words / phrases included in the description part of the side effect in the package insert of the drug recognized as the similar drug is acquired as a comparison target word / phrase set. A comparison target set acquisition means;
For at least a part of the analogy drug, the phrase similarity evaluation unit evaluates the combination of the word / phrase included in the comparison target word / phrase acquired for the analogy drug and the judgment target side effect word / phrase. A determination means for determining whether or not the side effect indicated by the determination target side effect phrase is a known side effect in the related drug, using the result and the first threshold;
An information processing apparatus comprising: output means for outputting a determination result by the determination means.
(Appendix 2)
The attached document of the determination target drug is read from the storage unit, and a phrase extraction process is performed on the indication of the effect or effect in the attached document to obtain a first effect-effect phrase set, and the plurality of other medicines The efficacy and effect of reading the package insert of the drug for each from the storage means and performing the phrase extraction process on the indication of the effect or effect in the package insert to obtain the second set of effect and effect phrases related to the other drug Word extraction means;
For at least some of the other pharmaceutical products, respectively
The phrase similarity evaluation means evaluates a combination of a phrase included in the second efficacy effect phrase set acquired for the other pharmaceutical product and a phrase included in the first efficacy effect phrase set,
By calculating the evaluation for a plurality of combinations, a value indicating the efficacy or similarity of the effect between the determination target drug and the other drug is calculated,
Comparing the calculated value with a second threshold;
When the similarity indicated by the calculated value is higher than the similarity indicated by the second threshold, the other medicine is determined as an analog of the determination target drug, and the identification information of the determination target drug and the other First analog medicine determination means for updating the analog medicine learning result information on the storage means so as to associate the medicines of
Further comprising
The information processing apparatus according to appendix 1, wherein the similar drug recognition unit recognizes the similar drug of the determination target drug according to a determination result of the first similar drug determination unit.
(Appendix 3)
Among the plurality of other medicinal products, a medicinal property classification name, a reference name, a common name, a chemical name or a structural formula described in the package insert are determined to be a similar drug, and the medicinal product is determined as the similar drug, A second similar medicine determination means for updating the similar medicine learning result information on the storage means so as to associate the medicine determined as the similar medicine with the identification information of the determination target medicine;
The first similar drug determination means determines whether or not it is the similar drug for a drug that has not been determined as the similar drug by the second similar drug determination means among the plurality of other drug drugs,
The additional drug recognition unit according to claim 2, wherein the similar drug recognition unit recognizes the similar drug of the determination target drug according to a determination result of both the second similar drug determination unit and the first similar drug determination unit. Information processing device.
(Appendix 4)
The phrase similarity evaluation means aggregates evaluations for all combinations of division patterns that can be divided into any two characters adjacent to each other in the two phrases. The information processing apparatus according to any one of appendices 1 to 3, wherein the similarity between the two words is obtained.
(Appendix 5)
Any one of Supplementary notes 1 to 3, wherein the word similarity evaluation means obtains the similarity between the two words using only a division pattern for dividing each of the two words at a morpheme break position. The information processing apparatus according to item 1.
(Appendix 6)
On the computer,
Accepts information to identify medicines,
The drug indicated by the information is identified as a determination target drug,
Obtain a report document describing the side effects of the determination target drug,
By reading from the storage means the identification information that uniquely identifies the drug and the similar drug learning result information from the storage means, or for each of a plurality of drugs, the identification information of the drug and the side effects of the drug And a medicinal product similar to the determination target drug among a plurality of other medicinal products by reading out the package insert including the efficacy or effect of the medicinal product from the storage unit,
From the report document, the phrase indicating the side effect of the determination target drug is extracted as a determination target side effect phrase,
By reading out a package insert of a medicine recognized as the similar drug from the storage means, and extracting a set of phrases from the side effect description part of the package insert by a phrase extraction process, or from the storage section, the class By reading out the side effect learning result information that associates the set of phrases obtained by the phrase extraction process from the description part of the side effects in the package insert of the drug recognized as a drug with the identification information of the drug recognized as the similar drug, A set of phrases included in the description part of the side effect in the package insert of the drug recognized as the similar drug is obtained as a set of phrases to be compared;
For at least a part of the similar drugs, the combination of the phrase included in the set of comparison target words acquired for the similar drug and the determination target side effect phrase is evaluated, and the result of the evaluation and the first threshold value To determine whether the side effect indicated by the determination target side effect phrase is a known side effect in the related drug,
Executing a side effect determination process including outputting a determination result of whether or not the side effect indicated by the determination target side effect phrase is a known side effect;
In order to evaluate the combination of the phrase included in the comparison target phrase set and the determination target side effect phrase, as a phrase similarity evaluation process,
When the partial character strings of the third length obtained by adding the first length and the second length match the similarity between the partial character strings obtained by dividing each of the two phrases respectively. The evaluation is performed so as to give a higher evaluation than the evaluation obtained by adding the evaluation when the partial character strings of the first length match and the evaluation when the partial character strings of the second length match. ,
Evaluating the combination of division patterns that divide each of the two phrases into one or more partial character strings by counting the similarities evaluated for the partial character strings,
The determination program which performs the process including evaluating the similarity of two said phrases using evaluation about multiple combinations of the division | segmentation pattern of each said two phrases.
(Appendix 7)
The side effect determination process further includes
Reading the attached document of the determination target drug from the storage unit, performing a phrase extraction process on the effect or effect description part in the attached document to obtain a first efficacy effect phrase set,
For each of the plurality of other medicines, a package insert of the drug is read from the storage means, and a phrase extraction process is performed on a description of the effect or effect in the package insert to obtain a second efficacy effect phrase related to the other drug Including obtaining a set,
The determination program is stored in the computer.
For at least some of the other pharmaceutical products, respectively
Evaluating a combination of a phrase included in the second efficacy effect phrase set acquired for the other pharmaceutical product and a phrase included in the first efficacy effect phrase set by the phrase similarity evaluation process,
By calculating the evaluation for a plurality of combinations, a value indicating the efficacy or similarity of the effect between the determination target drug and the other drug is calculated,
The calculated value is compared with a second threshold, and when the similarity indicated by the calculated value is higher than the similarity indicated by the second threshold, the other drug is determined as an analog of the determination target drug And further executing a first drug determination process including updating the drug learning result information on the storage means so as to associate the identification information of the drug to be determined with the other drug.
Recognizing the similar drug of the determination target drug according to the determination result of the first similar drug determination process,
The determination program according to appendix 6, characterized in that:
(Appendix 8)
The determination program is:
Among the plurality of other medicinal products, a medicinal property classification name, a reference name, a common name, a chemical name or a structural formula described in the package insert are determined to be a similar drug, and the medicinal product is determined as the similar drug, Causing the computer to execute a second similar drug determination process for updating the similar drug learning result information on the storage means so as to associate the drug determined to be the similar drug with the identification information of the determination target drug;
For a drug that has not been determined as the similar drug by the second similar drug determination process among the plurality of other drug products, the computer performs the first similar drug determination process,
The determination program according to appendix 7, wherein the computer recognizes the similar drug of the determination target drug according to determination results of both the second similar drug determination process and the first similar drug determination process.
(Appendix 9)
The phrase similarity evaluation process totals evaluations for all combinations of division patterns that can be divided into arbitrary two characters adjacent to each other in each of the two phrases. The determination program according to any one of supplementary notes 6 to 8, further comprising: calculating the similarity between the two words.
(Appendix 10)
The phrase similarity evaluation processing includes obtaining the similarity between the two phrases by using only a division pattern that divides each of the two phrases at a morpheme segmentation position. The determination program according to any one of the above.
(Appendix 11)
Computer
Accepts information to identify medicines,
The drug indicated by the information is identified as a determination target drug,
Obtain a report document describing the side effects of the determination target drug,
By reading from the storage means the identification information that uniquely identifies the drug and the similar drug learning result information from the storage means, or for each of a plurality of drugs, the identification information of the drug and the side effects of the drug And a medicinal product similar to the determination target drug among a plurality of other medicinal products by reading out the package insert including the efficacy or effect of the medicinal product from the storage unit,
From the report document, the phrase indicating the side effect of the determination target drug is extracted as a determination target side effect phrase,
By reading out a package insert of a medicine recognized as the similar drug from the storage means, and extracting a set of phrases from the side effect description part of the package insert by a phrase extraction process, or from the storage section, the class By reading out the side effect learning result information that associates the set of phrases obtained by the phrase extraction process from the description part of the side effects in the package insert of the drug recognized as a drug with the identification information of the drug recognized as the similar drug, A set of phrases included in the description part of the side effect in the package insert of the drug recognized as the similar drug is obtained as a set of phrases to be compared;
For at least a part of the similar drugs, the combination of the phrase included in the set of comparison target words acquired for the similar drug and the determination target side effect phrase is evaluated, and the result of the evaluation and the first threshold value To determine whether the side effect indicated by the determination target side effect phrase is a known side effect in the related drug,
Outputs a determination result as to whether the side effect indicated by the determination target side effect phrase is a known side effect,
In order to evaluate the combination of the phrase included in the comparison target phrase set and the determination target side effect phrase, as a phrase similarity evaluation process,
When the partial character strings of the third length obtained by adding the first length and the second length match the similarity between the partial character strings obtained by dividing each of the two phrases respectively. The evaluation is performed so as to give a higher evaluation than the evaluation obtained by adding the evaluation when the partial character strings of the first length match and the evaluation when the partial character strings of the second length match. ,
Evaluating the combination of division patterns that divide each of the two phrases into one or more partial character strings by counting the similarities evaluated for the partial character strings,
A method including: evaluating a similarity between the two words / phrases using an evaluation of a plurality of combinations of division patterns of the two words / phrases.
(Appendix 12)
The computer further comprises:
Reading the attached document of the determination target drug from the storage unit, performing a phrase extraction process on the effect or effect description part in the attached document to obtain a first efficacy effect phrase set,
For each of the plurality of other medicines, a package insert of the drug is read from the storage means, and a phrase extraction process is performed on a description of the effect or effect in the package insert to obtain a second efficacy effect phrase related to the other drug Get a set,
For at least some of the other pharmaceutical products, respectively
Evaluating a combination of a phrase included in the second efficacy effect phrase set acquired for the other pharmaceutical product and a phrase included in the first efficacy effect phrase set by the phrase similarity evaluation process,
By calculating the evaluation for a plurality of combinations, a value indicating the efficacy or similarity of the effect between the determination target drug and the other drug is calculated,
The calculated value is compared with a second threshold, and when the similarity indicated by the calculated value is higher than the similarity indicated by the second threshold, the other drug is determined as an analog of the determination target drug And performing a first medicine determination process including updating the medicine learning result information on the storage means so as to associate the identification information of the medicine to be judged with the other medicine.
Recognizing the similar drug of the determination target drug according to the determination result of the first similar drug determination process,
The determination method according to supplementary note 11, characterized by:
(Appendix 13)
The computer is
Among the plurality of other medicinal products, a medicinal property classification name, a reference name, a common name, a chemical name or a structural formula described in the package insert are determined to be a similar drug, and the medicinal product is determined as the similar drug, Further executing a second similar drug determination process for updating the similar drug learning result information on the storage means so as to associate the drug determined as the similar drug with the identification information of the determination target drug,
For the drug that has not been determined as the similar drug by the second similar drug determination process among the plurality of other drugs, the first similar drug determination process is performed,
13. The determination method according to appendix 12, wherein the similar drug of the determination target drug is recognized according to both determination results of the second similar drug determination process and the first similar drug determination process.
(Appendix 14)
The phrase similarity evaluation process totals evaluations for all combinations of division patterns that can be divided into arbitrary two characters adjacent to each other in each of the two phrases. The determination method according to any one of appendices 11 to 13, further comprising: calculating the similarity between the two words.
(Appendix 15)
The phrase similarity evaluation processing includes obtaining the similarity between the two phrases using only a division pattern that divides each of the two phrases at a morpheme segmentation position. The determination method according to any one of the above.

DESCRIPTION OF SYMBOLS 100 Determination apparatus 101 Storage part 102 Side effect processing part 103 Similar drug processing part 104 Determination target medicine designation | designated part 105 Similar drug recognition part 106 Report document acquisition part 107 Side effect keyword extraction part 108 Side effect determination / learning part 109 Keyword similarity evaluation part 110 part Character string similarity evaluation unit 111 Efficacy / effect keyword extraction unit 112 Similarity determination / learning unit 113 Preprocessing control unit 201 Attached document group 202 Attached document 203 Synonym dictionary 204 Learning result table 205 Safety information report document 206 Drug to be determined ID
300 Computer 301 CPU
302 ROM
303 RAM
304 Communication Interface 305 Input Device 306 Output Device 307 Storage Device 308 Drive Device 309 Bus 310 Portable Storage Medium 311 Network 312 Other Computer 400 Side Effect Judgment Result Screen 401 Judgment Target Drug Display Field 402 Judgment Result List 403 Radio Button 404 Learning Button 501 Scoring information 502 Reference value information 601 and 602 Keywords 603a to 603c, 604a to 604c Division pattern

Claims (5)

A means for receiving information for identifying a drug, and identifying the drug indicated by the information as a determination target drug;
Report document acquisition means for acquiring a report document describing the side effects of the determination target drug;
By reading from the storage means the identification information that uniquely identifies the drug and the similar drug learning result information from the storage means, or for each of a plurality of drugs, the identification information of the drug and the side effects of the drug And a medicinal product recognition means for recognizing a similar drug to the determination target drug among a plurality of other medicinal products by reading out a package insert including the efficacy or effect of the medicinal product from the storage unit,
To evaluate the similarity between partial character strings included in a phrase, the first length and the second length are added to each other when the first and second partial character strings match. A partial character string similarity evaluation unit that gives a higher evaluation than the evaluation obtained by adding the evaluation when the partial character strings having the same length match each other and the evaluation when the partial character strings having the second length match. ,
By allowing the character string similarity evaluation means to evaluate the similarity between partial character strings obtained by dividing each of the two phrases, and summing up the evaluation results by the character string similarity evaluation means, the 2 Evaluate a combination of division patterns that divide each of the two phrases into one or more partial character strings, and use the evaluation of a plurality of combinations of the division patterns of each of the two phrases to determine the similarity between the two phrases A word similarity evaluation means for evaluating
From the report document, a side effect phrase extracting means for extracting a phrase indicating the side effect of the determination target drug as a determination target side effect phrase;
By reading out a package insert of a medicine recognized as the drug by the drug recognizing means from the storage means and extracting a set of words by a word extraction process from a side effect description part in the package attached, or The medicinal product recognized from the storage means as the medicinal product by a set of words and phrases obtained by the word extraction process from the side effect description part in the package insert of the medicinal product recognized by the medicinal product recognition unit as the medicinal product By reading out the side effect learning result information associated with the identification information, the set of words / phrases included in the description part of the side effect in the package insert of the drug recognized as the similar drug is acquired as a comparison target word / phrase set. A comparison target set acquisition means;
For at least a part of the analogy drug, the phrase similarity evaluation unit evaluates the combination of the word / phrase included in the comparison target word / phrase acquired for the analogy drug and the judgment target side effect word / phrase. A determination means for determining whether or not the side effect indicated by the determination target side effect phrase is a known side effect in the related drug, using the result and the first threshold;
An information processing apparatus comprising: output means for outputting a determination result by the determination means. The attached document of the determination target drug is read from the storage unit, and a phrase extraction process is performed on the indication of the effect or effect in the attached document to obtain a first effect-effect phrase set, and the plurality of other medicines The efficacy and effect of reading the package insert of the drug for each from the storage means and performing the phrase extraction process on the indication of the effect or effect in the package insert to obtain the second set of effect and effect phrases related to the other drug Word extraction means;
For at least some of the other pharmaceutical products, respectively
The phrase similarity evaluation means evaluates a combination of a phrase included in the second efficacy effect phrase set acquired for the other pharmaceutical product and a phrase included in the first efficacy effect phrase set,
By calculating the evaluation for a plurality of combinations, a value indicating the efficacy or similarity of the effect between the determination target drug and the other drug is calculated,
Comparing the calculated value with a second threshold;
When the similarity indicated by the calculated value is higher than the similarity indicated by the second threshold, the other medicine is determined as an analog of the determination target drug, and the identification information of the determination target drug and the other First analog medicine determination means for updating the analog medicine learning result information on the storage means so as to associate the medicines of
Further comprising
The information processing apparatus according to claim 1, wherein the similar drug recognition unit recognizes the similar drug of the determination target drug according to a determination result of the first similar drug determination unit. Among the plurality of other medicinal products, a medicinal property classification name, a reference name, a common name, a chemical name or a structural formula described in the package insert are determined to be a similar drug, and the medicinal product is determined as the similar drug, A second similar medicine determination means for updating the similar medicine learning result information on the storage means so as to associate the medicine determined as the similar medicine with the identification information of the determination target medicine;
The first similar drug determination means determines whether or not it is the similar drug for a drug that has not been determined as the similar drug by the second similar drug determination means among the plurality of other drug drugs,
The said analogy drug recognition means recognizes the said analogy drug of the said determination object drug according to the determination result of both said 2nd analogy drug determination means and said 1st analogy drug determination means. The information processing apparatus described. On the computer,
Accepts information to identify medicines,
The drug indicated by the information is identified as a determination target drug,
Obtain a report document describing the side effects of the determination target drug,
By reading from the storage means the identification information that uniquely identifies the drug and the similar drug learning result information from the storage means, or for each of a plurality of drugs, the identification information of the drug and the side effects of the drug And a medicinal product similar to the determination target drug among a plurality of other medicinal products by reading out the package insert including the efficacy or effect of the medicinal product from the storage unit,
From the report document, the phrase indicating the side effect of the determination target drug is extracted as a determination target side effect phrase,
By reading out a package insert of a medicine recognized as the similar drug from the storage means, and extracting a set of phrases from the side effect description part of the package insert by a phrase extraction process, or from the storage section, the class By reading out the side effect learning result information that associates the set of phrases obtained by the phrase extraction process from the description part of the side effects in the package insert of the drug recognized as a drug with the identification information of the drug recognized as the similar drug, A set of phrases included in the description part of the side effect in the package insert of the drug recognized as the similar drug is obtained as a set of phrases to be compared;
For at least a part of the similar drugs, the combination of the phrase included in the comparison target phrase set acquired with respect to the similar drug and the judgment target side effect phrase is evaluated, and the evaluation result and the threshold value are used. Determining whether the side effect indicated by the determination side effect phrase is a known side effect in the related drug,
Executing a side effect determination process including outputting a determination result of whether or not the side effect indicated by the determination target side effect phrase is a known side effect;
In order to evaluate the combination of the phrase included in the comparison target phrase set and the determination target side effect phrase, as a phrase similarity evaluation process,
When the partial character strings of the third length obtained by adding the first length and the second length match the similarity between the partial character strings obtained by dividing each of the two phrases respectively. The evaluation is performed so as to give a higher evaluation than the evaluation obtained by adding the evaluation when the partial character strings of the first length match and the evaluation when the partial character strings of the second length match. ,
Evaluating the combination of division patterns that divide each of the two phrases into one or more partial character strings by counting the similarities evaluated for the partial character strings,
The determination program which performs the process including evaluating the similarity of two said phrases using evaluation about multiple combinations of the division | segmentation pattern of each said two phrases. Computer
Accepts information to identify medicines,
The drug indicated by the information is identified as a determination target drug,
Obtain a report document describing the side effects of the determination target drug,
By reading from the storage means the identification information that uniquely identifies the drug and the similar drug learning result information from the storage means, or for each of a plurality of drugs, the identification information of the drug and the side effects of the drug And a medicinal product similar to the determination target drug among a plurality of other medicinal products by reading out the package insert including the efficacy or effect of the medicinal product from the storage unit,
From the report document, the phrase indicating the side effect of the determination target drug is extracted as a determination target side effect phrase,
By reading out a package insert of a medicine recognized as the similar drug from the storage means, and extracting a set of phrases from the side effect description part of the package insert by a phrase extraction process, or from the storage section, the class By reading out the side effect learning result information that associates the set of phrases obtained by the phrase extraction process from the description part of the side effects in the package insert of the drug recognized as a drug with the identification information of the drug recognized as the similar drug, A set of phrases included in the description part of the side effect in the package insert of the drug recognized as the similar drug is obtained as a set of phrases to be compared;
For at least a part of the similar drugs, the combination of the phrase included in the comparison target phrase set acquired with respect to the similar drug and the judgment target side effect phrase is evaluated, and the evaluation result and the threshold value are used. Determining whether the side effect indicated by the determination side effect phrase is a known side effect in the related drug,
Outputs a determination result as to whether the side effect indicated by the determination target side effect phrase is a known side effect,
In order to evaluate the combination of the phrase included in the comparison target phrase set and the determination target side effect phrase, as a phrase similarity evaluation process,
When the partial character strings of the third length obtained by adding the first length and the second length match the similarity between the partial character strings obtained by dividing each of the two phrases respectively. The evaluation is performed so as to give a higher evaluation than the evaluation obtained by adding the evaluation when the partial character strings of the first length match and the evaluation when the partial character strings of the second length match. ,
Evaluating the combination of division patterns that divide each of the two phrases into one or more partial character strings by counting the similarities evaluated for the partial character strings,
A method including: evaluating a similarity between the two words / phrases using an evaluation of a plurality of combinations of division patterns of the two words / phrases.

Images