JP7402008B2 - Disease name inference system, disease name inference method, disease name inference program, and data structure - Google Patents

Description

Application of Article 30, Paragraph 2 of the Patent Act 1. [Publisher name] EM Systems Co., Ltd. [Publication name] Press conference information for ``MAPs for PHARMACY'', a business support system for pharmacies [Date of publication] April 16, 2019 2. [Meeting name] Press conference announcing “MAPs for PHARMACY” [Venue] EM Systems Co., Ltd. Presentation Room, Kyodo Building 2F, 3-6 Kanda Nishikicho, Chiyoda-ku, Tokyo [Date] May 15, 2019 3. [Publisher name] Yakuji Nipposha Co., Ltd. [Publication name] Yakuji Nippo page 8 [Date of publication] June 5, 2019

The present invention relates to a disease name inference system, a disease name inference method, and a disease name inference program, and particularly to a disease name inference system, a disease name inference method, a disease name inference program, and a disease name inference system, a disease name inference method, and a disease name inference program that infer a disease name from prescription data using machine learning. Regarding data structures.

Patients who receive treatment at a hospital or clinic bring a prescription issued by the hospital or clinic to a pharmacy, receive the medicine based on the prescription, and receive explanations about dosage and medication instructions from the pharmacist.

However, prescriptions issued by hospitals and clinics do not include the name of the patient's injury or disease, making it difficult to provide appropriate medication guidance according to the patient's injury or disease. For example, if a drug has side effects, the explanation of the side effects differs depending on the injury or disease, and the precautions regarding dosage and administration also differ, but it is difficult to know the name of the injury or disease from the prescription, and it is difficult to provide appropriate medication guidance. Ta.

In view of such problems, a pathological condition estimation database has been disclosed in which a prescription database, a drug database, and an injury and disease database can be searched in association with each other (Patent Document 1).

Japanese Patent Application Publication No. 2004-185196

However, in the pathological condition estimation database shown in Patent Document 1, it is only possible to search from databases that are associated with each other, and it is not possible to infer the name of injury or disease from prescription data with high accuracy. For example, when multiple drugs are prescribed, verification is performed for each drug, so it is not possible to infer the name of injury or disease with high accuracy by comprehensively considering the multiple drugs.

In addition, in the pathological state estimation database shown in Patent Document 1, by storing a patient information database that stores each patient's constitution, medical history, drug history, etc., it is possible to provide medication guidance to the patient. The condition table only displays the severity of the patient's symptoms and cannot infer the name of the disease or disease from the patient's constitution.

The disease name inference system of the present invention includes a processor and a storage device, the processor infers the name of the disease, and the processor infers the name of the disease and the prescription data. The apparatus includes a machine learning unit that generates a learned model by machine learning based on learning data, and an inference unit that infers the name of the injury or disease from prescription data for inference input to the learned model.

According to the present invention, since the name of injury or disease is comprehensively inferred from prescription data by machine learning, the name of injury or disease can be inferred with high accuracy.

FIG. 1 is a block diagram showing an example of a system configuration of a disease name inference system according to the present embodiment. FIG. 1 is a block diagram showing an example of a system configuration of an artificial intelligence server according to the present embodiment. It is a block diagram showing an example of the system configuration of the learning database server of this embodiment. FIG. 3 is a diagram showing an example of data stored as learning data. It is a figure explaining the data for learning expanded by classifying at least one of prescription data into a plurality of groups. 2 is a flowchart illustrating an example of a disease name inference method according to the present embodiment. FIG. 3 is a sequence diagram illustrating an example of operations for acquiring and preprocessing various data, and generating and storing a learned model. FIG. 3 is a sequence diagram illustrating an example of operations of inference using a trained model and determination of an inference result. FIG. 3 is a diagram showing a first example of an inference result displayed on a computer display. FIG. 6 is a diagram showing a second example of the inference result displayed on the computer display. FIG. 7 is a diagram illustrating a third example of inference results displayed on a computer display.

A disease name inference system according to an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing an example of the system configuration of the injury/illness name inference system of this embodiment. The injury/illness name inference system 1 infers the injury/illness name.

As shown in FIG. 1, the disease name inference system 1 includes an artificial intelligence server 3, a learning database server 4, and a computer (PC) 5, which are electrically connected via a network 2 and can communicate with each other.

FIG. 2 is a block diagram showing an example of the system configuration of the artificial intelligence server of this embodiment. As shown in FIG. 2, the artificial intelligence server 3 includes a processor 21, a storage device (for example, ROM, RAM, HDD, etc.) 22, an input device 23, an interface 24, and an output device 25.

The processor 21 is a control device such as a CPU, MPU, or GPU, and includes a data acquisition section 32, a machine learning section 34, a learned model storage section 35, and an inference section 36. The data acquisition section 32, preprocessing section 33, machine learning section 34, learned model storage section 35, and inference section 36 are electrically connected by a bus (not shown) and can communicate with each other.

FIG. 3 is a block diagram showing an example of the system configuration of the learning database server of this embodiment. As shown in FIG. 3, the learning database server 4 includes a processor 37, a learning database 41, a storage device (for example, ROM, RAM, HDD, etc.) 42, an input device 43, an interface 44, and an output device 45. The processor 37 is a control device such as a CPU, MPU, or GPU, and includes a data storage section 38, a data acquisition section 39, and a preprocessing section 40. The learning database 41 includes medical record data 46, prescription data 48, and learning data 80.

The medical record data 46 is medical record data obtained from a hospital or clinic, and includes the injury/disease name data 47 of the patient for whom a prescription has been issued, including the injury/disease name, prescribed medicine, and patient attributes (name, insurer number, and insurance card). (symbols, numbers, etc.) (not shown). The prescription data 48 includes prescription drugs and patient attribute data 49 (name, insurer number, symbol/number of insured person's card, etc.). By associating the patient attributes of the medical record data 46 with the patient attribute data 49 of the prescription data 48, the injury/illness name data 47 and the prescription data 48 are associated for each injury/illness name or prescription. The associated disease name data 47 and prescription data 48 are stored in the learning database 41 as learning data 80. The learning data 80 is data used for machine learning by the machine learning unit 34 of the artificial intelligence server 3.

In addition, in FIG. 3, the medical record data 46 and the prescription data 48 include drug data (drug name and drug code) of prescription drugs, but instead of the medical record data 46 and the prescription data 48 including drug data, the medical record data 46 and the prescription data 48 46 and the prescription data 48, drug data may be stored in the learning database 41, and the drug data may be associated with the medical record data 46 and the prescription data 48, respectively. Further, the prescription data 48 includes efficacy/effect data of prescribed drugs, but instead of the prescription data 48 including efficacy/effect data, the efficacy/effect data is stored in the learning database 41 separately from the prescription data 48. The efficacy/effect data may be associated with drug data of the prescription data 48 or with drug data different from the prescription data 48.

FIG. 4 is a diagram showing an example of data stored as learning data 80. As shown in FIG. 4, the injury/illness name data 47 and the prescription data 48 are associated with each other using an ID 50 (patient's name, insurer number, symbol/number of an insured card, etc.) that identifies the patient, prescription, etc. The data acquisition unit 39 acquires injury/illness name data 47 and prescription data 48 . The preprocessing unit 40 associates the injury/illness name data 47 with the prescription data 48 based on the ID of the patient's name, insurer number, symbol/number of the insured person's card, etc. in the medical record data 46 . The associated data is stored in the learning database 41 as learning data 80.

Based on the injury/illness name data 47 of the medical record data 46, the injury/illness name data 51 is stored as learning data 80. Based on prescription data 48, drug name (drug name) 52, drug code 53, efficacy/effect 54, usage/dose 55, medical treatment date 56, medical department 57, and patient attribute 58 are stored as learning data 80. . In this way, the prescription data 52 to 58 including patient attribute data (patient attributes) 58 are associated with the injury/illness name data 51 for each injury/disease name or prescription, and the learning data 80 is generated.

In FIG. 4, regarding the learning data 80 whose ID 50 is "1", the injury/disease name data 51 "hypercholesterolemia" is associated with a plurality of prescription drugs (drug names (drug names) 52 or drug codes 53). This is an example of one prescription containing multiple prescription drugs.

Dosage/Administration 55 includes the number of days the prescription drug is prescribed (the number of days it is taken), the number of times it is prescribed (the number of times it is taken), the time of taking the drug (before meals, after meals, between meals, etc.), and the route/site of administration (drop into the right eye, instillation into the affected area). Application, etc.), dosage, etc. are stored as data, and the dosage field stores as data the dosage for one day, one dose, or one day's worth of dosage, and the number of times of administration per day.

In this embodiment, the usage/dose 55 stores the number of prescription days, number of prescriptions, and prescription type of the prescription drug. For example, the prescription type is classified into internal drops, internal use, infusion, hot water, temporary use, external use, injection, self-pay, equipment, and the like.

As patient attributes 58, the patient's age, sex, weight, pregnancy status, past medical conditions, history of side effects, allergy history, concomitant medications, lifestyle, preferences, and family medical history are stored. In this way, attribute data (patient attributes) 58 including at least one of the patient's age, gender, weight, pregnancy status, past medical conditions, history of side effects, history of allergies, concomitant medications, lifestyle, preferences, and family medical history are collected. The learning data 80 is expanded by associating it with the injury/illness name data 51.

FIG. 5 is a diagram illustrating learning data 80 expanded by classifying at least one of prescription data 52 to 58 into a plurality of groups.

As shown in FIG. 5(a), the learning data 80 is expanded by classifying the age 64 of the attribute data (patient attributes) 58 into multiple groups (newborn 65, infant 66, child 67, elderly 68). Ru. Furthermore, weight may be classified into multiple groups. Furthermore, lifestyle habits (frequency of drinking and smoking, sleeping hours, etc.) may be classified into a plurality of groups. In this way, by classifying the attribute data (patient attributes) 58 into a plurality of groups, the learning data 80 is expanded.

As shown in FIG. 5(b), the learning data 80 is expanded by classifying the prescription days 70 into a plurality of groups (not applicable=2, acute=1, and chronic=0). If data representing prescription type, such as occasional use or injection, is stored, or if data representing prescription days (number of days for taking medication) is not stored in usage, data "0" representing "not applicable" is stored. . In addition, data representing the number of prescription days (number of days for taking medication) is stored in the usage, and if the number of prescription days (number of days for taking medication) is less than a predetermined number of days (for example, 14 days), data ``1'' representing ``acute'' is set. If the number of days is greater than or equal to a predetermined number of days (for example, 14 days), data "2" representing "chronic" is stored. The predetermined number of days can vary depending on prescription drugs and the like. Further, the learning data 80 may be expanded by classifying the number of prescriptions into a plurality of groups.

As shown in FIG. 5C, the learning data 80 is expanded by classifying the medical treatment days 56 into a plurality of groups (season or quarter 61, month 62, and day of the week 63). The learning data 80 is expanded by using data that classifies the medical treatment days 56 into at least one of seasons, quarters, months, and days of the week.

As shown in FIG. 5(d), the learning data 80 is expanded by classifying the medical department 57 into a plurality of groups (a plurality of medical treatment divisions 69). When the predetermined standard medical department name partially matches the medical department name of the medical department 57, the learning data 80 is expanded by setting the medical department name for each matching part and classifying it into multiple groups. be done. As the standard medical department name, the Ministry of Health, Labor and Welfare's medical department classifications without "department" may be used, or other clinical department classifications compiled in a database may be used. The medical departments classified by the Ministry of Health, Labor and Welfare include internal medicine (department), non-orthopedics (department), ophthalmology (department), psychosomatic (department), non-plastic (department)), otorhinolaryngology (department), and psychiatry (department). There is a division. For example, as shown in FIG. 5(d), if the medical department 57 is "Neurology", the reference medical department names "Internal Medicine" and "Psychiatry (Department)" partially match "Neurology". Therefore, "neurology" is classified into multiple groups "internal medicine" and "psychiatry", and the medical treatment category 69 stores data representing "internal medicine" and "psychiatry".

Further, the learning data 80 may be expanded by using a drug code as prescription drug data and extracting a part of the drug code to classify it into a plurality of groups. As the drug code, a drug price list drug code or a YJ code may be used. When using the YJ code, training data can be created by classifying the YJ code into first 1 digit, first 2 digits, first 3 digits, first 4 digits, first 5 to 7 digits, first 8 digits, and first 9 digits. 80 may be expanded. For example, the YJ code for 60 mg/1 tablet of loxoprofen sodium hydrate tablet is "1149019F1242", so "1", "11", "114", "1149", "019", "F", and By being classified as "1149019F1", the learning data 80 is expanded.

In this way, by classifying the prescription data 52 to 58 into a plurality of groups, the learning data 80 is expanded.

FIG. 6 is a flowchart illustrating an example of the disease name inference method according to the present embodiment. 7 and 8 are sequence diagrams showing examples of operations executed by the disease name inference program of this embodiment. FIG. 7 is a sequence diagram illustrating an example of operations for acquiring and preprocessing various data, and generating and storing a trained model. FIG. 8 is a sequence diagram illustrating an example of operations for inference using a learned model and determining an inference result. The injury/illness name inference program may be installed on a recording medium.

As shown in FIGS. 6 and 7, in step S1, the processor (not shown) of the computer 5 transmits an instruction to store various data to the learning database server 4 (110), and stores the data in the learning database server 4. The storage unit 38 executes a command to store various data (111), and stores various data input from the computer 5 or the input device 43 in the learning database 41. The various data stored include medical record data 46, disease name data 47, prescription data 48, and attribute data 49.

The processor of the computer 5 transmits various data acquisition commands and preprocessing commands to the learning database server 4 (112). In step S2, the data acquisition unit 39 of the learning database server 4 executes various data acquisition commands (113) and acquires various data from the learning database 41. In step S3, the preprocessing unit 40 of the learning database server 4 executes a command for associating various data (113), and associates the injury/illness name data 47 and the prescription data 48 as the learning data 80.

In step S4, the data storage unit 38 executes the instruction to store the learning data 80 (115), and stores the learning data 80 in the learning database 41. Further, the preprocessing unit 40 executes an extension instruction for the learning data 80 (115) to extend the learning data 80. The preprocessing unit 40 expands the learning data 80 by associating the patient attribute data (patient attributes) 58 with the prescription data 52 to 57. The preprocessing unit 40 generates attribute data (patient attributes) including at least one of the patient's age, gender, weight, pregnancy status, past medical conditions, side effect history, allergy history, concomitant medications, lifestyle, preferences, and family medical history. 58 with the prescription data 52 to 57, the learning data 80 is expanded.

Further, the preprocessing unit 40 expands the learning data 80 by classifying at least one of the prescription data 52 to 57 and the attribute data (patient attributes) 58 into a plurality of groups. The preprocessing unit 40 classifies, in addition to prescription drugs, at least one of the efficacy, effect, usage, dose, prescription type, medical treatment date, medical department name, and patient attribute data of the prescription drug. Data 80 is expanded. For example, the preprocessing unit 40 expands the learning data 80 by classifying the age 64 of the attribute data (patient attributes) 58 into a plurality of groups (newborn 65, infant 66, child 67, and elderly 68). The preprocessing unit 40 expands the learning data 80 by classifying the prescription days 70 into a plurality of groups (not applicable=2, acute=1, and chronic=0). The preprocessing unit 40 expands the learning data 80 by using data that classifies the medical treatment days 56 into at least one of seasons, quarters, months, and days of the week. The preprocessing unit 40 expands the learning data 80 by classifying the clinical department 57 into a plurality of groups (a plurality of medical treatment divisions 69). The preprocessing unit 40 expands the learning data 80 by using the drug code as prescription drug data and extracting a part of the drug code.

In step S5, the preprocessing unit 40 executes a filter processing command for the learning data 80 (116), and performs filter processing for each ID50. If the prescription data 52 to 58 are inconsistent with the disease name data 51 or other prescription data 52 to 58, or if a predetermined condition is satisfied, the preprocessing unit 40 replaces the disease name data 51 or the prescription data 52 to 58 with the ID 50. Filter processing is performed by excluding the data 80 from the learning data 80. Note that the preprocessing unit 40 may exclude only contradictory data portions. As a result, the name of the injury or disease can be inferred with high accuracy.

The preprocessing unit 40 performs processing when the group of prescription data 52 to 58 is inconsistent with at least one of the injury/disease name data, the prescription drug, the efficacy, effect, usage, dosage, and prescription type of the prescription drug, or when a predetermined condition is satisfied. , filter processing is performed by excluding the injury/illness name data 51 or the prescription data 52 to 58 with the ID 50 from the learning data 80.

The age 64 of the attribute data (patient attributes) 58 is classified into multiple groups (newborn 65, infant 66, child 67, elderly 68), the learning data 80 is expanded, and the classification is changed to the injury/disease name data 51. , efficacy/effect 54, and usage/dose 55, or when a predetermined condition is satisfied, the preprocessing unit 40 converts the disease name data 51 or the prescription data 52 to 58 into the learning data 80. Exclude from. For example, if the age 64 of a predetermined ID50 is classified as a newborn baby 65, but the efficacy/effect 54 includes injury/disease name data 51 related to pregnant women, the preprocessing unit 40 /Exclude the injury/illness name data 51 regarding pregnant women from the effect 54.

In addition, the number of prescription days 70 is classified into multiple groups (not applicable = 2, acute = 1, and chronic = 0), the learning data 80 is expanded, and the classification is changed to injury/disease name data 51, efficacy/ If it contradicts at least one of the effect 54 and the usage/dose 55, or if a predetermined condition is satisfied, the preprocessing unit 40 converts the injury/disease name data 51 or the prescription data 52 to 58 of the ID 50 into the learning data 80. Exclude from. For example, if the prescribed number of days of prescription for ID50 is 70, it is classified as "chronic = 0", but if the injury/disease name data 51 includes character string data of "acute" such as "acute gastritis", if the The processing unit 40 excludes the classification data with ID 50 and prescription days 70.

Furthermore, the learning data 80 has been expanded by classifying gender into multiple groups (male = 1 and female = 2), and the classification has been changed to injury/disease name data 47, efficacy/effect 54, and usage/dose 55. If it contradicts at least one of the following, or if a predetermined condition is satisfied, the preprocessing unit 40 removes the injury/illness name data 51 or the prescription data 52 to 58 of the ID 50 from the learning data 80. For example, if the gender of a predetermined ID50 is classified as "female = 2", but the efficacy/effect 54 includes data on the name of injury or disease related to the prostate, the preprocessing unit 40 Injury and disease name data 51 related to the prostate gland is excluded from the effect 54.

Further, if the disease name data for the drug name or drug code includes a character string that satisfies a predetermined condition, the preprocessing unit 40 learns the disease name data 51 or the prescription data 52 to 58 of the ID 50. data 80. For example, if a stomach medicine is prescribed to prevent stomach irritation caused by analgesics, and the injury/disease name data includes the character string "chronic gastritis," the purpose of the prescription is different from the injury/disease name. The processing unit 40 determines that a combination of analgesic and gastric medicine has been prescribed, and excludes data on the name of injury or disease that includes the character string "chronic gastritis" from the name data 51 of the name of injury or disease with ID 50.

In this way, by performing machine learning while excluding inappropriate data combinations, it is possible to infer the name of injury or disease with high accuracy. The contradictory data combinations may be automatically generated based on the injury/illness name data 47 or the prescription data 48, or may be stored in the learning database 41 as a previously created database.

In step S6, a trained model is generated. The processor of the computer 5 sends a machine learning model generation command to the artificial intelligence server 3 (120). The data acquisition unit 32 of the artificial intelligence server 3 transmits a command to acquire the learning data 80 to the learning database server 4 via the interface 24 (121), and the learning database server 4 transmits the learning data 80. The instruction is executed and the learning data 80 stored in the learning database 41 is transmitted to the artificial intelligence server 3 via the interface 44 (122). The data acquisition unit 32 acquires the learning data 80 from the learning database 41, and the machine learning unit 34 acquires the learning data (including expanded learning data) in which the injury/disease name data 47 and the prescription data 48 are associated. ) 80, a machine learning trained model generation/storage instruction is executed (123).

The machine learning unit 34 generates a learned model using a neural network that inputs the prescription data 52 to 58 into the input layer and inputs the disease name data 51 as the correct value in the output layer. An affine layer or a convolution layer is provided as an intermediate layer of the neural network. Downsampling processing or the like may be performed as appropriate. Further, the number of layers, the number of neurons, and the activation function of the intermediate layer are optimally selected so that the inference result is highly accurate. As the neural network, a feed forward neural network (FFNN), a recurrent neural network (RNN), etc. are used. The machine learning unit 34 uses parameters (weights) initialized with predetermined values such as random numbers to compare the values output to the output layer and the output layer when the prescription data 52 to 58 are input to the input layer. A loss function representing the deviation from the correct value (injury/illness name data 51) is calculated, and the differential value of the loss function is used as a gradient to reduce the deviation between the value output to the output layer and the correct value of the output layer. A trained model is generated by changing parameters (weights).

Since the medicine is prescribed according to the injury or disease, there is a certain relationship between the prescribed medicine in the prescription data 52 and 53 and the name of the injury or disease. Therefore, by generating a trained model that infers the name of injury or disease using prescription drugs as learning data, it is possible to infer the name of injury or disease with high accuracy. In particular, when multiple drugs are prescribed in one prescription, the name of the disease or illness can be inferred with higher accuracy by machine learning the combination of the multiple prescription drugs. That is, in the learning data 80, at least a plurality of drug names 52 or a plurality of drug codes 53 need only be associated with the disease name data 47.

In addition, the indications for drugs are determined by the efficacy and effects of the prescribed drug, and the usage and dosage of prescription drugs change depending on the injury and disease, so the prescription data 54 to 57 include efficacy, effects, usage, and dosage. There is a certain relationship between the type of prescription, the date of treatment, the name of the department, and the name of the disease. For example, even if the same antibiotic is prescribed by an ophthalmologist, it is more likely to be a stye, and if it is prescribed by a gynecologist, it is more likely to be caused by cystitis.

In addition, because some injuries and illnesses are seasonal (for example, influenza is prevalent from January to February), and the patient's behavioral characteristics may be affected depending on the injury or illness, the date of treatment56 and the name of the injury or illness may be changed. have a certain relationship. Therefore, in addition to prescription drugs, prescription data including at least one of the prescription drug's efficacy, effect, usage, dose, prescription type, medical treatment date, and medical department name is used as the learning data 80 for learning to infer the name of injury or disease. By generating a complete model, it is possible to comprehensively infer the name of an injury or disease with high accuracy.

In addition, even if the same prescription drug is used, it will depend on the patient's attribute data such as age, gender, weight, pregnancy status, past medical conditions, history of side effects, allergy history, concomitant medications, lifestyle habits, preferences, and family medical history. In some cases, the name of the injury or disease may be different. Therefore, by generating a trained model for inferring the name of injury or disease using the patient's attribute data 58 as learning data 80 in addition to prescription drugs, the name of injury or disease can be inferred with high accuracy.

In step S7, the trained model storage unit 35 stores the generated trained model in the storage device 22 by executing a command to store the trained model (123).

In step S8, if the process proceeds to the step of inferring the name of the injury or disease, the process proceeds to step S9; if the process does not proceed to the step of inferring the name of the injury or disease, the process ends.

As shown in FIGS. 6 and 8, in step S9, the processor of the computer 5 inputs prescription data for inference (including patient attribute data) through the input device 23, and inputs prescription data for inference (including patient attribute data). data) to the artificial intelligence server 3 (129). Then, the processor of the computer 5 transmits an instruction for inferring the name of the injury or disease to the artificial intelligence server 3 (130).

In step 10, the inference unit 36 executes a data check command, checks the data format of the inference prescription data (including patient attribute data) sent from the computer 5, and determines whether the data format is correct. is determined (131). If the data format is incorrect, the inference unit 36 transmits data indicating this to the computer 5.

If the data format is correct, in step S11, the inference unit 36 executes a data preprocessing instruction and encodes prescription data for inference (including patient attribute data) according to the encoding policy at the time of learning (131). . If a value that is not used during learning is included in the inference prescription data (including patient attribute data), the inference unit 36 either excludes the value or assigns a predetermined value according to predetermined conditions. . For example, if the upper limit of the patient's age 64 in the learning data 80 is "100 years old" and the patient's age for inference is "103 years old", the inference unit 36 calculates "103 years old" from the prescription data for inference. Either the age data of ``years old'' is excluded, or the upper limit of the learning data 80, ``100 years old,'' is assigned instead of ``103 years old.''

In step S12, the inference unit 36 executes the inference command, reads out the learned model stored in the storage device 22, inputs prescription data for inference (including patient attribute data) into the learned model, and The disease name is inferred from the prescription data (including patient attribute data) input into the completed model (132).

In this case, when the inferred disease name is inconsistent with the prescription data (including patient attribute data) input to the learned model, the inference unit 36 excludes the contradictory disease name from the inference result. The inference unit 36 generates a mask for the output so that contradictory injury/illness names are not output. For example, if the age of patient attribute data input for inference is "72 years old," names of injuries and illnesses related to newborns, infants, and children are masked and excluded. Furthermore, if the gender of the patient attribute data input for inference is "male," the name of the disease related to women's diseases is masked and excluded. As a result, the name of the injury or disease can be inferred with high accuracy.

The inference unit 36 executes the inference result transmission command via the interface 24 and transmits the inference result to the computer 5 (133).

In step S13, the processor of the computer 5 executes a transmission result display command to display the inference result on a display or the like (134).

9 to 11 are diagrams showing examples of inference results displayed on the display (not shown) of the computer 5. As shown in FIGS. 9 to 11, the screens 500, 600, and 700 include patient attribute areas 501, 601, and 701, clinical department/medical treatment date areas 502, 602, and 702, prescription drug/usage/dose areas 505, 605, 705, upper inference result area 506, 606, 706, all display instruction area 507, 607, 707, all inference result area 508, past data area 509, 609, 709, selection area 510, 610, 710, pharmacy/pharmacist Areas 511, 611, 711 and information update areas 512, 612, 712 are displayed. Note that Amlozin, Crestor, Valtrex, and Flomox are registered trademarks.

In the medical department/medical day area 502, 602, 702, the medical institution, medical department (medical department), and doctor name are displayed in the medical institution area 503, 603, 703, and the medical day is displayed in the medical day area 504, 604, 704. Prescription drug/usage/dose areas 505, 605, and 705 display the drug name, usage, and dose. In the upper inference result areas 506, 606, and 706, the top three injury and disease names with high inference probabilities among the inference results are displayed. When the all-display instruction area 507 is designated with a cursor or the like, all inference results (10 results) are displayed in the all-inference result area 508. The past data areas 509 and 609 display the medical institution, department (medical department), doctor name, drug name, usage, and dose in past prescriptions. Past prescriptions can be selected in selection areas 510, 610, and 710. In FIGS. 9 to 11, past prescriptions with matching medical institutions and medical departments are selected. Pharmacy/pharmacist areas 511, 611, and 711 display pharmacy names and pharmacist names. When the information update area 512, 612, 712 is specified with a cursor or the like, the information displayed on the screen 500, 600, 700 is updated.

In both FIGS. 10 and 11, "Valtrex Granules 50%" is prescribed, but the name of the disease or disease resulting from the inference differs depending on the usage, dosage, and patient attributes (age, etc.). In this way, by generating a trained model using machine learning based on the learning data 80 in which prescription data 48 including drug names and injury/disease name data 47 are associated, injury/disease names can be determined from prescription data with high accuracy. can be inferred.

In step S14, the input device (not shown) of the computer 5 inputs the correct injury or disease name, and the processor of the computer 5 executes an instruction to determine the inference result (135). For example, when a pharmacist asks a patient, the correct name of the disease or injury is input from an input device (not shown) of the computer 5. In this case, the correct name of the disease or disease may be input by specifying the name of the disease or disease displayed in the upper inference result area 506 or the total inference result area 508 with a cursor or the like.

In step S15, the computer determines whether or not the correct injury/illness name has been stored as injury/illness name data 51 in the learning data 80 of the learning database 41 in association with inference prescription data (including patient attribute data). 5 decides. If it is not stored in the learning data 80 of the learning database 41, the processor (not shown) of the computer 5 learns data in which the correct disease name and prescription data (including patient attribute data) are associated. The data storage unit 38 of the learning database server 4 executes the instruction to store the learning data (137), and stores the data sent from the computer 5 in the learning database 41. . This increases the amount of learning data 80 in the learning database 41, making it possible to infer the name of injury or disease with even higher accuracy.

If data in which the correct disease name and prescription data (including patient attribute data) are associated is stored in the learning data 80 of the learning database 41, the process ends.

As described above, according to the present embodiment, it is possible to infer the name of an injury or disease from prescription data with high accuracy, it is possible to understand the name of an injury or disease even from a prescription that does not include the name of an injury or disease, and it is possible to provide appropriate medication guidance. It can be performed.

Although the embodiments according to the present invention have been described above, the present invention is not limited to these, and can be modified and modified within the scope of the claims.

This embodiment has a data structure of learning data 80 in which prescription data 52 to 58 and injury/disease name data 51 are associated, and the prescription data 52 to 58 that can be input to the input layer of the neural network and the neural network The learning data 80 includes prescription data 52 to 58 input to the input layer using parameters (weights) initialized with predetermined values. In order to generate a trained model by changing the parameters (weights) so that the deviation becomes smaller, using the differential value of the loss function that represents the deviation between the value output to the output layer and the correct value when Data structures characterized in that they are used are included. In this case, the learning data 80 includes, in addition to prescription drugs, prescription data 52 to 57 including at least one of the prescription drug's efficacy, effect, usage, dose, prescription type, medical treatment date, and medical department name, and the name of the disease. It may be a data structure characterized by being associated with data 51.

INDUSTRIAL APPLICABILITY The present invention is useful as a disease name inference system that can infer the name of a disease or disease with high accuracy because it comprehensively infers the name of a disease or disease from prescription data using machine learning. In particular, it is useful as a disease/injury name inference system that can infer the name of a disease/injury with high accuracy by comprehensively considering a plurality of drugs.

1...Injury and disease name inference system 3...Artificial intelligence server 4...Learning database server 5...Computer 32...Data acquisition section 33, 40...Preprocessing section 34...Machine learning section 35...Model storage section 36...Inference section 38...Data storage Section 39...Data acquisition section 41...Learning database 46...Medical record data 47...Injury and disease name data 48...Prescription data 49...Attribute data 80...Learning data

Claims (19)

An injury/illness name inference system including a processor and a storage device, wherein the processor infers an injury/illness name,
The processor includes:
Based on the learning data that is expanded by associating the prescription drug of the prescription data with the injury/disease name data and classifying at least one of the number of prescription days, the number of prescriptions, and the medical treatment date of the usage of the prescription drug into a plurality of groups. A machine learning unit that generates a trained model by machine learning,
an inference unit that infers the name of the injury or disease from the inference prescription data input to the trained model;
An injury/illness name inference system comprising: An injury/illness name inference system including a processor and a storage device, wherein the processor infers an injury/illness name,
The processor includes:
Machine learning is performed based on learning data that is expanded by using data that associates prescription drugs in prescription data with injury and disease name data and classifies medical treatment days into at least one of seasons, quarters, months, and days of the week. a machine learning unit that generates a trained model;
an inference unit that infers the name of the injury or disease from the inference prescription data input to the trained model;
An injury/illness name inference system comprising: An injury/illness name inference system including a processor and a storage device, wherein the processor infers an injury/illness name,
The processor includes:
When prescription drugs in prescription data and injury/illness name data are associated, and a predetermined standard medical department name partially matches the medical department name, the medical department name can be set for each matching part to classify it into multiple groups. a machine learning unit that generates a trained model by machine learning based on the training data expanded by
an inference unit that infers the name of the injury or disease from the inference prescription data input to the trained model;
An injury/illness name inference system comprising: An injury/illness name inference system including a processor and a storage device, wherein the processor infers an injury/illness name,
The processor includes:
Prescription drugs in prescription data and injury/illness name data are associated, a drug code is used as the prescription drug data, a part of the drug code is extracted, and the data is classified into multiple groups for expanded learning purposes. a machine learning unit that generates a trained model by machine learning based on the data;
an inference unit that infers the name of the injury or disease from the inference prescription data input to the trained model;
An injury/illness name inference system comprising: In addition to the prescription drug, the machine learning unit is configured to perform the learning in which at least one of the efficacy, effect, usage, dose, prescription type, medical treatment date, and medical department name of the prescription drug is associated with the injury/illness name data. The injury/disease name inference system according to any one of claims 1 to 4, wherein the learned model is generated by the machine learning based on the data. The machine learning unit generates the learned model by the machine learning based on the learning data expanded by associating the prescription data including patient attribute data with the disease name data,
6. The inference unit infers the disease name by inputting the prescription data for inference including patient attribute data for inference into the learned model. The injury/illness name inference system described in item 1. The machine learning unit analyzes the attribute data including at least one of the patient's age, gender, weight, pregnancy status, past medical conditions, side effect history, allergy history, concomitant medications, lifestyle, preferences, and family medical history. 7. The injury/illness name inference system according to claim 6, wherein the learned model is generated by the machine learning based on the learning data expanded by associating with injury/illness name data. The machine learning unit is configured to use the prescription data or the injury/illness name data for the learning when the prescription data is inconsistent with the injury/illness name data or other prescription data, or when a predetermined condition defining the contradiction is satisfied. The injury/illness name inference system according to any one of claims 1 to 7, wherein the learned model is generated by the machine learning without using it as data. The machine learning unit is configured to detect a case where the group of prescription data is inconsistent with at least one of the injury/disease name data, the prescription drug, the efficacy, effect, usage, dosage, and prescription type of the prescription drug, or a predetermined method that defines the contradiction. 9. If the following condition is satisfied, the learned model is generated by the machine learning without using the prescription data or the injury/illness name data as the learning data. The injury/illness name inference system described in item 1. The disease name inference according to any one of claims 1 to 9, wherein the machine learning unit generates the learned model by a neural network using the prescription data as an input layer and the disease name data as an output layer. system. When the inferred injury/illness name is inconsistent with the patient attribute data of the prescription data input to the learned model or satisfies a predetermined condition that defines the inconsistency, The injury/illness name inference system according to any one of claims 1 to 10, wherein the injury/illness name is excluded from the inference results. A disease name inference method for inferring a disease name using a disease name inference system in which a processor infers a disease name, the method comprising:
The prescription drug of the prescription data and the injury/illness name data are associated with each other, and the training data is expanded by categorizing at least one of the number of days of prescription, number of prescriptions, and medical treatment date of the usage of the prescription drug into a plurality of groups. a step of generating a trained model by machine learning based on the
inferring the name of the injury or disease from the prescription data input to the learned model;
A method for inferring names of injuries and illnesses. A disease name inference method for inferring a disease name using a disease name inference system in which a processor infers a disease name, the method comprising:
Machine learning is performed based on learning data that is expanded by using data that associates prescription drugs in prescription data with injury and disease name data and classifies medical treatment days into at least one of seasons, quarters, months, and days of the week. generating a trained model;
inferring the name of the injury or disease from the prescription data input to the learned model;
A method for inferring names of injuries and illnesses. A disease name inference method for inferring a disease name using a disease name inference system in which a processor infers a disease name, the method comprising:
When prescription drugs in prescription data and injury/illness name data are associated, and a predetermined standard medical department name partially matches the department name, the department name can be set for each matching part to classify it into multiple groups. a step of generating a trained model by machine learning based on the training data expanded by
inferring the name of the injury or disease from the prescription data input to the learned model;
A method for inferring names of injuries and illnesses. A disease name inference method for inferring a disease name using a disease name inference system in which a processor infers a disease name, the method comprising:
Prescription drugs in prescription data and injury/illness name data are associated, a drug code is used as the prescription drug data, a part of the drug code is extracted, and the data is classified into multiple groups for expanded learning purposes. generating a trained model by machine learning based on the data;
inferring the name of the injury or disease from the prescription data input to the learned model;
A method for inferring names of injuries and illnesses. A disease name inference program executed on a computer that infers a disease name using a disease name inference system in which a processor infers a disease name,
The computer,
The prescription drug of the prescription data and the injury/illness name data are associated with each other, and the training data is expanded by categorizing at least one of the number of days of prescription, number of prescriptions, and medical treatment date of the usage of the prescription drug into a plurality of groups. A machine learning function that generates a trained model by machine learning based on the
an inference function that infers the name of the injury or disease from the prescription data input to the learned model;
A disease name inference program that is characterized by realizing the following. A disease name inference program executed on a computer that infers a disease name using a disease name inference system in which a processor infers a disease name,
The computer,
Machine learning is performed based on learning data that is expanded by using data that associates prescription drugs in prescription data with injury and disease name data, and classifies medical treatment days into at least one of seasons, quarters, months, and days of the week. A machine learning function that generates a trained model,
an inference function that infers the name of the injury or disease from the prescription data input to the learned model;
A disease name inference program that is characterized by realizing the following. A disease name inference program executed on a computer that infers a disease name using a disease name inference system in which a processor infers a disease name,
The computer,
When prescription drugs in prescription data and injury/illness name data are associated, and a predetermined standard medical department name partially matches the department name, the department name can be set for each matching part to classify it into multiple groups. A machine learning function that generates a trained model by machine learning based on the training data expanded by
an inference function that infers the name of the injury or disease from the prescription data input to the learned model;
A disease name inference program that is characterized by realizing the following. A disease name inference program executed on a computer that infers a disease name using a disease name inference system in which a processor infers a disease name,
The computer,
Prescription drugs in prescription data and injury/illness name data are associated, a drug code is used as the prescription drug data, a part of the drug code is extracted, and the data is classified into multiple groups for expanded learning purposes. A machine learning function that generates a trained model using machine learning based on data,
an inference function that infers the name of the injury or disease from the prescription data input to the learned model;
A disease name inference program that is characterized by realizing the following.

Images