JP2021197005A - Information processor, information processing method, and information processing program - Google Patents

Abstract

To appropriately perform evaluation concerning a side effect of medicine.SOLUTION: An information processor comprises: a storage section for storing a prediction model generated on the basis of learning data that is generated from actual result data on a side effect of medicine and that is learning data obtained by associating a multidimensional feature vector including feature quantities of medicine information on medicine, object person information on an object person, time information on time with information on occurrence results of the side effect; a reception section for receiving input information including one or more feature quantities of the medicine information, the object person information, and the time information; a calculation section for calculating occurrence information on the occurrence of the side effect on the basis of the prediction model and the input information; and an output section for outputting the occurrence information.SELECTED DRAWING: Figure 3

Description

The present invention relates to an information processing apparatus, an information processing method and an information processing program.

Conventionally, in drug safety evaluation, information on side effects voluntarily reported to the authorities by medical personnel, patients, pharmaceutical companies, etc. when side effects or adverse events (hereinafter collectively referred to as "side effects") occur. A database that manages the information (hereinafter referred to as "spontaneous report database") is used. As the spontaneous report database, for example, the Japanese Adverse Drug Event Report database (JADER) of the Pharmaceuticals and Medical Devices Agency (PMDA) is known.

For example, Non-Patent Document 1 describes that signal detection using a spontaneous report database is performed for the safety evaluation of a drug. Here, signal detection is the discovery of spontaneous reports that require detailed investigation and the prioritization of the need for investigation. For signal detection, for example, an analysis method called imbalance analysis such as Proportion Reporting Ratio (PRR), Reporting Odds Ratio (ROR), Bayesian Confidence Propagation Neural Network method (BCPNN) or Multi-Item Gamma-Poisson Shrinker (MGPS). Is used.

Hiroshi Susuda, Yukio Takahashi. Planning / New utilization of PMDA's spontaneous report database and future issues 4. Search for drug interactions that cause serious drug eruption, Drug Epidemiology, 2014, Vol. 19, No. 1, p. 39-49, Internet <URL: https://www.jstage.jst.go.jp/article/jjpe/19/1/19_39/_pdf/-char/ja>

However, for example, when a spontaneous report database is used as described in Non-Patent Document 1, there is a possibility that the side effects due to the administration of the drug cannot be properly evaluated. For example, drugs for which sufficient data on the occurrence of adverse drug reactions (hereinafter referred to as "actual data") cannot be obtained (for example, drugs that have not been reported in the spontaneous report database, compounds that are not approved as drugs, etc.), subjects. (For example, subjects in their 90s, etc.) may not be able to properly evaluate side effects. In addition, since the spontaneous reporting database is susceptible to bias such as reporting bias, there is a risk that side effects cannot be properly evaluated.

Therefore, one of the objects of the present invention is to provide an information processing apparatus, an information processing method and an information processing program capable of appropriately evaluating the side effects of a drug.

The information processing apparatus according to one aspect of the present invention is learning data generated from actual data on side effects of a drug, and includes a feature amount of drug information on the drug, subject information on a subject, and time information on time. A storage unit that stores a prediction model generated based on the learning data in which a dimensional feature vector and information on the occurrence record of side effects are associated with each other, and one or more feature quantities of drug information, subject information, and time information are included. It includes a reception unit that receives input information, a calculation unit that calculates expression information regarding the occurrence of side effects based on the prediction model and the input information, and an output unit that outputs the expression information.

According to this aspect, each using a prediction model generated based on training data in which a multidimensional feature vector including feature quantities of drug information, subject information, and time information is associated with information on the actual occurrence of side effects. Since the information on the occurrence of side effects is calculated, the information on the occurrence of each side effect can be appropriately calculated even for drugs, subjects, side effects, etc. for which sufficient actual data cannot be obtained.

In the above embodiment, a calculation unit for calculating evaluation information regarding side effects may be further provided based on the expression information, and the output unit may output the evaluation information. According to this aspect, since the evaluation information regarding the side effect is calculated based on the expression information, the reliability of the evaluation information can be improved.

In the above aspect, the calculation unit may calculate the evaluation information based on the expression information and the weighting coefficient based on the severity of the side effect. According to this aspect, since the evaluation information is calculated based on the weighting coefficient based on the severity of the side effect, the evaluation information can be calculated more appropriately.

In the above embodiment, the reception unit further receives specific target person information regarding the specific target person, and the calculation unit expresses side effects of the specific target person based on the expression information about the specific target person. The evaluation information regarding the above may be calculated. According to this aspect, since it is possible to recognize the susceptibility of side effects to a specific subject before administration of the drug, it is possible to improve the safety at the time of administration to the specific subject.

In the above embodiment, the reception unit further receives specific drug information regarding the specific drug, and the calculation unit uses the expression information for the specific drug as an index value for the safety of the specific drug. The evaluation information may be calculated. According to this aspect, the index value of the safety of a specific drug can be appropriately calculated.

The learning device according to another aspect of the present invention includes an acquisition unit that acquires actual data regarding side effects of a drug, learning data generated from the actual data, such as drug information regarding the drug, subject information regarding a subject, and subject information. A predictive model for the occurrence of side effects by a generator that generates the training data that associates a multidimensional feature vector including the feature amount of time information with respect to time and information on the actual occurrence of side effects, and machine learning based on the training data. It is equipped with a learning unit that generates data.

The information processing method according to another aspect of the present invention includes a step of accepting input information including one or more feature quantities of drug information, subject information, and time information, and learning data generated from actual data on side effects of the drug. Prediction generated based on the learning data in which a multidimensional feature vector including feature quantities of drug information regarding the drug, subject information regarding the subject, and time information regarding time is associated with information regarding the occurrence record of side effects. An information processing method in an information processing apparatus, comprising a step of calculating expression information regarding the occurrence of side effects based on a model and the input information, and a step of outputting the expression information.

The information processing program according to another aspect of the present invention accepts input information including one or more feature quantities of drug information, subject information, and time information to the information processing apparatus, and results data on side effects of the drug. The learning data generated from the above, in which a multidimensional feature vector including drug information about the drug, subject information about the subject, and feature amounts of time information about time is associated with information about the occurrence record of side effects. Based on the prediction model generated based on the above and the input information, calculation of expression information regarding the occurrence of side effects and output of the expression information are executed.

According to the present invention, it is possible to appropriately evaluate the side effects of a drug.

It is a figure which shows an example of the derivation of the occurrence information of the side effect which concerns on this embodiment. It is a figure which shows an example of the occurrence probability of a specific side effect which concerns on this embodiment. It is a figure which shows an example of the occurrence probability of each side effect about a specific drug which concerns on this embodiment. It is a figure which shows an example of the occurrence probability of each side effect about the specific subject which concerns on this embodiment. It is a figure which shows an example of the functional structure of the information processing system which concerns on this embodiment. It is a figure which shows an example of the performance data which concerns on this embodiment. It is a figure which shows an example of the hardware composition of each apparatus in the information processing system which concerns on this embodiment. It is a flowchart which shows an example of the operation of the learning apparatus which concerns on this embodiment. It is a flowchart which shows an example of the operation of the information processing apparatus which concerns on this embodiment.

An embodiment of the present invention will be described with reference to the accompanying drawings. In each figure, those having the same reference numerals have the same or similar configurations.

In this embodiment, an Adverse Event (AE) is an undesired or unintended effect (eg, sign, symptom, illness, etc.) that occurs in a patient to whom a drug has been administered and is referred to as an adverse event. May be good. The side effect is not limited to those having a causal relationship with the drug, and may include those having an unclear causal relationship. In the following, side effects are assumed to be described, for example, in the Medical Dictionary for Regulatory Activities (MedDRA) of the International Council for Harmonization of Technical Requirements for Pharmaceuticals for Human Use (ICH). However, it is not limited to this.

Further, in the present embodiment, the drug may include not only an approved drug but also a compound not approved as a drug.

FIG. 1 is a diagram showing an example of deriving information on the occurrence of side effects according to the present embodiment. As shown in FIG. 1, information on a drug (hereinafter referred to as "drug information") D, information on a subject to whom the drug is administered (hereinafter referred to as "subject information") P, and information on time (hereinafter referred to as "subject information"). , Information on the occurrence of side effects at a predetermined time when a predetermined drug is taken by a predetermined subject (hereinafter referred to as "expression information") may be derived based on T (referred to as "time information").

Here, the expression information may be, for example, information indicating whether or not a side effect occurs (that is, the presence or absence of expression), or the probability that a side effect will occur (hereinafter, referred to as “expression probability”) or. It may be information indicating the probability that it does not appear. Further, the expression information may be information indicating the number of reports of future side effects in a predetermined period (for example, monthly or semi-annual unit).

As shown in FIG. 1, the expression information may be derived based on a feature vector including the feature amounts of the drug information D, the subject information P, and the time information T. The feature vector indicates a predetermined point and / or region (hereinafter referred to as "segment") on a feature space having at least three dimensions including a drug, a subject and time. The three-dimensional or higher feature space may be referred to as a "multidimensional feature space", and the feature vector may be referred to as a "multidimensional feature vector". Thus, the expression information can be thought of as a function on the feature space consisting of the drug, the subject and time, and may be expressed as "f (P, D, T)".

Here, the drug information D may include or indicate, for example, information regarding at least one of the following. The drug may be specified by the drug information D.
・ Name
・ Chemical Structure
・ Dosage Form
・ Indication Disease for which the drug has been approved
-Reason for use-Effectiveness or safety of clinical trials (eg, clinical trials) -Information on clinical trials (eg, study purpose, design, usage, etc.)
Pharmacokinetics (also referred to as pharmacokinetics (PK) and / or pharmacodynamics (PD), eg, PK and / or PD submitted when approving a drug).
・ Presence / absence of already known cytochrome P450 (CYP) inhibition ・ Information on non-clinical studies (information on LD50, ED50, etc., physical property information, etc.)
・ Molecular weight ・ Drug interaction with other drugs ・ Drug efficacy classification ・ Purpose of use or reason for use of target diseases when using drugs ・ Continuous time information (for example, predetermined unit (for example, month or year)) from the start of sales number)
・ Number of past side effects reported ・ Number of patients using drugs ・ Prescribing medical institutions or prescribing doctors ・ Dispensing pharmacies or pharmacists ・ Drug managers (persons, families, care providers, etc.)
・ Medical staff (opinions of doctors, pharmacists, etc.)
・ Opinions of drug users ・ Opinions of side effect reporters ・ Instructions from the government (contents and frequency of blue letters and yellow letters, revision of package inserts, implementation of side effect relief system and its contents, etc.)
-Drug administration period-Manufacturing and sales company-Manufacturing location and company-Included additives-Manufacturing drug substance information-Manufacturing process-Drug price Here, the above chemical structure is also called a chemical structural formula, for example, a simplified molecular input line entry system. It may be expressed in (SMILES) notation. Further, the dosage form may be, for example, a tablet, a capsule, a powder, a liquid medicine, an injection, or the like. The above-mentioned medicinal effect classification may be, for example, a medicinal effect classification number attached to a drug price-based listed drug code according to the Anatomical Therapeutic Chemical (ATC) Classification System. Further, the number of patients using the drug may be obtained from, for example, National Data Base, receipt information, shipping information, and the like.

Further, the subject information P may include, or may indicate, information relating to at least one of the following, for example. The target person may be specified by the target person information P.
・ Gender (Sex)
・ Age
・ Race
・ Occupation
・ Weight / height / clinical course (place of hospitalization, date of hospitalization, date of onset of side effects, etc.)
・ Presence / absence of affected disease, if there is a disease, the disease / history / complications, if there are complications, the complications / presence / absence of concomitant drugs, if there are concomitant drugs, the concomitant drugs and the number of drugs ・Gene, lifestyle, personality, health literacy, relationship with doctors, consultation status or medical treatment behavior, medication behavior such as adherence, personal or group SNS usage information, socio-economic status (for example, education, income, etc.)
Here, the disease may be, for example, a disease described at the International Classification of Diseases (ICD) 10 level. Further, the above-mentioned combination drug is not limited to a drug that has been used as a drug, and may be a drug or a chemical compound that has been used or has not been used as a drug at the component name level of the drug.

Further, the time information T may include, or may indicate, information relating to at least one of the following, for example. A predetermined time may be specified by the time information T.
・ Period until side effects appear ・ Presence / absence of side effects at each time point over time ・ Duration of drug administration (Treatment Duration)
-Period during which side effects are occurring-Start date of occurrence of side effects Here, the above period is, for example, a number of predetermined time units (for example, 1 second, 1 minute, 1 hour, etc.), and a predetermined time category (for example, 1 hour). , Variables indicating (immediately after administration, half a day, several days later), etc.), start date and time, end date and time, start date and end date, etc. may be indicated.

The drug information D, subject information P, and time information T include data based on the actual occurrence of each side effect, such as information in clinical trials at the time of drug development and information registered in a spontaneous report database (hereinafter, "" Actual data ") can be used. In the actual data, the drug information D, the subject information P, the time information T, and the expression information may be associated with each other. In addition, at least one of the above-mentioned drug information D, subject information P, and time information T includes data items (also referred to as "E2B" data items, etc.) for transmitting an individual case safety report based on an agreement in ICH. It can also be used.

FIG. 2 is a diagram showing an example of the occurrence probability of a specific side effect according to the present embodiment. In FIG. 2, the probability of occurrence of vomiting calculated based on the above-mentioned actual data is shown for each segment. For example, in FIG. 2, each segment is specified by subject information P, drug information D, and time information T, but is not limited to this as described above.

As shown in FIG. 2, it is assumed that the probability of vomiting of the drug X differs depending on the subject information P (for example, age, disease, etc.) and / or the time information T (for example, elapsed time after administration). To. For example, as shown in FIG. 2, even if the elapsed time after administration of the same drug X is the same, it is assumed that the onset probabilities differ among subjects in their teens, 50s, and 90s. Similarly, even if the subjects are the same, it is assumed that the probability of onset differs between immediately after administration of drug X and within 6 hours after administration.

However, for example, actual data such as information in clinical trials and information registered in a spontaneous report database are limited. Therefore, there is a possibility that the expression probabilities of all the segments in the feature space cannot be comprehensively derived only from the drug information D, the subject information P, and the time information T obtained from the actual data. For example, FIG. 2 does not show the probability of occurrence for subjects and time segments that have not been administered drug X. In addition, for drugs with a virtual chemical structure (for example, new drugs) and side effects that rarely occur, the probability of occurrence of each segment may not be derived.

Therefore, in the present embodiment, machine learning is performed using learning data in which a multidimensional feature vector including feature quantities of drug information D, subject information P, and time information T and information on the occurrence record of side effects are associated with each other. Therefore, a prediction model regarding whether or not each side effect occurs in a predetermined time due to administration of the drug to the subject is generated, and the expression information of each segment is calculated based on the prediction model.

In the present embodiment, further, by deriving at least one expression information of drug information D, subject information P, and time information T using a prediction model, for example, (1) each side effect of a specific drug. (2) Information on the expression of each side effect for a specific subject can be derived. (1) Information on the occurrence of one or more side effects of a specific drug can be used, for example, to derive safety evaluation information (hereinafter referred to as “safety score”) of the specific drug. In addition, (2) information on the occurrence of one or more side effects for a specific subject is used, for example, for deriving evaluation information on the likelihood of side effects for a specific individual (hereinafter referred to as "side effect score"). can. In this embodiment, the expression probability will be mainly described as an example of the expression information, but as described above, the expression information is not limited to the expression probability. Further, the evaluation information may be called an evaluation index or the like.

With reference to FIGS. 3 and 4, (1) the derivation of the safety score based on the probability of occurrence of each side effect for a specific drug, and (2) the side effect score based on the probability of occurrence of each side effect for a specific subject. Derivation will be described. In FIGS. 4 and 5, it is assumed that the expression probabilities of all segments are calculated using the prediction model obtained as a result of machine learning for each of vomiting, blood pressure increase, and drowsiness.

FIG. 3 is a diagram showing an example of the occurrence probability of each side effect for a specific drug according to the present embodiment. For example, FIG. 3 shows the probability of developing vomiting, increased blood pressure, and drowsiness due to administration of drug X to each subject. In FIG. 3, vomiting, increased blood pressure, and drowsiness are shown as side effects, but it goes without saying that the side effects are not limited to those shown in the illustration.

As shown in FIG. 3, the safety score of the drug X may be derived based on the subject information P, the time information T, and the probability of occurrence of each side effect aggregated for the drug X. In addition, a safety score may be derived based on the probability of occurrence and a weighting factor based on the severity of each side effect.

For example, in FIG. 3, the safety score of the drug X may be calculated by adding the average values of the expression probabilities for each of the subject information P and the time information T for each side effect. Further, the average value may be multiplied by the above weighting factor. For example, since an increase in blood pressure is considered to be more serious than drowsiness and vomiting, the average value of the probability of occurrence of an increase in blood pressure is a weighting coefficient having a larger value than the average value of the probability of occurrence of drowsiness and vomiting. May be multiplied.

In this way, by deriving the safety score of the drug X based on the subject information P, the time information T, and the onset probability for each side effect aggregated for the drug X, the influence of bias such as bias of the actual data can be exerted. It can be mitigated and the reliability of the safety score can be improved. It is also possible to calculate the safety score of drug X for each subject and / or time. In addition, it is possible to derive an absolute evaluation value for each drug instead of a relative evaluation value as in the existing safety evaluation. The safety score is not limited to each drug, and may be calculated for each combination of one or a plurality of drugs. By calculating the safety score of a combination of a plurality of drugs, it is possible to appropriately evaluate the safety of a combination of multiple drugs (polypharmacy).

FIG. 4 is a diagram showing an example of the occurrence probability of each side effect for a specific subject according to the present embodiment. For example, FIG. 4 shows the probability of developing vomiting, increased blood pressure and drowsiness due to administration of drugs X, Y and Z to a specific subject. It should be noted that the side effects and drugs shown in FIG. 4 are merely examples, and of course, they are not limited to those shown in the drawings.

As shown in FIG. 4, the side effect score of the specific subject may be derived based on the side effect, the drug information D, and the occurrence probability for each time information T aggregated for the specific subject. For example, in FIG. 4, the average value of the expression probabilities for each drug information D and time information T for each side effect may be used as the side effect score of a specific subject.

In this way, by deriving the side effect score of a specific subject based on the side effect, the drug information D, and the probability of occurrence for each time information T aggregated for the specific subject, the administration of the drug to the specific subject is performed. Since it is possible to recognize the likelihood of side effects before, it is possible to improve the safety of medication for a specific subject. The side effect score may be calculated for a specific drug (for example, a combination of one or a plurality of drugs) for a specific subject. By calculating the side effect score of a combination of a plurality of drugs, it is possible to improve the safety of a multidrug combination (polypharmacy) for a specific subject.

Hereinafter, the information processing system according to the present embodiment will be described in detail.

(Configuration of information processing system)
FIG. 5 is a diagram showing an example of the functional configuration of the information processing system according to the present embodiment. As shown in FIG. 5, the information processing system 1 includes a learning device 10, a storage device 20, and an information processing device 30. In this example, the learning device 10, the storage device 20, and the information processing device 30 are shown as separate devices, but at least two of these may be configured as an integrated device. A plurality of devices configured as an integrated device may be realized, for example, as different operations of PLC (Programmable Logic Controller). Further, at least one of the learning device 10, the storage device 20, and the information processing device 30 may be divided into a plurality of devices.

<Learning device>
As shown in FIG. 5, the learning device 10 includes an acquisition unit 11, a generation unit 12, and a learning unit 13.

The acquisition unit 11 acquires actual data regarding the actual occurrence of side effects. The actual data is data on whether or not each side effect actually occurred in a predetermined time by administration of a certain drug to a certain subject. The acquisition unit 11 may acquire, for example, the actual data stored in the storage device 20. The acquisition unit 11 may acquire actual data from another device via the communication unit 10c.

FIG. 6 is a diagram showing an example of actual data according to the present embodiment. As shown in FIG. 6, the actual data is data in which the subject information P, the drug information D, and the time information T are associated with the information regarding the actual occurrence of each side effect (hereinafter referred to as “expression actual information”). There may be. The drug information D, subject information P, and time information T may be, for example, information in a clinical test at the time of drug development, information registered in a spontaneous report database (for example, JADER), or original information. It may be information on the occurrence of side effects collected in.

Note that FIG. 6 is merely an example, and the subject information P, the drug information D, and the time information T are not limited to those shown in FIG. Further, in FIG. 6, the actual occurrence information of each side effect indicates the presence or absence of the occurrence of each side effect, but the occurrence probability of each side effect may be indicated.

The generation unit 12 extracts from the actual data acquired by the acquisition unit 11 a multidimensional feature vector including the feature amounts of the subject information P, the drug information D, and the time information T, and the occurrence record information of the side effect, and extracts these. Generate the associated learning data.

For example, in the case of the actual data shown in FIG. 6, subject information P (gender "male", age "60s", weight "50 kg", height "160 cm"), drug information D (name "magnesium oxide", For the feature vector including the reason for use "constraint") and time information T (onset time "immediately after taking"), the actual occurrence information of each side effect (vomiting "yes", blood pressure increase "no", drowsiness "yes" as teacher data. ”) To generate training data associated with it. In this way, a training data set is generated from a large number of actual data.

The learning unit 13 performs machine learning using the learning data set generated by the generation unit 12. As a result, when a multidimensional feature vector including the feature quantities of the subject information P, the drug information D, and the time information T is input, a prediction model 13a is generated in which the occurrence information of each side effect corresponding to the feature vector is output. Will be done.

In machine learning by the learning unit 13, the presence / absence (class) of the occurrence of each side effect, the probability of occurrence of each side effect, and the like may be derived by class identification. In the case of class identification, for example, the prediction model 13a may be configured by a neural network. In this case, the side effect occurrence record information is used as a label in machine learning, and the learning unit 13 may perform the learning process of the neural network by, for example, the error back propagation method. Alternatively, in machine learning by the learning unit 13, the probability of occurrence of each side effect may be derived by value prediction. In the case of value prediction, the prediction model 13a may be constructed using, for example, regression analysis or multiple regression analysis.

<Information processing device>
The information processing device 30 includes a reception unit 31, an extraction unit 32, a calculation unit 33, a calculation unit 34, and an output unit 35. The reception unit 31 (or the reception unit 31 and the extraction unit 32) may constitute the "reception unit" of the present invention.

The reception unit 31 receives input information including one or more feature quantities of drug information D, subject information P, and time information T. Specifically, the reception unit 31 receives input of at least one feature amount among the feature amounts contained in the drug information D, the subject information P, and the time information T, which are the prediction targets of the expression information.

For example, if a man in his 70s wants to predict the side effects one day after taking magnesium oxide as a laxative, the subject information P includes gender "male", age "70s", and drug information. Input information including the name "magnesium oxide", the reason for use "constipation" as D, and the onset time "1 day later" as the time information T is accepted. The drug information D, the subject information P, and the time information T may correspond to, for example, a segment in which the expression probability cannot be calculated based on the actual data in the feature space shown in FIG. Further, the drug information D may indicate a virtual drug (for example, a new drug, a chemical substance before approval, etc.).

The input information may be input from the input means included in the information processing device 30, may be received from another device via the communication unit 10c, or may be received from the storage device 20.

In addition, the reception unit 31 may receive specific target person information regarding a specific target person and / or specific drug information regarding a specific drug. The specific target person information and / or the specific drug information may be used for calculation of evaluation information (for example, side effect score, safety score, etc.) described later.

The extraction unit 32 extracts a feature vector indicating the feature amount of the subject information P, the drug information D and / or the time information T from the input information received by the reception unit 31. For example, in the above example, from the input information, subject information P (gender "male", age "70s"), drug information D (name "magnesium oxide", reason for use "constipation"), time information T ( A feature vector containing the onset time “1 day later”) is extracted.

The calculation unit 33 calculates the occurrence information of each side effect by inputting the feature vector extracted by the extraction unit 32 into the prediction model 33a. For example, the calculation unit 33 determines the probability of occurrence of each side effect at a predetermined time by administering a predetermined drug to a predetermined subject based on the subject information P, the drug information D and / or the time information T indicated by the feature vector. Calculated as expression information. Further, the probability of occurrence of a predetermined side effect may be calculated for each specific drug information D, subject information P and / or time information T.

Specifically, the prediction model 33a may be a prediction model 13a generated by machine learning in the learning device 10. Further, it may be a new prediction model generated by further distillating the prediction model 13a. In distillation, a new prediction model may be generated using input / output data to the prediction model 13a machine-learned by the learning device 10.

The calculation unit 33 may generate prediction data in which the calculated expression information is associated with the drug information D, the subject information P and / or the time information T, and store the generated prediction data in the storage device 20. ..

The calculation unit 34 calculates evaluation information regarding side effects based on the expression information calculated by the calculation unit 33. For example, the calculation unit 34 may aggregate the probability of occurrence of a side effect calculated by the calculation unit 33 as evaluation information for at least one of a specific drug, a specific subject, a specific time, and a specific side effect. When the request information for the aggregation is received, the calculation unit 34 may perform the aggregation based on the request information.

The calculation unit 34 may calculate a safety score for a specific drug based on the probability of occurrence of each side effect aggregated for the specific drug. That is, the calculation unit 34 calculates the overall safety score for the side effect of the specific drug from the probability of occurrence of all or arbitrary side effects that can occur with the specific drug. Further, the calculation unit 34 may calculate the safety score based on the occurrence probability of each side effect calculated by the calculation unit 33 and the weighting coefficient based on the severity of each side effect.

The calculation unit 34 may calculate the side effect score of the specific target person based on the occurrence probability of each side effect aggregated for the specific target person.

The output unit 35 outputs the side effect occurrence information calculated by the calculation unit 33 and / or the evaluation information regarding the side effect calculated by the calculation unit 34. Specifically, the output unit 35 provides the expression information and the evaluation information to the user via the communication unit 10c or the output unit 10e, which will be described later.

<Hardware configuration>
Next, the hardware configuration of each device (for example, each of the learning device 10, the storage device 20, and the information processing device 30) in the information processing system 1 will be described. As shown in FIG. 7, each device in the information processing system 1 has a processor 10a such as a CPU (Central Processing Unit) corresponding to an arithmetic unit, a storage device 10b, a communication unit 10c, an input unit 10d, and an output. It has a part 10e. Each of these configurations is connected to each other via a bus so that data can be transmitted and received. In this example, the learning device 10, the storage device 20, and the information processing device 30 are composed of separate computers, but at least two of them may be configured by one computer.

The processor 10a is a control unit that controls execution of a program stored in the storage device 10b, calculates data, and processes data. The processor 10a may be an arithmetic unit (arithmetic unit) that executes a program (information processing program) for calculating the occurrence probability of each side effect. The processor 10a receives various input data from the input unit 10d and / or the communication unit 10c, outputs (for example, displays) the calculation result of the input data to the output unit 10e, stores it in the storage device 10b, or It may be transmitted via the communication unit 10c.

The storage device 10b is at least one of a memory, an HDD (Hard Disk Drive), and an SSD (Solid State Drive). The storage device 10b of the information processing device 30 may store the information processing program executed by the processor 10a. Further, the storage device 10b of the learning device 10 and the information processing device 30 may store the prediction models 13a and 33a. The storage device 10b may constitute the "storage unit" of the present invention.

The communication unit 10c is an interface for connecting each device in the information processing system 1 to an external device. When the learning device 10 and the information processing device 30 are configured as an integrated device, the communication unit 10c performs interprocess communication between the process operating as the learning device 10 and the process operating as the information processing device 30. May include.

The input unit 10d receives data input from the user, and may include, for example, at least one of a keyboard, a mouse, a touch panel, and a microphone.

The output unit 10e outputs the calculation result of the processor 10a, and may be composed of at least one of a display such as an LCD (Liquid Crystal Display) and a speaker.

The information processing program may be stored and provided in a storage medium readable by a computer such as the storage device 10b, or may be provided via a network connected by the communication unit 10c. The storage medium in which the determination program is stored may be a non-transitory computer readable medium. The non-temporary storage medium is not particularly limited, but may be, for example, a storage medium such as a USB memory, a CD-ROM, or a DVD.

In the information processing apparatus 30, when the processor 10a executes the information processing program, the operations of the reception unit 31, the extraction unit 32, the calculation unit 33, the calculation unit 34, and the output unit 35 described with reference to FIG. 5 are realized. .. It should be noted that these physical configurations are examples and do not necessarily have to be independent configurations. For example, the information processing device 30 may include an LSI (Large-Scale Integration) in which a processor 10a and a storage device 10b are integrated.

(Operation of information processing system)
<Operation of learning device>
FIG. 8 is a flowchart showing an example of the operation of the learning device according to the present embodiment. The operation of the learning device 10 shown in FIG. 8 is merely an example, and is not limited to the one shown in the figure.

As shown in FIG. 8, in step S101, the learning device 10 acquires actual data (for example, FIG. 6) in which the subject information P, the drug information D, and the time information T are associated with the actual occurrence information of each side effect. do. The manifestation performance information associated with the subject information P, the drug information D, and the time information T in the performance data may be used as the teacher data for machine learning in step S103. The learning device 10 may store the actual data in the storage device 20.

In step S102, the learning device 10 extracts a multidimensional feature vector including the feature amounts of the subject information P, the drug information D, and the time information T included in the actual data, and uses the multidimensional feature vector and the actual data as the multidimensional feature vector. Generate learning data associated with the actual occurrence information of each included side effect.

In step S103, the learning device 10 performs machine learning using the generated learning data, and when a feature vector is input, generates a prediction model 13a that outputs information on the occurrence of side effects corresponding to the feature vector.

<Operation of information processing device>
FIG. 9 is a flowchart showing an example of the operation of the information processing apparatus according to the present embodiment. The operation of the information processing apparatus 30 shown in FIG. 9 is merely an example, and is not limited to the one shown in the figure. For example, the information processing apparatus 30 may omit step S204.

As shown in FIG. 9, in step S201, the information processing apparatus 30 selects one or more feature amounts among the feature amounts contained in the drug information D, the subject information P, and the time information T, which are the prediction targets of the expression information. Accepts input information including. The drug information D, the subject information P, and the time information T may correspond to, for example, a segment in which the expression probability cannot be calculated based on the actual data in the feature space shown in FIG. Further, the drug information D may indicate a virtual drug (for example, a new drug, a chemical substance before approval, etc.).

In step S202, the information processing apparatus 30 extracts a feature vector including the feature amounts of the drug information D, the subject information P, and / or the time information T, which are the prediction targets of the expression information.

In step S203, the information processing apparatus 30 inputs the feature vector extracted in step S202 into the prediction model 33a, and each side effect corresponding to the drug information D, the subject information P, and / or the time information T indicated by the feature vector. The expression probability of is calculated as expression information. The information processing device 30 may generate prediction data in which the drug information D, the subject information P and / or the time information T are associated with the calculated expression information, and store the prediction data in the storage device 20.

In step S204, the information processing apparatus 30 calculates the evaluation information regarding the side effect based on the calculated occurrence information of each side effect. For example, the information processing apparatus 30 aggregates the probability of occurrence of each side effect calculated in step S203 for at least one of a specific drug, a specific subject, a specific time, and a specific side effect.

As the evaluation information, the information processing apparatus 30 may calculate the subject information P, the time information T, and the occurrence probability for each side effect aggregated for the drug X, as shown in FIG. Further, the information processing apparatus 30 may calculate the time information T aggregated for the drug X and the specific subject and the probability of occurrence for each side effect. Further, the information processing apparatus 30 may calculate the occurrence probability for each of the subject information P and the time information T aggregated for the drug X and a specific side effect (for example, vomiting).

Further, as the evaluation information, the information processing apparatus 30 may calculate the drug information D, the time information T, and the occurrence probability for each side effect aggregated for the specific subject information P as shown in FIG. Further, the information processing apparatus 30 may calculate the occurrence probability for each of the specific subject information P, the drug information D aggregated for the specific side effect, and the time information T.

Further, the information processing apparatus 30 may calculate at least one of the safety score of a specific drug and the side effect score of a specific subject based on the probability of occurrence of each side effect.

Specifically, as shown in FIG. 3, the information processing apparatus 30 calculates the safety score of the drug X by using the subject information P, the time information T, and the probability of occurrence for each side effect aggregated for the drug X. You may. For example, in FIG. 3, the safety score of the drug X may be derived by adding the average values of the expression probabilities for each of the subject information P and the time information T of each side effect.

Further, as shown in FIG. 4, the information processing apparatus 30 calculates the side effect score of the specific subject using the drug information D, the time information T, and the occurrence probability for each side effect aggregated for the specific subject. May be. For example, in FIG. 4, the side effect score of a specific subject may be derived from the average value of the expression probabilities for each drug information D and time information T for each side effect.

In step S205, the information processing apparatus 30 outputs at least one of the expression information calculated in step S203 and the evaluation information calculated in step S204 and provides it to the user.

As described above, according to the information processing system 1 according to the present embodiment, learning data in which a multidimensional feature vector including feature quantities of drug information, subject information, and time information is associated with information on the occurrence record of side effects. Since the side effect occurrence information (for example, the occurrence probability) is calculated using the prediction model generated based on the above, the side effect expression information can be obtained even for drugs, subjects, side effects, etc. for which sufficient actual data cannot be obtained. Can be calculated properly.

Further, according to the information processing system 1 according to the present embodiment, the side effect score of a specific subject is calculated based on the above-mentioned expression information, so that the susceptibility to side effects before administration of the drug to the specific subject is determined. It is recognizable and can improve the safety of medication for specific subjects.

Further, according to the information processing system 1 according to the present embodiment, since the safety score of a specific drug is calculated based on the expression information, the index value of the safety of the specific drug can be appropriately calculated, and a specific drug can be calculated. It is possible to improve the safety when administering a drug.

The embodiments described above are for facilitating the understanding of the present invention, and are not for limiting the interpretation of the present invention. Each element included in the embodiment and its arrangement, material, condition, shape, size, and the like are not limited to those exemplified, and can be appropriately changed. Further, it is possible to partially replace or combine the configurations shown in different embodiments.

For example, in the above embodiment, the drug has been described as being a drug or a compound, but the drug is not limited to this, and the drug may include, for example, a drug prescribed by a doctor, such as a medical device or a therapeutic application. ..

Further, in the above embodiment, a prediction model generated based on learning data in which a multidimensional feature vector including feature quantities of drug information D, subject information P, and time information T and information on the occurrence record of side effects are associated with each other. Allowed appropriate evaluation of the drug by using, but is not limited to this. For example, medical evaluation is performed by using a prediction model generated based on learning data using medical information (also referred to as "medical information") such as information on medical devices and medical applications instead of drug information D. (For example, evaluation of medical equipment or medical application) can also be appropriately performed. As described above, in the present embodiment, the multidimensional feature vector including the feature quantities of the medical information, the subject information P, and the time information T, and the information regarding the predetermined performance (for example, the treatment performance using the medical device, the medical application, etc.) By using the prediction model generated based on the training data associated with), it can be applied to various evaluations related to medical treatment.

1 ... Information processing system, 10 ... Learning device, 20 ... Storage device, 30 ... Information processing device, 11 ... Acquisition unit, 12 ... Generation unit, 13 ... Learning unit, 13a ... Prediction model, 31 ... Reception unit, 32 ... Extraction Unit, 33 ... Calculation unit, 33a ... Prediction model, 34 ... Calculation unit, 35 ... Output unit, D ... Drug information, P ... Subject information, T ... Time information, 10a ... Processor, 10b ... Storage device, 10c ... Communication Unit, 10d ... Input unit, 10e ... Output unit

Claims (8)

Learning data generated from actual data on drug side effects, including a multidimensional feature vector including drug information on the drug, subject information on the subject, and time information on time, and information on the actual occurrence of side effects. A storage unit that stores a prediction model generated based on the associated training data, and a storage unit.
A reception unit that accepts input information including one or more feature quantities of drug information, subject information, and time information,
A calculation unit that calculates expression information regarding the occurrence of side effects based on the prediction model and the input information.
An output unit that outputs the expression information and
Information processing device equipped with. Further, a calculation unit for calculating evaluation information regarding side effects based on the expression information is provided.
The output unit outputs the evaluation information.
The information processing apparatus according to claim 1. The calculation unit calculates the evaluation information based on the expression information and the weighting coefficient based on the severity of the side effect.
The information processing apparatus according to claim 2. The reception unit further receives specific target person information regarding a specific target person, and receives information on the specific target person.
The calculation unit calculates the evaluation information regarding the occurrence of side effects of the specific subject based on the expression information of the specific subject.
The information processing apparatus according to claim 2 or 3. The reception department further receives specific drug information regarding a specific drug,
The calculation unit calculates the evaluation information as an index value of the safety of the specific drug based on the expression information of the specific drug.
The information processing apparatus according to any one of claims 2 to 4. The acquisition department that acquires actual data on the side effects of drugs,
The learning data generated from the actual data, which is a combination of a multidimensional feature vector including drug information about the drug, subject information about the subject, and time information about time, and information about the actual occurrence of side effects. And the generator that generates
A learning unit that generates a predictive model for the occurrence of side effects by machine learning based on the learning data.
A learning device equipped with. A step of accepting input information including one or more feature quantities of drug information, subject information and time information,
Learning data generated from actual data on drug side effects, including a multidimensional feature vector including drug information on the drug, subject information on the subject, and time information on time, and information on the actual occurrence of side effects. A step of calculating expression information regarding the occurrence of side effects based on the prediction model generated based on the associated learning data and the input information.
It has a step of outputting the expression information.
Information processing method in an information processing device. For information processing equipment
Accepting input information including one or more features of drug information, subject information and time information,
Learning data generated from actual data on drug side effects, including a multidimensional feature vector including drug information on the drug, subject information on the subject, and time information on time, and information on the actual occurrence of side effects. To calculate the expression information based on the prediction model generated based on the associated learning data and the input information.
Outputting the expression information and
Information processing program to execute.

Images