JP2023009004A - Apparatus and method for predicting demand for medicine in medical institution, and program thereof - Google Patents

Abstract

To provide an unprecedented method for predicting demand for medicines in a medical institution.SOLUTION: An apparatus 100 acquires prescription information corresponding to a plurality of prescriptions received by a pharmacy (S201). Then, the apparatus 100 distinguishes demand of patients coming to the pharmacy in the future again for treatment of the same diseases, from demand of first-time patients, and makes predictions separately (S202 and S203). The apparatus generates final time series of predicted demand quantities for every medicine based on each-day values of prediction results related to the respective medicines (S204).SELECTED DRAWING: Figure 2

Description

TECHNICAL FIELD The present invention relates to an apparatus, method, and program for predicting drug demand in medical institutions.

Medical institutions such as pharmacies purchase medicines from distributors such as wholesalers, prepare and deliver one or more prescribed medicines to patients who visit the hospital as necessary. Since medical institutions need to dispense drugs in response to patient requests, they strive to secure sufficient inventories of both types and quantities.

A code for identifying each product is attached to each medicine, and even if the medicinal ingredients are the same, there are some medicines with different codes set as different products if the manufacturer is different, for example. In actual distribution, each product is sold in a certain unit, and there are a sales packaging unit that is the minimum order unit and an original packaging unit that packs a plurality of products in the sales packaging unit. A code for identifying these packaging units is also assigned, and a code system called GS1 is used internationally. For example, a sales packaging unit of 10 PTP packaging sheets containing 10 "Loxonin tablets 60 mg" is assigned a code "14987081105400" according to the GS1 system. A medical institution designates such a code and purchases the necessary kind and quantity of goods.

JP 2020-166859 A

However, at present, medical institutions cannot predict when and how much demand will occur for which type of drug, and order processing is based on past experience, so inventory is insufficient. Inefficiency is occurring, such as requesting delivery in a hurry and conversely having excessive inventory.

Patent Literature 1 describes a technology for learning the transition of delivery of pharmaceuticals in medical institutions and predicting the demand for the pharmaceuticals according to the learning result, but the inventors found a new idea. .

The present invention has been made in view of such problems, and its object is to provide an unprecedented device, method, and program for predicting demand for drugs in medical institutions.

To achieve these objects, a first aspect of the present invention is a method for predicting drug demand in a medical institution, comprising obtaining prescription information corresponding to a first prescription received by the medical institution. a second step of calculating, for each drug contained in said prescription information, a predicted patient visit date and predicted demand quantity associated with said prescription information; and said first and second steps is repeated for a second prescription received by said medical institution; a fourth step of calculating a demand quantity, and a time series of the predicted demand quantity for each drug based on the value of the predicted demand quantity for each drug obtained in the third step and the fourth step for each day; and a fifth step of generating

A second aspect of the present invention is the method of the first aspect, wherein the second step estimates one or more diseases for which one or more drugs included in the prescription information are used. calculating at least one of the predicted visit date and the predicted quantity of demand using a prediction model for each disease.

A third aspect of the present invention is the method of the first aspect, wherein the second step is the step of estimating a single disease for which one or more drugs included in the prescription information are used. and calculating the predicted number of times the patient will revisit for one or more drugs used for the disease; and ending the prediction process if the predicted number of times is 0, and the predicted number of times is 1. If the above is the case, the step of calculating the predicted visit date for the next and subsequent visits and the predicted demand quantity of one or more medicines to be used for the disease.

A fourth aspect of the present invention is the method of the first aspect, wherein the second step is a step of estimating a plurality of diseases for which one or more drugs included in the prescription information are used. , for each of the plurality of diseases, calculating the predicted number of times the patient will revisit for one or more drugs used for the disease; and terminating the prediction process if the predicted number of times is 0. , calculating the predicted visit date for the next and subsequent visits and the predicted demand amount for each disease of one or more drugs used for the disease when the predicted number of times is one or more, and is calculated by multiplying each disease-specific predicted demand quantity by a value corresponding to the weight assigned to each of the plurality of diseases, and totaling the results for the plurality of diseases.

A fifth aspect of the present invention is the method of the third or fourth aspect, wherein in the second step, at least part of the patient's past prescription information is used to calculate the predicted number of times. .

A sixth aspect of the present invention is the method according to any one of the first to fifth aspects, wherein the second step calculates the predicted visit date using the visit history for each disease.

A seventh aspect of the present invention is the method according to any one of the first to sixth aspects, wherein the second step converts the predicted visit date into the week and day of the week of the next and subsequent visits. Calculated based on individual prediction results.

Moreover, an eighth aspect of the present invention is the method according to any one of the first to seventh aspects, wherein the fourth step includes obtaining a plurality of prescriptions received by a plurality of pharmacies in the same area as the medical institution. a step of generating a drug-specific time series of the quantity demanded by a patient visiting for the first time in the area based on a plurality of prescription information based on; , generating a clustered time series of quantity demanded; and for each drug contained in prescription information corresponding to said first and second prescriptions received by said medical institution during a first period of a year. In addition to the prescription information, the predicted visit date and predicted demand quantity of the first-visit patient are calculated using the feature amount obtained from the cluster-specific time series in the second period preceding the first period of the cluster to which the drug belongs. and calculating

A ninth aspect of the present invention is a program for causing a computer to execute a method of predicting drug demand in a medical institution, wherein the method corresponds to a first prescription received by the medical institution. a first step of obtaining prescription information for the prescription, a second step of calculating a predicted patient visit date and a predicted demand quantity associated with the prescription information, for each drug contained in the prescription information; and the second step for a second prescription received by the medical institution; a fourth step of calculating the predicted visit date and the predicted demand quantity; and a fifth step of generating a time series for each drug.

A tenth aspect of the present invention is a device for predicting drug demand in a medical institution, which acquires prescription information corresponding to a first prescription received by the medical institution, and obtains prescription information included in the prescription information. calculating a predicted visit date and a predicted demand quantity for a return visit of the patient associated with the prescription information for each drug that is received by the medical institution; Repeatedly, for each drug contained in the prescription information corresponding to the first and second prescriptions, the predicted visit date and the predicted demand quantity of the first-visit patient are calculated, and each day of the obtained predicted demand quantity for each drug A time series of predicted demand quantity is generated for each drug based on the values in .

According to one aspect of the present invention, prescription information corresponding to a plurality of prescriptions received by a pharmacy is obtained, and demand by patients who visit the pharmacy again in the future and demand by patients who visit the pharmacy for the first time are separately predicted. By doing so, it is possible to provide a new method of predicting the demand for drugs in medical institutions, which has never existed before.

It is a figure which shows the demand prediction apparatus concerning one Embodiment of this invention. It is a figure which shows the outline of the demand forecast method concerning one Embodiment of this invention. It is a figure which shows the flow of the demand forecast method regarding the revisiting patient concerning one Embodiment of this invention.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

FIG. 1 is a demand forecasting device according to one embodiment of the present invention. The device 100 acquires prescription information corresponding to a plurality of prescriptions and generates a time series of predicted demand quantity of each drug included in the prescription information. Prescription information is generated by imaging a two-dimensional code written on a prescription received by a pharmacy with a device having an imaging device installed at the pharmacy, and inputting the generated prescription information to the device 100 by wire or wirelessly. Alternatively, the device 100 can acquire the prescription information corresponding to the prescription received by the pharmacy by electronically receiving it from a device capable of communicating via a computer network such as an IP network. . The forecast demand quantity time series can be a series of daily forecast demand quantities for each drug. Device 100 is preferably a server that provides demand forecasting services to multiple pharmacies.

The device 100 includes a communication unit 101 such as a communication interface, a processing unit 102 such as a processor and a CPU, and a storage unit 103 including a storage device or storage medium such as a memory or hard disk. can be configured by running a program to do so. Device 100 may include one or more devices, computers or servers. The program may include one or more programs, and may be recorded on a computer-readable storage medium as a non-transitory program product. The program can be stored in a storage device or storage medium such as the database 104 accessible from the storage unit 103 or the device 100 via the IP network, and can be executed by the processing unit 102 . Data described below as being stored in storage 103 may be stored in database 104, and vice versa.

FIG. 2 shows an outline of a demand forecasting method according to one embodiment of the present invention. The device 100 acquires prescription information corresponding to multiple prescriptions received by the pharmacy (S201). Next, the device 100 separately predicts demand by patients who will visit the pharmacy again for treatment of the same disease and demand by patients who visit the pharmacy for the first time (S202, S203). Then, by calculating the sum of the values of the forecast results for each drug for each day, the time series of the final predicted demand quantity is generated for each drug (S204). In FIG. 2, predictions for first-visit patients are shown next to predictions for revisiting patients, but the order does not matter. In addition, although FIG. 2 shows the order in which the prediction processing is performed after obtaining the prescription information first, the prescription information may be sequentially obtained when necessary for the prediction processing. In addition, although it was explained that the time series is generated by calculating the sum for each day, more generally, the final forecast demand quantity for each day is calculated based on the value for each day of the forecast results for each drug. Thus, a time series of predicted demand quantity can be generated.

In the following, predictions for returning patients are described first, followed by predictions for first-time patients. The reason why predictions for returning patients and first-time patients are made separately in this way is that the inventors have found that these predictions are highly independent.

Any of the various predictions or estimates described in this specification should preferably take into account probabilistic elements and calculate expected values and upper and lower limits of confidence intervals as a result. For example, a normal distribution can be assumed for the forecasted demand quantity, so that the expected value of the forecasted demand quantity is the predicted expected value for returning patients and the predicted expected value for first-time patients. It is calculated by the sum of In addition, from the additivity of the variance in the normal distribution, the standard deviation is obtained from the square root of the variance obtained for the returning patients and the variance obtained for the first-time patients, and the confidence interval is determined using the standard deviation. However, probabilistic factors need not be considered in all of the various predictions or estimates described herein.

Forecasting for Returning Patients FIG. 3 shows the flow of a demand forecasting method for returning patients according to one embodiment of the present invention. The device 100 performs demand forecasting on prescription information corresponding to each of a plurality of prescriptions received by a pharmacy targeted for demand forecasting, and the results form a time series as a series of forecasted demand quantities. FIG. 3 shows the processing performed on the prescription information corresponding to each prescription.

First, the device 100 estimates a disease for which one or more drugs included in the acquired prescription information are used (S301). Although the prescription r does not mention the disease d for which the patient is being treated, one or more drugs are prescribed to treat one or more diseases of the patient, so the prescription information includes 1 Alternatively, it is theoretically possible to infer one or more diseases of the patient based on multiple drugs. In this embodiment, as an example, the correspondence between one or more drugs and the strength of connection between one or more diseases for which each drug can be used is stored in the storage unit 103, and each drug included in the prescription information is stored in the storage unit 103. It is possible to determine the strength of the connection with each disease by referring to the association based on. Then, by summing up the strength of association with each disease determined for all drugs included in the prescription information for each disease, and estimating the top N diseases (N is an integer equal to or greater than 1) as the patient's disease good. The strength of association value with one or more diseases for which each drug may be used can be determined, for example, according to the frequency with which each drug has been used in the past for each disease.

The strength of the connection between the drug m and the disease d held by the association can be expressed as a weight ω _m,d . A set of one or more drugs included in a prescription r, Mr, and a total value ω _d for each disease can be expressed by the following equation.

The table below shows that drug _{mA was associated with disease d 1 with a strength of 2.0, disease d 2 with a strength of 1.5, and drug m B was associated with} disease d ₁ with a strength of 3.0. , indicating that it is associated with disease d ₃ with a strength of 1.0.

Summed for _each disease, 5.0 for disease d1, _1.5 for disease d2 _, and 1.0 for disease d3. Disease d ₁ is estimated based on the top one, and diseases d ₁ and d ₂ are estimated based on the top two. In the following, an example in which the top one is estimated as a disease will be mainly described. When _N ( _N is an integer equal to or greater than 2) diseases are estimated, a value corresponding to the weight ω _dn assigned to the disease dn in the prediction demand quantity by disease for each disease dn By _multiplying by an and calculating the sum, it is possible to obtain the predicted demand quantity of each medicine based on one prescription r. As an example, the value a _n according to the weight ω _dn assigned to each disease d _n can be expressed as:

Next, the device 100 inputs at least part of the prescription information into the disease-specific prediction model to predict the number of times the patient will visit the clinic again in association with the prescription information (S202). The data to be input includes age, sex, medical department, and the like. If it is 0 times, the prediction process ends, and if it is n times (n is an integer of 1 or more), the date of the first visit to the office is predicted (S303), and the demand quantity is predicted (S304). If n is 2 or more, the date of the second to n-th visit to the office and demand quantity are predicted as necessary. For example, for acute illnesses such as colds and influenza, treatment may be completed in the first visit, with no repeat visits. If the patient has a chronic disease such as diabetes, it is expected that the patient will visit the hospital several times or more.

A predictive model for the number of revisit stations can be generated by supervised learning, more specifically by supervised learning using gradient boosting such as LightGBM and XGBoost, but it is currently available. Yes, it can be generated by any machine learning method that will be available in the future. In addition, the prediction model of the number of revisits is not limited to one generated by machine learning, and may be any model in which at least part of the prescription information is input and the number of revisits is output accordingly. For higher accuracy, it may be preferable to include at least a portion of the patient's past prescription information associated with the prescription information in the input data.

Prediction of return visit dates may also be made by predictive models generated by machine learning or any other currently available and future available method, but the patient or prescription with which the prescription information is associated It can also be performed using a visit interval distribution obtained from past visit histories for each disease of a patient having a disease with which information is associated. The station visit interval distribution may be calculated in units of days, but may be calculated in units of weeks, and by using the station visit day-of-the-week distribution, a specific date of visit may be predicted.

Based on the study of the inventors, it is preferable to use the fact that the prediction of the interval between visits to the station in units of weeks and the prediction of the day of the week to visit the station are highly independent. Based on the results of predicting these separately, it is possible to predict the return date, which is a product event.

Demand volume forecasts can also be made by forecasting models generated by machine learning or any other currently available or later available technique, e.g. It can also be performed using the sales volume distribution obtained from the past sales history for each disease for patients.

Prediction of First-Time Patients The arrival date and demand quantity of first-time patients may be predicted using an appropriate time-series prediction algorithm based on past prescription information associated with the pharmacy.

Prescription information associated with a pharmacy may include information based on prescriptions received by the pharmacy that is the target of the demand forecast, as well as information based on prescriptions received by other pharmacies associated with the pharmacy. For example, by using prescription information based on prescriptions received by one or more other pharmacies in the same area as the pharmacy, the seasonality of drug demand in the area is determined, and based on the seasonality, first-time patients can make predictions about

In more detail, the following examples are given. First, using multiple prescription information based on multiple prescriptions received by multiple pharmacies in the same area as the target pharmacy for demand forecast over a period of one year or more, we estimate the amount of drugs demanded by first-time patients in that area. Generate another time series. Each drug-specific time series spans at least one year and reflects differences at different times of the year. Here, when there is not enough data, when the need is not high, etc., it is not necessary to generate a time series for some drugs. Also, less than one year is not necessarily insufficient. Next, a plurality of drugs for which time series have been generated by a clustering method such as the k-medoids method are classified into a plurality of clusters, and cluster-specific time series of demand quantities are generated. Each time series described above can be a series of demand quantities on a daily basis, or can be a series of demand quantities on a predetermined unit basis, such as weekly basis. Then, from the multiple cluster-specific time series, the cluster-specific time series of the cluster to which the drug targeted for demand forecasting belongs is obtained, and the cluster-specific time series is used to predict the expected number of patients visiting the hospital for the first time for the drug. Calculate the station date and forecast demand quantity. It should be added that this prediction method for first-time patients is, in itself, an unprecedented drug demand prediction method for medical institutions, separate from the prediction for revisiting patients.

Specifically, for the drugs targeted for demand forecast, in addition to prescription information received by the pharmacy targeted for demand forecast in the first period of one year, Using the quantity of demand contained in the time series for each cluster in period 2 or the value obtained by performing the required calculation on the quantity of demand, the predicted date of visit to the hospital and the predicted quantity of demand for the first-visit patient can be calculated. As an example, in a pharmacy, when only prescription information based on prescriptions received between January and March for a certain drug is given, the cluster to which the drug belongs is included in the cluster-by-cluster time series from April onwards. By using the quantity demanded or a value obtained by subjecting it to a required calculation as a feature amount, it is possible to predict with high accuracy how the quantity demanded will change even after April.

In the above explanation, the pharmacy is mainly explained as an example, but the prescription information corresponding to the prescription received by the medical institution can be used without being limited to the pharmacy. It should be understood as "visit".

In the above-described embodiments, if there is no description of "only" such as "based on XX only", "only in response to XX", or "only in case of XX", in this specification, Note that it is envisioned that additional information may also be considered. Also, as an example, it should be noted that the statement ``do b in the case of a'' does not necessarily mean ``always do b in the case of a'' or ``do b immediately after a'' unless explicitly stated otherwise. want to be Moreover, the description of "each a constituting A" does not necessarily mean that A is composed of a plurality of constituent elements, but includes that the constituent elements are singular.

Also, just in case, any method, program, terminal, device, server, or system (hereinafter "method, etc."), even if there is an aspect that operates differently from the operations described in this specification, each of the present invention Aspects are directed to acts that are identical to any of the acts described herein, and that there are acts that are different from those described herein, the methods, etc. It should be added that it is not out of the scope of the embodiments.

In addition, "start" and "end" shown in FIGS. 2 and 3 are merely examples, and the method according to the present embodiment is always started by the illustrated procedure and always ends by the illustrated procedure. does not mean that

100 device 101 communication unit 102 processing unit 103 storage unit 104 database

Claims (10)

A method of predicting drug demand in a medical institution, comprising:
a first step of obtaining prescription information corresponding to a first prescription received by the medical institution;
a second step of calculating, for each drug contained in the prescription information, a predicted patient visit date and a predicted demand quantity associated with the prescription information;
a third step of repeating said first and second steps for a second prescription received by said medical institution;
Fourth, for each drug included in the prescription information corresponding to the first and second prescriptions, calculating the predicted visit date and the predicted demand quantity of the first-visit patient based on the prescription information associated with the medical institution. a step of
and a fifth step of generating a time series of predicted demand quantity for each drug based on the daily value of the predicted demand quantity for each drug obtained in the third step and the fourth step. 2. The method of claim 1, wherein
The second step includes
estimating one or more diseases for which one or more drugs included in the prescribing information are used;
At least one of the predicted visit date and the predicted demand quantity is calculated using a prediction model for each disease. The method of claim 1, wherein
The second step includes
estimating a single disease for which one or more drugs included in the prescribing information are used;
calculating a predicted number of return visits for said patient for one or more medications used for said disease;
When the predicted number of times is 0, the prediction process is terminated, and when the predicted number of times is 1 or more, the predicted visit date for the next and subsequent visits and the predicted demand amount of one or more medicines used for the disease are calculated. step. The method of claim 1, wherein
The second step includes
estimating a plurality of diseases for which one or more drugs included in the prescription information are used;
For each of the plurality of diseases,
calculating a predicted number of return visits for the patient for one or more medications for the disease;
When the number of predictions is 0, the prediction process is terminated, and when the number of predictions is 1 or more, the predicted visit date for the next and subsequent visits and the predicted demand quantity for each disease of one or more drugs used for the disease are calculated. and calculating
The predicted demand quantity calculated in the second step is calculated by multiplying the predicted demand quantity for each disease by a value corresponding to the weight assigned to each of the plurality of diseases, and totaling the results for the plurality of diseases. do. A method according to claim 3 or 4,
In the second step, at least part of the patient's past prescription information is used to calculate the predicted number of times. A method according to any one of claims 1 to 5,
The second step calculates the predicted visit date using the visit history for each disease. A method according to any one of claims 1 to 6,
In the second step, the predicted visit date is calculated based on results of individually predicting visit weeks and visit days of the week from the next time onward. A method according to any one of claims 1 to 7,
The fourth step is
generating a drug-specific time series of quantity demanded by patients visiting for the first time in the region based on a plurality of prescription information based on a plurality of prescriptions received by a plurality of pharmacies in the same region as the medical institution;
Classifying a plurality of drugs for which time series by drugs have been generated into a plurality of clusters to generate a time series of demand quantity by clusters;
For each drug included in the prescription information corresponding to the first and second prescriptions, in addition to the prescription information received by the medical institution during the first period of the year, and calculating a predicted visit date and a predicted demand quantity of the first-visit patient using the feature values obtained from the cluster-specific time series in the second period prior to the first period. A program for causing a computer to execute a method of predicting drug demand in a medical institution, the method comprising:
a first step of obtaining prescription information corresponding to a first prescription received by the medical institution;
a second step of calculating, for each drug contained in the prescription information, a predicted patient visit date and a predicted demand quantity associated with the prescription information;
a third step of repeating said first and second steps for a second prescription received by said medical institution;
Fourth, for each drug included in the prescription information corresponding to the first and second prescriptions, calculating the predicted visit date and the predicted demand quantity of the first-visit patient based on the prescription information associated with the medical institution. a step of
and a fifth step of generating a time series of predicted demand quantity for each drug based on the daily value of the predicted demand quantity for each drug obtained in the third step and the fourth step. A device for predicting drug demand in a medical institution, comprising:
Prescription information corresponding to the first prescription received by the medical institution is obtained, and for each drug included in the prescription information, a predicted visit date and a predicted demand quantity of the patient's return visit associated with the prescription information are calculated. calculate,
repeating said calculation of predicted visit dates and predicted quantity demanded for a second prescription received by said medical institution;
For each drug contained in the prescription information corresponding to the first and second prescriptions, based on the prescription information associated with the medical institution, calculate the predicted visit date and predicted demand quantity of the first-visit patient;
Based on the obtained values of the predicted demand quantity for each drug for each day, a time series of the predicted demand quantity is generated for each drug.

Images