JP2022035097A - Receipt data significance assessment program, receipt data significance assessment method, and information processing device - Google Patents

Abstract

To appropriately assess the significance of medical care items included in receipt data.SOLUTION: The present invention causes a computer to execute the process of performing an assessment process of extracting second data 32c from first data 32a related to a plurality of receipts in accordance with a condition 32b for extracting the calculation items of receipts and assessing, on the basis of third data 32d derived by extracting data for which first calculation items were requested from the first or second data 32a or 32c and fourth data 32e extracted at random from data derived from the second data 32c by excluding the data for which the first calculation items were requested, a significant difference pertaining to the first calculation items with regard to the plurality of first medical care items included in the third and fourth data 32d and 32e, and generating learning data 32f in which one or more of second medical care items having been assessed as having a significant difference are extracted from the third and fourth data 32d and 32e.SELECTED DRAWING: Figure 7

Description

The present invention relates to a receipt data significance determination program, a receipt data significance determination method, and an information processing apparatus.

The medical management system used in medical institutions such as hospitals shall have a function to generate and output medical receipt data based on electronic medical record information including medical records, nursing records and operation information input by doctors or nurses. There is. A "receipt" is a monthly medical fee statement submitted by a medical institution to an insurance institution.

The receipt data will be scrutinized by the insurance institution, and if there is an incorrect receipt, the receipt will be returned to the medical institution. The medical receipts returned from the insurance institution are mainly the medical receipts that are overcharged, and in many cases, the medical receipts for which the medical fee is not calculated are not returned. Therefore, if the medical institution overlooks the medical treatment fee that can be calculated based on the medical treatment result due to omission of the check, the medical profit that should be originally obtained cannot be obtained.

Since the medical receipt data generated by the medical management system may contain thousands of items or more, it is not possible for the medical office staff to thoroughly examine (check) the medical receipt data before submitting it to the insurance institution. Have difficulty.

By the way, there is also known a method of detecting whether there is a possibility of calculation omission using AI (Artificial Intelligence) such as machine learning based on the uncorrected medical receipt and electronic medical record information output from the medical management system. There is.

International Publication No. 2008/047835 Pamphlet

In the above-mentioned method using machine learning, data is extracted from the receipt data for each calculation item to be learned, and machine learning is performed by selecting medical items with significant differences to obtain a highly accurate receipt calculation omission detection model. Can be generated.

However, some calculation items may be limited to gender, age, etc., and medical care items unrelated to the calculation items may be selected as significant items. For example, when learning about a calculation item peculiar to a male (for example, an adult male), a female or pediatric receipt data may be used as data for determining a significant difference. In such a case, the accuracy of receipt calculation omission detection may decrease due to machine learning using learning data that includes medical items unrelated to the calculation items.

In one aspect, one of the objects of the present invention is to appropriately determine the significance of the medical treatment items included in the receipt data.

In one aspect, the receipt data significance determination program may cause the computer to perform the following processing. In the process, the second data may be extracted from the first data regarding a plurality of receipts according to the conditions for extracting the calculation items of the receipts. In the processing, the third data obtained by extracting the data for which the first calculation item is requested from the first data or the second data, and the data excluding the data for which the first calculation item is requested among the second data. Based on the fourth data randomly extracted from the third data, a determination process for determining a significant difference in the first calculation item for a plurality of first medical care items included in the third data and the fourth data may be performed. .. The process may generate learning data obtained by extracting one or more second medical care items determined to have a significant difference by the determination process from the third data and the fourth data.

In one aspect, the significance of medical items included in the receipt data can be appropriately determined.

It is a figure for demonstrating an example of the system which concerns on one Embodiment. It is a figure for briefly explaining an example of the process of the receipt calculation omission detection system which concerns on one Embodiment. It is a block diagram which shows the functional structure example of the calculation omission detection system which concerns on one Embodiment. It is a figure which shows an example of the input data. It is a figure explaining an example of the machine learning by the system which concerns on the comparative example of one Embodiment. It is a figure explaining an example of the machine learning of the receipt calculation omission detection by the system which concerns on the comparative example of one Embodiment. It is a figure explaining an example of the learning data generation processing by the detection system (generation part) which concerns on one Embodiment. It is a figure which shows an example of the rule applied data. It is a figure which shows an example of the calculation item billed data. It is a figure which shows an example of the data for significant difference calculation. It is a figure for demonstrating an example of a significant difference determination process. It is a figure which shows an example of the learning data. It is a figure which shows an example of the determination data. It is a flowchart explaining the operation example of the learning process of the detection system which concerns on one Embodiment. It is a flowchart explaining the operation example of the determination process of the detection system which concerns on one Embodiment. It is a block diagram which shows the hardware (HW) configuration example of the computer which realizes the function of a detection system.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. However, the embodiments described below are merely examples, and there is no intention of excluding the application of various modifications or techniques not specified below. For example, the present embodiment can be variously modified and implemented without departing from the spirit of the present embodiment. In the drawings used in the following description, the parts with the same reference numerals represent the same or similar parts unless otherwise specified.

[1] Embodiment [1-1] Explanation of a system FIG. 1 is a diagram for explaining an example of a system 1 according to an embodiment. As shown in FIG. 1, the system 1 may include a medical management system 1a and a receipt calculation omission detection system (hereinafter, may be simply referred to as a “detection system”) 10.

The medical management system 1a is a system used in a medical institution such as a hospital for managing and supporting operations such as medical office work. For example, the medical management system 1a may generate receipt information 1b including a plurality of receipts and output it to the detection system 10 as an output of the medical fee billing office work.

Information regarding medical receipts may be input to the medical management system 1a by a doctor, a nurse, a medical staff, or the like.

Further, the system 1 may include the electronic medical record information 1c and output the information of the electronic medical record information 1c to the detection system 10. Information on a plurality of electronic medical records such as medical records 1d, nursing records, and operation records may be input to the electronic medical record information 1c by a doctor, a nurse, a medical staff, or the like. The electronic medical record information 1c may be stored in, for example, the medical management system 1a or may be stored in another system.

The detection system 10 detects the omission of the receipt calculation by the AI 11 based on the input receipt information 1b and the electronic medical record information 1c, and outputs the detection result 12. As an example, the detection result 12 may include information indicating the probability of omission of calculation of the medical fee for each receipt included in the receipt information 1b.

For example, in the system 1, an operator such as a medical office worker or a system such as the medical management system 1a generates the modified receipt information 1e obtained by modifying the receipt information 1b based on the detection result 12, and the modified receipt information 1e. Submit 1e to the insurance institution.

FIG. 2 is a diagram for briefly explaining an example of processing of the receipt calculation omission detection system 2 according to the embodiment. The detection system 2 is an example of the detection system 10 shown in FIG.

As shown in FIG. 2, in the processing by the detection system 2, the model creation phase 2A for creating the detection model 23 and the model application phase 2B for outputting the calculation omission candidate recommendation 26 using the created detection model 23 are exemplified. And may be included.

(Model creation phase 2A)
In the model creation phase 2A, the detection system 2 executes machine learning by the machine learning unit 22 based on the past calculation result 21, and generates the detection model 23. For example, the detection system 3 may learn different calculation rules for each calculation item obtained from the calculation result 21 by the machine learning unit 22, as shown by the reference numeral (i) in FIG.

The calculation result 21 may include, for example, receipt information 21a including a plurality of receipts and electronic medical record information 21b including a plurality of electronic medical records. For example, as shown by reference numeral (ii) in FIG. 2, the detection system 2 refers to the electronic medical record information 21b in order to extract learning information such as a numerical value and a character string from the text of the electronic medical record information 21b. Language processing using natural language processing technology may be performed.

Further, the calculation result 21 may include correct answer (teacher) data, for example, a calculation result as a correct answer label. The calculation result may include, for example, information indicating the calculation item 21c calculated for the data included in the receipt information 21a and the electronic medical record information 21b. As described above, the calculation result 21 is an example of the training data set as shown by the reference numeral (iii).

The detection system 2 may extract appropriate data to be used for machine learning from a plurality of receipt information 21a and a plurality of electronic medical record information 21b based on predetermined conditions regarding the calculation item 21c by a method described later.

The machine learning unit 22 learns the training data set. For example, the machine learning unit 22 ends the learning when the learning is executed a predetermined number of times, or when the error between the output value of the neural network and the correct answer data becomes smaller than the predetermined value. Then, the machine learning unit 22 stores various parameters and the like as the detection model 23.

(Model application phase 2B)
In the model application phase 2B, the detection system 2 acquires a calculation omission candidate recommendation 26 by inputting a new uncorrected receipt 24 and a new electronic medical record 25 into the detection model 23.

As shown in FIG. 2, the calculation omission candidate recommendation 26 may optionally include the target patient, the detection date, the calculation omission item, and the calculation omission probability item. The target patient may be the patient's name in the uncorrected receipt 24 and the electronic medical record 25 in which the calculation omission is detected. The detection date may be a time stamp of the date and time when the calculation omission was detected. The calculation omission item is information indicating a calculation item 21c in which a calculation omission is detected, for example, a label. The calculation omission probability is information indicating the probability that a calculation omission has occurred in a calculation omission item.

[1-2] Functional Configuration Example FIG. 3 is a block diagram showing a functional configuration example of the calculation omission detection system 3 according to the embodiment. As shown in FIG. 3, the detection system 3 may optionally include a communication unit 31, a memory unit 32, and a control unit 33.

The communication unit 31 communicates information with an information processing device such as another computer or server via a network (not shown). The communication unit 31 may receive various data such as data used in the learning phase (model creation phase) and data used in the determination phase (model application phase) from the information processing device used by the operator, for example. Further, the communication unit 31 may transmit various information such as a learning result and a determination result to the information processing device.

The memory unit 32 is an example of a storage area, and stores various data used by the detection system 3. As shown in FIG. 3, the memory unit 32 is exemplified by input data 32a, extraction rule 32b, rule applied data 32c, calculation item requested data 32d, significant difference calculation data 32e, learning data 32f, detection model. 32 g and the determination data 32h may be storable.

In the following description, for convenience, each of the data 32a to 32f and 32h stored in the memory unit 32 will be described as table format data, but the present invention is not limited to this, and various types such as a DB (Database) or an array are used. It may be in a data format.

The control unit 33 performs various controls related to machine learning. As shown in FIG. 1, the control unit 33 may optionally include an acquisition unit 33a, a generation unit 33b, a learning unit 33c, and a determination unit 33d.

The acquisition unit 33a may acquire the input data 32a based on the data input via the communication unit 31, and store the acquired input data 32a in the memory unit 32, for example. The input data 32a is data used in the learning phase, and may be based on, for example, the receipt information 21a shown in FIG. 2, the electronic medical record information 21b, and the information of the calculation item 21c.

The input data 32a is an example of the first data relating to a plurality of receipts, and is, for example, acquired (for example, received or received) based on a plurality of data input from an external information processing device such as a terminal (not shown) via the communication unit 31. May be generated). As an example of the information processing apparatus, one or both of the medical management system 1a shown in the explanation of FIG. 1 and the system for outputting the electronic medical record information 1c can be mentioned.

FIG. 4 is a diagram showing an example of input data 32a. As shown in FIG. 4, the input data 32a may optionally include an ID, a label, a gender, an age, and items 1 to N (where N is an integer of 1 or more).

The ID is the learning target data for which the receipt calculation is omitted, in other words, the identification information for identifying one entry. For example, the ID may be the identification information of the entry of the receipt information 21a.

The label is information indicating whether or not each entry corresponds to a calculation item to be detected for omission of calculation, in other words, whether or not a calculation item is requested. The label is an example of correct answer data, and may be set based on the information of the calculation item 21c, for example. In the example of FIG. 4, information indicating whether or not each entry corresponds to "calculation item 1" is set in the label.

"Calculation item 1" is a calculation item to be detected and is an example of the first calculation item. In one embodiment, an example of the input data 32a focusing on "calculation item 1" as a detection target is shown, but the present invention is not limited to this. The memory unit 32 may store input data 32a (input data 32a including a label indicating whether or not it corresponds to another calculation item) focusing on another calculation item as a detection target.

Gender and age are information indicating the gender and age of the medical receipt subject. For example, the gender and age may be set based on the receipt information 21a and the electronic medical record information 21b.

The medical treatment item 1 to the medical treatment item N is an example of one or more first medical treatment items to which the entry data is related, and is an example of an explanatory variable item. Various items related to billing of medical fees may be set as medical items. For example, the medical treatment item 1 to the medical treatment item N may be set based on the receipt information 21a.

Further, the acquisition unit 33a may acquire the determination data 32h based on the data input via the communication unit 31, and store the acquired determination data 32h in the memory unit 32, for example. The determination data 32h is data used in the determination phase, and may be based on, for example, the information of the new uncorrected receipt 24 and the new electronic medical record 25 shown in FIG. The determination data 32h may have, for example, a data structure excluding the label (correct answer data) of the input data 32a.

The generation unit 33b generates the learning data 32f based on the input data 32a. Hereinafter, an example of the generation processing of the learning data 32f by the generation unit 33b will be described.

FIG. 5 is a diagram illustrating an example of machine learning by the system 100 according to the comparative example of one embodiment. As shown in FIG. 5, the system 100 generates the model 130 by executing the learning 120 using the data 110 in which a large amount of data of each label exists. In the example of FIG. 5, since a large amount of data 110 exists, the learning 120 can be appropriately performed even when there are many explanatory variable items.

FIG. 6 is a diagram illustrating an example of machine learning of receipt calculation omission detection by the system 200 according to the comparative example of one embodiment. As shown in FIG. 6, the input data 32a may contain only a small amount of data for each calculation item (label) (data corresponding to calculation item 1 in the example of FIG. 6). Since the amount of data corresponding to the calculation item 1 is small, it is difficult for the system 200 to perform appropriate learning when there are a large amount of medical care items (explanatory variable items).

Therefore, the system 200 performs a data extraction process (see reference numeral a) from the input data 32a, and extracts the calculation item requested data 210 and the significant difference calculation data 220 from the input data 32a. The calculation item requested data 210 is data obtained by extracting (see reference numeral b) only the data for which the calculation item to be detected is requested from the input data 32a, for example, “the calculation item 1 is requested”. The significance difference calculation data 220 is data randomly extracted from the input data 32a other than the data for which the calculation item 1 has been requested (see reference numeral c).

The system 200 performs a significant difference determination process (see reference numeral d) for each medical treatment item, and uses only the medical treatment item having "significant difference" in the input data 32a as the learning data 230 for learning (see reference numeral f). By doing so, the detection model 240 is generated.

As described above, the system 200 according to the comparative example can reduce the number of medical care items to be learned by extracting only the medical care items that are effective for learning. In the example of FIG. 6, the system 200 extracts only "medical treatment item 2" and "medical treatment item 4" from "medical treatment item 1" to "medical treatment item N" as medical treatment items having "significant difference". Is used as the learning data 230. As a result, it is possible to generate the detection model 240 of the receipt calculation omission with high accuracy.

However, there may be items limited to gender, age, etc. in the calculation items, and medical care items unrelated to the calculation items may be selected as significant items. For example, when learning (see reference numeral f) about a calculation item peculiar to an adult male, receipt data of a woman or a child may be randomly extracted as data 220 for determining a significant difference. In such a case, medical care items unrelated to the calculation items may be extracted as “significantly different” (see reference numeral e). As described above, since the learning data 230 includes medical treatment items that are not related to the calculation items to be detected, the accuracy of the receipt calculation omission detection may decrease.

FIG. 7 is a diagram illustrating an example of a learning data 32f generation process by the detection system 3 (generation unit 33b) according to the embodiment.

As shown in FIG. 7, the generation unit 33b may perform data extraction processing (see reference numeral A) from the input data 32a based on the extraction rule 32b to generate the rule-applied data 32c and store it in the memory unit 32. ..

The extraction rule 32b is an example of a condition for extracting the calculation item of the receipt from the input data 32a, and as illustrated in FIG. 7, a rule for extracting data (entry) from the input data 32a is provided for each calculation item. show. The extraction rule 32b may be set in advance in the detection system 3, for example.

The rule may include, for example, a limitation of the attributes corresponding to the calculated items. The attribute may include, for example, one or both of the sex and age of the subject patient. The attribute restriction may include, for example, designation of gender classification, age range, and the like. In the example of FIG. 7, the rule of "calculation item 1" is gender: male and age: 20 years or older. The attributes are not limited to gender and age, and may include various items. Further, in the extraction rule 32b, for example, a rule that a restriction of at least one or all attributes is not set (see "calculation item 2" and "calculation item 5") may be permitted.

For example, the generation unit 33b may generate the rule-applied data 32c by extracting an entry satisfying the extraction rule 32b of the calculation item to be detected from the input data 32a. The rule-applied data 32c is an example of the second data.

FIG. 8 is a diagram showing an example of rule-applied data 32c. In the example of FIG. 8, the generation unit 33b corresponds to the input data 32a according to the rule of gender: male and age: 20 years or older corresponding to “calculation item 1” in the extraction rule 32b of FIG. The case where only the data (entry) is extracted is shown. As illustrated in FIG. 8, the rule-applied data 32c includes data (entry) extracted from the input data 32a shown in FIG. 4 by applying the rule corresponding to the “calculation item 1”. In other words, the rule-applied data 32c is data in which the data (entries) of "ID002", "ID004", "ID005", and "ID010" that are not subject to the rule are excluded from the input data 32a. It can be said that.

In this way, the generation unit 33b sets conditions such as the age and gender of the target for each calculation item in advance, and before performing the significant difference determination for each medical treatment item, the generation unit 33b is based on the age, gender and the like. Select the receipt data to be used.

As described above, the generation unit 33b is an example of the extraction unit that extracts the rule-applied data 32c from the input data 32a relating to the plurality of receipts according to the extraction rule 32b that extracts the calculation items of the receipts.

As illustrated in FIG. 7, the generation unit 33b has a significant difference between the calculation item requested data 32d, which is an example of the third data, and the calculation item requested data 32d, which is an example of the fourth data, from the rule-applied data 32c generated as described above. The calculation data 32e may be extracted and stored in the memory unit 32.

FIG. 9 is a diagram showing an example of calculation item requested data 32d. The calculation item requested data 32d is data obtained by extracting only the data "corresponding to the calculation item" (requesting the calculation item) to be detected from the rule applied data 32c. In the example of FIG. 9, the calculation item requested data 32d is data obtained by extracting only the data of “ID001”, “ID003”, “ID006”, and “ID008” corresponding to the calculation item 1 from the rule applied data 32c. (See reference numeral B in FIG. 7).

The generation unit 33b may generate the calculation item requested data 32d by extracting only the data "corresponding to the calculation item" (requesting the calculation item) to be detected from the input data 32a. This is because the data for which the calculation item of the detection target is requested is the data that matches the extraction rule 32b of the calculation item of the detection target.

FIG. 10 is a diagram showing an example of significant difference calculation data 32e. The data 32e for significant difference calculation is narrowed down by the extraction rule 32b from the rule applied data 32c, and only the data that "does not correspond to the calculation item" (the calculation item is not requested) to be detected is randomly selected (FIG. 10). (All in the example) It is the extracted data. In other words, the significant difference calculation data 32e is data randomly extracted from the data obtained by excluding the data for which the calculation item to be detected is requested from the rule-applied data 32c. In the example of FIG. 10, the significant difference calculation data 32e is data obtained by extracting only the data of “ID007” and “ID009” that do not correspond to the calculation item 1 from the rule-applied data 32c (see reference numeral C in FIG. 7). ..

As illustrated in FIG. 7, the generation unit 33b performs significance difference determination processing (see reference numeral D) for each medical treatment item based on the calculation item requested data 32d and the significance difference calculation data 32e.

FIG. 11 is a diagram for explaining an example of the significant difference determination process. As shown in FIG. 11, the generation unit 33b is related to the medical care items diagnosed when the calculation item (medical practice) to be detected is calculated from the past calculation results, such as the disease name and other medical care practices. Test sex. As the past calculation results, for example, the calculation item requested data 32d and the significant difference calculation data 32e may be used.

In FIG. 11, "special diet addition" is taken as an example of the calculation item of the detection target, and each item related to the name of injury / illness: forearm fracture, headache, and diabetes will be described as an example. As shown in FIG. 11, the generation unit 33b determines whether or not there is a medical treatment item related to the name of injury or illness (presence or absence) for each medical treatment item regarding the “special food addition” which is the calculation item of the detection target, and “special food”. The presence or absence of a relationship may be tested (determined) based on whether or not "addition" has been calculated. Whether or not the "special meal addition" has been calculated can be obtained from the calculation item requested data 32d if it is calculated, and from the significance difference calculation data 32e if it is not calculated.

For example, as shown in FIG. 11, the generation unit 33b aggregates the number of data (entries) corresponding to the combination of the presence / absence of medical treatment items and whether or not the calculation items have been calculated, and the relationship is based on the aggregation result. It may be determined whether or not there is a sex, in other words, whether or not there is a significant difference.

Various methods such as a significant difference test may be used to determine the presence or absence of a significant difference. For example, the generation unit 33b has at least one of the total number when there is a medical treatment item and the calculation item is calculated, the ratio of the total number to the total of any one or more of the other totals, and the like. Based on this, the presence or absence of a significant difference may be determined.

As illustrated in FIG. 11, the generation unit 33b indicates that the total number of the following (i) exceeds the threshold value and / or the total number of the following (i) and the following (ii) to (i) for the medical treatment item of diabetes. If the ratio to the total number of aggregates of any one or more of iv) exceeds the threshold value, it may be determined that there is a significant difference. The generation unit 33b may determine the presence or absence of a significant difference based on a combination other than the above.
(I) Total number “800” when diabetic and calculated
(Ii) Total number “1200” when there is no diabetes and it is calculated
(Iii) Total number “500” when there is diabetes and it is not calculated
(Iv) Total number “10000” when there is no diabetes and it is not calculated

On the other hand, as illustrated in FIG. 11, the generation unit 33b determines whether or not there is a significant difference in each of the medical treatment items of forearm fracture and headache, and determines that there is no relationship, in other words, there is no significant difference. You can do it.

As described above, the determination process is based on the calculation item requested data 32d and the significant difference calculation data 32e, and is a detection target for a plurality of medical care items included in the calculation item requested data 32d and the significant difference calculation data 32e. It can be said that this is the process of determining the significant difference in the calculation items of. Therefore, the generation unit 33b is an example of the significance determination unit that executes the significance determination process.

The detection system 3 according to the embodiment may output the result of the significant difference determination process by the generation unit 33b. The output of the significant difference determination processing result is, for example, display on an output device such as a monitor, storage in a memory unit 32 or a storage device, or to the outside of the detection system 3 (for example, an information processing device) via a communication unit 31. Transmission, etc.

As described above, the detection system 3 may be regarded as, for example, a significance determination device for determining a significance difference (significance) regarding the calculation item to be detected for a plurality of medical care items.

As illustrated in FIG. 7, the generation unit 33b extracts only the medical care items determined to be “significantly different” in the significant difference determination process from the calculation item requested data 32d and the significant difference calculation data 32e. (Refer to reference numeral E), the learning data 32f may be generated. Then, the generation unit 33b may store the learning data 32f in the memory unit 32. Calculation item The medical treatment item determined to have "significant difference" among the requested data 32d and the significant difference calculation data 32e is an example of one or more second medical treatment items.

The generation unit 33b may generate the learning data 32f by extracting only the medical treatment items determined to be "significantly different" from the rule-applied data 32c.

FIG. 12 is a diagram showing an example of learning data 32f. In the example of FIG. 12, the medical care items related to the injury / illness name: forearm fracture and headache shown in FIG. 11 are “medical treatment item 2” and “medical treatment item 4”, respectively, and the medical treatment item related to injury / illness name: diabetes is “medical treatment item 1”. Imagine a case. Further, in the example of FIG. 12, it is assumed that "medical treatment item 3", "medical treatment item 5" to "medical treatment item N" are also determined to be "significantly different" in the significant difference determination processing.

As shown in FIG. 12, the learning data 32f has "medical treatment item 1", "medical treatment item 3", "medical treatment item 5" to "medical treatment item N" determined to be "significantly different" from the rule-applied data 32c. Is the extracted data. In other words, the learning data 32f is data obtained by excluding "medical treatment item 2" and "medical treatment item 4" determined to be "no significant difference" from the rule-applied data 32c.

As described above, the generation unit 33b is an example of the learning data generation unit that generates the learning data 32f.

As described above, the generation unit 33b can construct a general-purpose detection model 32g by the learning unit 33c, which will be described later, by setting the medical care items that are highly related to the calculation items of the detection target in the learning data 32f. .. At this time, the generation unit 33b may suppress the use of the country definition or the like for the generation of the learning data 32f.

Further, the generation unit 33b determines the significance (significance) of the data under the same conditions regarding the calculation item to be detected (for example, calculation item 1) by the significance determination process using the significance calculation data 32e. , It is possible to reduce the number of cases where unrelated medical items are judged to be "significantly different".

Returning to the explanation of FIG. 7, the learning unit 33c is an example of the machine learning unit 22 shown in FIG. 2, and the machine learning of the detection model 32g, for example, the creation of the detection model 32g, based on the learning data 32f generated by the generation unit 33b. Or you may update. Machine learning of the detection model 32g may be performed by inputting the learning data 32f into the inference of machine learning.

For example, the learning unit 33c inputs the learning data 32f into the machine learning in response to the input of the learning instruction so that the error becomes smaller based on the error between the inference result of the machine learning and the teacher data (label). To learn. As an example, the learning unit 33c learns machine learning parameters so that the determination accuracy is high.

The learning unit 33c may end the learning when the learning is executed a predetermined number of times or when the error becomes smaller than the predetermined value, and various parameters and the like may be stored in the memory unit 32 as the detection model 32g. As a learning method, various methods such as an error back propagation method can be adopted.

As described above, the learning unit 33c is an example of the model learning unit that performs machine learning of the detection model 32g for detecting the omission of calculation of the medical treatment fee of the medical receipt using the learning data 32f.

After learning the detection model 32g by the learning unit 33d, the determination unit 33d acquires the determination result (detection result) by the learned detection model 32g with the new determination data 32h as an input. The determination data 32h is an example of the fifth data. The determination unit 33d may output a calculation omission candidate recommendation based on the determination result.

FIG. 13 is a diagram showing an example of the determination data 32h. As shown in FIG. 13, the determination data 32h may optionally include attributes (eg, gender, age) as well as a plurality of medical care items.

For example, the determination unit 33d refers to the memory unit 32 and executes the calculation omission detection for the determination data 32h using the learned detection model 32g. For example, the determination unit 33d inputs the determination data 32h into the machine learning inference process constructed by using the learned detection model 32g, acquires the determination result, and makes a calculation omission candidate recommendation based on the acquired determination result. You may output it. For example, the determination unit 33d outputs a calculation omission candidate recommendation including at least one item of the target patient, the detection date, the calculation omission item, and the calculation omission probability, as in the calculation omission candidate recommendation 26 shown in FIG. good.

Outputs of calculation omission candidate recommendations include, for example, display on an output device such as a monitor, storage in a memory unit 32 or a storage device, and transmission to the outside of the detection system 3 (for example, an information processing device) via a communication unit 31. , Etc. can be mentioned.

As described above, the determination unit 33d is an example of the detection unit that detects the omission of calculation of the medical treatment fee of the determination data 32h regarding the plurality of receipts by using the learned detection model 32g.

As described above, according to the detection system 3 according to the embodiment, the number of cases in which the medical item that is not related to the calculation item of the detection target (small relationship) is determined to be "significantly different" by the significant difference determination process. Can be reduced. As a result, it is possible to reduce the decrease in the detection accuracy of the medical receipt calculation omission, and it is possible to prevent the medical fee calculation omission, so that the medical profit can be improved.

[1-3] Operation Example Hereinafter, the operation example of the detection system 3 described above will be described with reference to the flowchart.

[1-3-1] Learning process (model creation phase)
FIG. 14 is a flowchart illustrating an operation example of the learning process of the detection system 3 according to the embodiment.

As illustrated in FIG. 14, in the control unit 33 of the detection system 3, the acquisition unit 33a is the input data 32a based on the learning data input from the information processing device used by the operator or the like via the communication unit 31. Is acquired (step S1) and stored in the memory unit 32. When the acquisition unit 33a acquires the input data 32a, the acquisition unit 33a may notify the generation unit 33b of the generation instruction of the learning data 32f. The data for learning may include receipt information, electronic medical record information, and information on calculation items (correct answer data). The input data 32a may include labels, attributes (eg, gender, age), as well as a plurality of medical care items, as illustrated in FIG.

When the generation instruction is input from the acquisition unit 33a, the generation unit 33b generates the rule-applied data 32c by extracting the data matching the rule of the calculation item to be detected in the extraction rule 32b from the input data 32a. (Step S2), and the data is stored in the memory unit 32.

The generation unit 33b generates (extracts) the calculation item requested data 32d and the significant difference calculation data 32e from the rule-applied data 32c (step S3), and stores the data in the memory unit 32. For example, the generation unit 33b generates the calculation item requested data 32d by extracting only the data "corresponding to the calculation item" (requesting the calculation item) to be detected from the rule applied data 32c or the input data 32a. You can do it. Further, the generation unit 33b randomly extracts only the data that is narrowed down by the extraction rule 32b and that does not correspond to the calculation item (does not request the calculation item) from the rule-applied data 32c. Data for calculating the significant difference 32e may be generated.

The generation unit 33b executes the significance difference determination process using the calculation item requested data 32d and the significance difference calculation data 32e (step S4). For example, the generation unit 33b determines whether or not each medical item is related to the calculation item by determining whether or not the calculation item to be detected is related to each of the plurality of medical items in the significant difference determination process. In other words, it may be determined whether or not there is a significant difference.

The generation unit 33b generates training data 32f from the input data 32a, the rule applied data 32c, the calculation item requested data 32d, and the significant difference calculation data 32e based on the processing result of the significance difference determination processing (step S5). ), Stored in the memory unit 32. The learning data 32f includes only the entries corresponding to the extraction rule 32b of the detection target in the input data 32a, and the medical care items are limited to the medical care items determined to have a significant difference in the significant difference determination process. It becomes the data.

The learning unit 33c generates a detection model 32g by machine learning using the learning data 32f (step S6), stores it in the memory unit 32, and completes the processing of the model creation phase. In the model creation phase, the acquisition unit 33a, the generation unit 33b, and the learning unit 33c may execute the processes of steps S1 to S6 for each calculation item to be detected.

[1-3-2] Judgment processing (model application phase)
FIG. 15 is a flowchart illustrating an operation example of the determination process of the detection system 3 according to the embodiment.

As illustrated in FIG. 15, the acquisition unit 33a acquires the determination data 32h based on the detection target data input from the information processing device used by the operator or the like via the communication unit 31 (step S11), and the memory. It is stored in the unit 32. When the acquisition unit 33a acquires the determination data 32h, the acquisition unit 33a may notify the determination unit 33d of the determination instruction. The data to be detected may include information such as receipt information and electronic medical record information.

For example, when a determination instruction is input from the acquisition unit 33a, the determination unit 33d refers to the memory unit 32 and determines the determination data 32h to be determined using the learned detection model 32g (step S12). For example, the determination unit 33d may input the determination data 32h into the inference process of machine learning to which the detection model 32g is applied, and acquire the output from the machine learning as the determination result.

The determination unit 33d generates a calculation omission candidate recommendation based on the determination result (step S13), outputs the output (step S14), and ends the processing of the model application phase.

[1-4] Hardware Configuration Example The detection system 3 according to the embodiment may be realized by, for example, a virtual server (VM; Virtual Machine) or a computer (information processing device) such as a physical server. Further, the function of the detection system 3 may be realized by one computer or may be realized by two or more computers. Further, at least a part of the functions of the detection system 3 may be realized by using HW (Hardware) resources and NW (Network) resources provided by the cloud environment.

FIG. 16 is a block diagram showing a hardware (HW) configuration example of the computer 30 that realizes the function of the detection system 3. When a plurality of computers are used as the HW resource that realizes the function of the detection system 3, each computer may have the HW configuration illustrated in FIG.

As shown in FIG. 16, the computer 30 optionally includes a processor 30a, a memory 30b, a storage unit 30c, an IF (Interface) unit 30d, an IO (Input / Output) unit 30e, and a reading unit 30f as an HW configuration. You may be prepared.

The processor 30a is an example of an arithmetic processing unit that performs various controls and operations. The processor 30a may be connected to each block in the computer 30 so as to be able to communicate with each other by the bus 30i. The processor 30a may be a multi-processor including a plurality of processors, a multi-core processor having a plurality of processor cores, or a configuration having a plurality of multi-core processors.

Examples of the processor 30a include integrated circuits (ICs) such as CPUs, MPUs, GPUs, APUs, DSPs, ASICs, and FPGAs. As the processor 30a, two or more combinations of these integrated circuits may be used. CPU is an abbreviation for Central Processing Unit, and MPU is an abbreviation for Micro Processing Unit. GPU is an abbreviation for Graphics Processing Unit, and APU is an abbreviation for Accelerated Processing Unit. DSP is an abbreviation for Digital Signal Processor, ASIC is an abbreviation for Application Specific IC, and FPGA is an abbreviation for Field-Programmable Gate Array.

The memory 30b is an example of HW that stores information such as various data and programs. Examples of the memory 30b include one or both of a volatile memory such as DRAM (Dynamic Random Access Memory) and a non-volatile memory such as PM (Persistent Memory).

The storage unit 30c is an example of HW that stores information such as various data and programs. Examples of the storage unit 30c include a magnetic disk device such as an HDD (Hard Disk Drive), a semiconductor drive device such as an SSD (Solid State Drive), and various storage devices such as a non-volatile memory. Examples of the non-volatile memory include flash memory, SCM (Storage Class Memory), ROM (Read Only Memory) and the like.

Further, the storage unit 30c may store a program 30g (receipt data significance determination program) that realizes all or a part of various functions of the computer 30.

For example, the processor 30a of the detection system 3 expands the program 30g stored in the storage unit 30c into the memory 30b and executes it as the detection system 3 (for example, the communication unit 31 and the control unit 33) illustrated in FIG. Functions can be realized. Further, the memory unit 32 illustrated in FIG. 3 may be realized by a storage area possessed by at least one of the memory unit 30b and the storage unit 30c. Further, the determination unit 33d shown in FIG. 3 may output and store information such as a significant difference determination result and a calculation omission candidate recommendation in the storage area of at least one of the memory 30b and the storage unit 30c.

The IF unit 30d is an example of a communication IF that controls connection and communication with a network. For example, the IF unit 30d may include an adapter compliant with LAN (Local Area Network) such as Ethernet (registered trademark) or optical communication such as FC (Fibre Channel). The adapter may support one or both wireless and wired communication methods. For example, the detection system 3 may be connected to an information processing device (not shown) so as to be able to communicate with each other via the IF unit 30d. At least a part of the functions of the communication unit 31 illustrated in FIG. 3 may be realized by the IF unit 30d. Further, for example, the program 30g may be downloaded from the network to the computer 30 via the communication IF and stored in the storage unit 30c.

The IO unit 30e may include one or both of an input device and an output device. Examples of the input device include a keyboard, a mouse, a touch panel, and the like. Examples of the output device include a monitor, a projector, a printer and the like. For example, the determination unit 33d shown in FIG. 3 may output and display the determination result on the output device of the IO unit 30e.

The reading unit 30f is an example of a reader that reads data and program information recorded on the recording medium 30h. The reading unit 30f may include a connection terminal or device to which the recording medium 30h can be connected or inserted. Examples of the reading unit 30f include an adapter compliant with USB (Universal Serial Bus), a drive device for accessing a recording disk, a card reader for accessing a flash memory such as an SD card, and the like. The program 30g may be stored in the recording medium 30h, or the reading unit 30f may read the program 30g from the recording medium 30h and store it in the storage unit 30c.

Examples of the recording medium 30h include a non-temporary computer-readable recording medium such as a magnetic / optical disk or a flash memory. Examples of the magnetic / optical disk include flexible discs, CDs (Compact Discs), DVDs (Digital Versatile Discs), Blu-ray discs, HVDs (Holographic Versatile Discs), and the like. Examples of the flash memory include semiconductor memories such as USB memory and SD card.

The HW configuration of the computer 30 described above is an example. Therefore, the increase / decrease of HW (for example, addition or deletion of arbitrary blocks), division, integration in any combination, addition or deletion of buses, etc. may be appropriately performed in the computer 30. For example, in the detection system 3, at least one of the IO unit 30e and the reading unit 30f may be omitted.

[2] Others The technique according to the above-described embodiment can be modified or modified as follows.

For example, the communication unit 31 and the control unit 33 included in the detection system 3 shown in FIG. 3 may be merged or separated from each other. The acquisition unit 33a, the generation unit 33b, the learning unit 33c, and the determination unit 33d included in the control unit 33 shown in FIG. 3 may be merged or divided in any combination.

Further, the detection system 3 shown in FIG. 3 may be configured to realize each processing function by coordinating a plurality of devices with each other via a network. As an example, the communication unit 31 may be a Web server, the control unit 33 may be an application server, the memory unit 32 may be a DB server, and the like. In this case, the Web server, the application server, and the DB server may cooperate with each other via the network to realize each processing function as the detection system 3.

[3] Additional notes The following additional notes will be further disclosed with respect to the above embodiments.

(Appendix 1)
From the first data regarding multiple receipts, the second data is extracted according to the conditions for extracting the calculation items of the receipts.
Randomly extracted from the third data obtained by extracting the data for which the first calculation item is requested from the first data or the second data, and the data excluding the data for which the first calculation item is requested among the second data. Based on the fourth data, a determination process is performed to determine a significant difference in the first calculation item for the third data and the plurality of first medical care items included in the fourth data.
From the third data and the fourth data, learning data is generated by extracting one or more second medical care items determined to have a significant difference by the determination process.
A receipt data significance determination program that causes a computer to perform processing.

(Appendix 2)
In the determination process, based on the presence / absence of a request for the first calculation item and the presence / absence of a diagnosis for each of the plurality of medical treatment items, the first or more second medical treatment related to the request for the first calculation item. Judging the item,
The receipt data significance determination program according to Appendix 1, which causes the computer to execute the process.

(Appendix 3)
The first data includes correct answer data indicating whether or not the first calculation item is requested for each of the plurality of receipts, and the plurality of medical treatment items.
In the determination process, the first or more second medical item is determined based on the aggregation result of each combination of the presence / absence of the request for the first calculation item and the presence / absence of each diagnosis of the plurality of medical items.
The receipt data significance determination program according to Appendix 2, which causes the computer to execute the process.

(Appendix 4)
The first data includes attribute information regarding a medical treatment subject for each of the plurality of receipts.
The condition is a condition related to the attribute information.
The extraction of the second data includes extracting data having attribute information satisfying the above conditions from the first data.
The receipt data significance determination program according to any one of Supplementary note 1 to Supplementary note 3.

(Appendix 5)
Using the learning data, machine learning of a detection model for detecting the omission of calculation of the medical fee of the medical receipt is performed.
The receipt data significance determination program according to any one of Supplementary note 1 to Supplementary note 4, wherein the processing is executed by the computer.

(Appendix 6)
Using the learned detection model, the omission of calculation of the medical fee of the fifth data regarding multiple receipts is detected.
The receipt data significance determination program according to Appendix 5, which causes the computer to execute the process.

(Appendix 7)
From the first data regarding multiple receipts, the second data is extracted according to the conditions for extracting the calculation items of the receipts.
Randomly extracted from the third data obtained by extracting the data for which the first calculation item is requested from the first data or the second data, and the data excluding the data for which the first calculation item is requested among the second data. Based on the fourth data, a determination process is performed to determine a significant difference in the first calculation item for the third data and the plurality of first medical care items included in the fourth data.
From the third data and the fourth data, learning data is generated by extracting one or more second medical care items determined to have a significant difference by the determination process.
Receipt data significance determination method in which the processing is executed by the computer.

(Appendix 8)
In the determination process, based on the presence / absence of a request for the first calculation item and the presence / absence of a diagnosis for each of the plurality of medical treatment items, the first or more second medical treatment related to the request for the first calculation item. Judging the item,
The method for determining the significance of receipt data according to Appendix 7, wherein the processing is executed by the computer.

(Appendix 9)
The first data includes correct answer data indicating whether or not the first calculation item is requested for each of the plurality of receipts, and the plurality of medical treatment items.
In the determination process, the first or more second medical item is determined based on the aggregation result of each combination of the presence / absence of the request for the first calculation item and the presence / absence of each diagnosis of the plurality of medical items.
The method for determining the significance of receipt data according to Appendix 8, wherein the processing is executed by the computer.

(Appendix 10)
The first data includes attribute information regarding a medical treatment subject for each of the plurality of receipts.
The condition is a condition related to the attribute information.
The extraction of the second data includes extracting data having attribute information satisfying the above conditions from the first data.
The method for determining the significance of the receipt data according to any one of Supplementary note 7 to Supplementary note 9.

(Appendix 11)
Using the learning data, machine learning of a detection model for detecting the omission of calculation of the medical fee of the medical receipt is performed.
The method for determining the significance of receipt data according to any one of Supplementary note 7 to Supplementary note 10, wherein the processing is executed by the computer.

(Appendix 12)
Using the learned detection model, the omission of calculation of the medical fee of the fifth data regarding multiple receipts is detected.
The method for determining the significance of receipt data according to Appendix 11, wherein the processing is executed by the computer.

(Appendix 13)
An extraction unit that extracts the second data from the first data related to a plurality of receipts according to the conditions for extracting the calculation items of the receipts.
Randomly extracted from the third data obtained by extracting the data for which the first calculation item is requested from the first data or the second data, and the data excluding the data for which the first calculation item is requested among the second data. A significant difference determination unit that performs a determination process for determining a significant difference in the first calculation item for a plurality of first medical care items included in the third data and the fourth data based on the fourth data.
An information processing device including a learning data generation unit that generates learning data obtained by extracting one or more second medical care items determined to have a significant difference by the determination process from the third data and the fourth data.

(Appendix 14)
The significance difference determination unit is based on the presence or absence of a request for the first calculation item and the presence or absence of a diagnosis for each of the plurality of medical treatment items, and the first or higher number related to the request for the first calculation item. 2 Judge medical items,
The information processing apparatus according to Appendix 13.

(Appendix 15)
The first data includes correct answer data indicating whether or not the first calculation item is requested for each of the plurality of receipts, and the plurality of medical treatment items.
The significant difference determination unit determines one or more of the second medical care items based on the aggregation result of each combination of the presence or absence of the request for the first calculation item and the presence or absence of each diagnosis of the plurality of medical care items. do,
The information processing apparatus according to Appendix 14.

(Appendix 16)
The first data includes attribute information regarding a medical treatment subject for each of the plurality of receipts.
The condition is a condition related to the attribute information.
The extraction unit extracts data having attribute information satisfying the above conditions from the first data.
The information processing apparatus according to any one of Supplementary note 13 to Supplementary note 15.

(Appendix 17)
The information processing apparatus according to any one of Supplementary note 13 to Supplementary note 16, comprising a model learning unit that performs machine learning of a detection model for detecting a medical fee calculation omission of the medical receipt using the learning data.

(Appendix 18)
The information processing apparatus according to Appendix 17, comprising a detection unit for detecting a medical fee calculation omission of the fifth data related to a plurality of receipts using the learned detection model.

1 System 1a Medical management system 10, 2, 3 Receipt calculation omission detection system 21 Calculation results 21a Receipt 21b Electronic chart 21c Calculation items 22 Machine learning department 23, 32g Detection model 24 New pre-correction receipt 25 New electronic chart 26 Calculation omission candidate recommendation 30 Computer 31 Communication unit 32 Memory unit 32a Input data 32b Extraction rule 32c Rule applied data 32d Calculation item Requested data 32e Significant difference calculation data 32f Learning data 32h Judgment data 33 Control unit 33a Acquisition unit 33b Generation unit 33c Learning unit 33d Judgment unit

Claims (8)

From the first data regarding multiple receipts, the second data is extracted according to the conditions for extracting the calculation items of the receipts.
Randomly extracted from the third data obtained by extracting the data for which the first calculation item is requested from the first data or the second data, and the data excluding the data for which the first calculation item is requested among the second data. Based on the fourth data, a determination process is performed to determine a significant difference in the first calculation item for the third data and the plurality of first medical care items included in the fourth data.
From the third data and the fourth data, learning data is generated by extracting one or more second medical care items determined to have a significant difference by the determination process.
A receipt data significance determination program that causes a computer to perform processing. In the determination process, based on the presence / absence of a request for the first calculation item and the presence / absence of a diagnosis for each of the plurality of medical treatment items, the first or more second medical treatment related to the request for the first calculation item. Judging the item,
The receipt data significance determination program according to claim 1, wherein the processing is executed by the computer. The first data includes correct answer data indicating whether or not the first calculation item is requested for each of the plurality of receipts, and the plurality of medical treatment items.
In the determination process, the first or more second medical item is determined based on the aggregation result of each combination of the presence / absence of the request for the first calculation item and the presence / absence of each diagnosis of the plurality of medical items.
The receipt data significance determination program according to claim 2, wherein the processing is executed by the computer. The first data includes attribute information regarding a medical treatment subject for each of the plurality of receipts.
The condition is a condition related to the attribute information.
The extraction of the second data includes extracting data having attribute information satisfying the above conditions from the first data.
The receipt data significance determination program according to any one of claims 1 to 3. Using the learning data, machine learning of a detection model for detecting the omission of calculation of the medical fee of the medical receipt is performed.
The receipt data significance determination program according to any one of claims 1 to 4, wherein the processing is executed by the computer. Using the learned detection model, the omission of calculation of the medical fee of the fifth data regarding multiple receipts is detected.
The receipt data significance determination program according to claim 5, wherein the processing is executed by the computer. From the first data regarding multiple receipts, the second data is extracted according to the conditions for extracting the calculation items of the receipts.
Randomly extracted from the third data obtained by extracting the data for which the first calculation item is requested from the first data or the second data, and the data excluding the data for which the first calculation item is requested among the second data. Based on the fourth data, a determination process is performed to determine a significant difference in the first calculation item for the third data and the plurality of first medical care items included in the fourth data.
From the third data and the fourth data, learning data is generated by extracting one or more second medical care items determined to have a significant difference by the determination process.
Receipt data significance determination method in which the processing is executed by the computer. An extraction unit that extracts the second data from the first data related to a plurality of receipts according to the conditions for extracting the calculation items of the receipts.
Randomly extracted from the third data obtained by extracting the data for which the first calculation item is requested from the first data or the second data, and the data excluding the data for which the first calculation item is requested among the second data. A significant difference determination unit that performs a determination process for determining a significant difference in the first calculation item for a plurality of first medical care items included in the third data and the fourth data based on the fourth data.
An information processing device including a learning data generation unit that generates learning data obtained by extracting one or more second medical care items determined to have a significant difference by the determination process from the third data and the fourth data.

Images