JP5799377B2 - Abnormal frequency estimation device, abnormal frequency estimation program, abnormal frequency estimation method, and abnormal frequency estimation system - Google Patents

Description

  The present invention relates to an abnormal frequency estimation device, an abnormal frequency estimation program, an abnormal frequency estimation method, and an abnormal frequency estimation system, and particularly, for example, an abnormal frequency for estimating the frequency (abnormal frequency) at which a test result of a test item in a clinical test shows an abnormality. The present invention relates to an estimation device, an abnormal frequency estimation program, an abnormal frequency estimation method, and an abnormal frequency estimation system.

  An example of this type of conventional abnormality frequency estimation device is disclosed in Patent Document 1. In the clinical test data analysis and display device disclosed in Patent Document 1, the health level is calculated from the test data of the client based on the test data stored in advance and the reference value consisting of the upper and lower limits of the range to be determined as normal. Using an interactive interface to judge and display, a hidden relationship between test items of clinical test data is extracted. In this relation extraction, attention is paid to the number of abnormalities in each inspection item. Therefore, a relation matrix calculation unit is added to the feature extraction unit of the interactive interface, and a relation matrix display interface is added to the display interface between the display control device and the display device. Thereby, another inspection item value is estimated from a certain inspection item value.

JP 2011-22857 [G06Q 50/00, G06F 17/30, A61B 5/00]

  However, in the clinical test data analysis display device of Patent Document 1, for example, the relationship between the average value of test values of test items {A1, A2, A3, ...} and the number of abnormal test results of test items B is linear. Assuming that the number of abnormalities of the inspection result of the inspection item B is linearly approximated (estimated) from the average value of the inspection values of the inspection items of the A group, (Abnormal frequency) cannot be estimated correctly.

  Therefore, a main object of the present invention is to provide a novel abnormality frequency estimation device, abnormality frequency estimation program, abnormality frequency estimation method, and abnormality frequency estimation system.

  Another object of the present invention is to provide an abnormality frequency estimation device, an abnormality frequency estimation program, an abnormality frequency estimation method, and an abnormality frequency estimation system that can improve estimation accuracy.

  The present invention employs the following configuration in order to solve the above problems. The reference numerals in parentheses, supplementary explanations, and the like indicate correspondence relationships with embodiments described later to help understanding of the present invention, and do not limit the present invention in any way.

  The first invention is a database in which test values of a predetermined number of test items for clinical tests are recorded for each of a plurality of clients, and of all test values of all test items recorded in the database, First average value calculating means for calculating an average value for a predetermined number of tests for each of some of the plurality of first test items using the test values, and using the test values of the previous and previous clinical tests, Abnormal frequency calculating means for calculating an abnormal frequency at a predetermined number of inspections for the second inspection item other than the inspection items, an average value vector having each average value calculated by the first average value calculating means as an element, and an abnormal frequency calculating means Extraction means for extracting the relationship between the abnormality frequency of the second inspection item calculated by the above, and for each of the plurality of first inspection items recorded in the database for the selected one client The second average value calculating unit calculates the average value of the test values for the predetermined number of times including the test value of the current clinical test, and the second average value calculating unit uses the relationship extracted by the extracting unit. An abnormality frequency estimation device comprising abnormality frequency estimation means for estimating an abnormality frequency in a predetermined number of examinations including the current clinical examination for the second examination item from the calculated average value for each of the plurality of first examination items. is there.

  In the first invention, the abnormality frequency estimation device (12) includes a database (14). The database records test values of a predetermined number of test items for clinical tests for each of a plurality of clients. The first average value calculation means (12a, S15) uses a test value of a clinical test before the previous time among all test values of all test items recorded in the database, and selects a plurality of first test items. The average value for the predetermined number of inspections for each is calculated. The abnormality frequency calculation means (12a, S15) calculates the abnormality frequency in the predetermined number of examinations for the second examination item other than the first examination item, using the examination value of the previous clinical examination. The extracting means (12a, S21) shows the relationship between the average value vector having each average value calculated by the first average value calculating means as an element and the abnormality frequency of the second inspection item calculated by the abnormality frequency calculating means. Extract. The second average value calculation means (12a, S19) performs a predetermined number of examinations including the examination values of the current clinical examination for each of the plurality of first examination items recorded in the database for the selected one client. Calculate the average value. The abnormality frequency estimation means (12a, S27) uses the relationship extracted by the extraction means to calculate the second inspection item from the average value for each of the plurality of first inspection items calculated by the second average value calculation means. The frequency of abnormalities in a predetermined number of tests including the current clinical test is estimated.

  According to the first invention, out of all past inspection values, the average value for the predetermined number of inspections for each of the inspection values of some of the plurality of first inspection items and the second inspection other than the first inspection item The relationship between the abnormal frequency at the predetermined number of times of the item is extracted, and using this relationship, the predetermined value including the current time is obtained from the average value for the predetermined number of times for each of the inspection values of the plurality of first inspection items including the current inspection value. Since the abnormal frequency of the second inspection item in the number of times is estimated, it is not necessary to assume the relationship between the second inspection item to be estimated and the first inspection item, and the estimation accuracy can be improved.

  The second invention is dependent on the first invention, and the abnormality frequency is the number of times determined to be abnormal in a predetermined number of clinical tests.

  In the second invention, the abnormal frequency is the number of times that each test item is determined to be abnormal in, for example, a predetermined number of consecutive clinical tests.

  According to the second invention, since the abnormal frequency is estimated based on the test values of a plurality of clinical tests, it is not affected by various factors such as seasonal variation, sleep time, time from meal, etc. as much as possible. Abnormal frequency can be estimated relatively accurately.

  The third invention is dependent on the first or second invention, and based on the abnormality frequency of the second inspection item estimated by the abnormality frequency estimation means and the inspection value of the second inspection item before the previous time, The apparatus further comprises test result estimating means for estimating normality or abnormality of the test result of the second test item.

  In the third invention, the abnormality frequency estimation device further includes test result estimation means (12a, S29). The inspection result estimation means is configured to determine the current inspection result of the second inspection item based on the abnormality frequency of the second inspection item estimated by the abnormality frequency estimation means and the inspection value of the second inspection item before the previous time. Estimate (determine) normality or abnormality.

  According to the third invention, when the test result for the second test item is estimated to be abnormal, a clinical test can be performed for the second test item. The risk of overlooking can be reduced as much as possible.

  A fourth invention is an abnormality frequency estimation program for an abnormality frequency estimation device, comprising a database in which test values of a predetermined number of test items for clinical tests are recorded for each of a plurality of clients, the processor of the abnormality frequency estimation device In addition, among all the test values of all test items recorded in the database, using the test values of the previous clinical test, the average value for the predetermined number of times for each of some of the first test items is calculated. A first average value calculating step for calculating, an abnormal frequency calculating step for calculating an abnormal frequency in a predetermined number of tests for a second test item other than the first test item, using a test value of a previous clinical test, a first average The relationship between the average value vector having each average value calculated in the value calculation step as an element and the abnormality frequency of the second inspection item calculated in the abnormality frequency calculation step Extraction step of extracting the average value of the test values for the predetermined number of tests including the test values of the current clinical test for each of the plurality of first test items recorded in the database for the selected one client Using the relationship extracted in the second average value calculating step and the extracting step, the current value for the second inspection item is calculated from the average value for each of the plurality of first inspection items calculated in the second average value calculating step. It is an abnormal frequency estimation program which performs the abnormal frequency estimation step which estimates the abnormal frequency in the predetermined frequency | count of an inspection including the clinical test of this.

  5th invention is the abnormality frequency estimation method of the abnormality frequency estimation apparatus provided with the database which recorded the test value of the predetermined number of test items about a clinical test about each of several clients, Comprising: The processor of abnormality frequency estimation apparatus (A) Of all the test values recorded in the database, among the test values of the previous test, the test values for the predetermined number of tests for each of some of the first test items are used. An average value is calculated, (b) using the test values of clinical tests before the previous time, the abnormal frequency in the predetermined number of tests for the second test item other than the first test item is calculated, and (c) step (a) The relationship between the average value vector having each average value calculated in step 1 and the abnormal frequency of the second inspection item calculated in step (b) is extracted, and (d) about the selected one client. The average value of the test values for the predetermined number of times including the test value of the current clinical test for each of the plurality of first test items recorded in the database is calculated, and (e) extracted in step (c) Using the relationship, an abnormal frequency in a predetermined number of tests including the current clinical test for the second test item is estimated from the average value for each of the plurality of first test items calculated in step (d). This is an abnormal frequency estimation method.

  The sixth invention relates to a database in which test values of a predetermined number of test items for clinical tests are recorded for each of a plurality of clients, and of all test values of all test items recorded in the database, First average value calculating means for calculating an average value for a predetermined number of tests for each of some of the plurality of first test items using the test values, and using the test values of the previous and previous clinical tests, Abnormal frequency calculating means for calculating an abnormal frequency at a predetermined number of inspections for the second inspection item other than the inspection items, an average value vector having each average value calculated by the first average value calculating means as an element, and an abnormal frequency calculating means Extraction means for extracting the relationship between the abnormality frequency of the second inspection item calculated by the above, and for each of the plurality of first inspection items recorded in the database for the selected one client The second average value calculating unit calculates the average value of the test values for the predetermined number of times including the test value of the current clinical test, and the second average value calculating unit uses the relationship extracted by the extracting unit. An abnormality frequency estimation system comprising abnormality frequency estimation means for estimating an abnormality frequency at a predetermined number of examinations including a current clinical examination for a second examination item from an average value for each of a plurality of calculated first examination items. is there.

  In the fourth to sixth inventions as well, the estimation accuracy can be improved as in the first invention.

  According to the present invention, among all the past inspection values, the average value for the predetermined number of inspections for each of the inspection values of some of the plurality of first inspection items and the second inspection items other than the first inspection item. The relationship with the abnormal frequency at the predetermined number of times is extracted, and using this relationship, the average value for the predetermined number of times for each of the inspection values of the plurality of first inspection items including the current inspection value at the predetermined number of times including this time Since the abnormality frequency of the second inspection item is estimated, it is not necessary to assume the relationship between the second inspection item to be estimated and the first inspection item, and the estimation accuracy can be improved.

  The above object, other objects, features and advantages of the present invention will become more apparent from the following detailed description of embodiments with reference to the drawings.

FIG. 1 is an illustrative view showing one example of an abnormality frequency estimation system of the present invention. FIG. 2 is an illustrative view showing one example of contents of the inspection database shown in FIG. FIG. 3 is an illustrative view showing specific contents of the inspection database shown in FIGS. 1 and 2. FIG. 4 is an illustrative view showing a normal distribution of inspection values and a reference value. FIG. 5 is an illustrative view showing one example of a reference value. FIG. 6 is an illustrative view showing an example of a change in the inspection value and its abnormality frequency. FIG. 7 is an illustrative view showing one example of a conversion table for converting a test result that is not a real value into a real value. FIG. 8 is an illustrative view for explaining how to obtain a likelihood function. FIG. 9 is an illustrative view showing a correspondence table for estimating normality and abnormality of the inspection result of the estimated inspection item from the previous two inspection results and the abnormality frequency estimation result. FIG. 10 is an illustrative view showing one example of a memory map of a RAM in the abnormality frequency estimation apparatus of FIG. FIG. 11 is a flowchart showing an abnormal frequency estimation process and a test result estimation process in the abnormal frequency estimation apparatus shown in FIG. FIG. 12 is an illustrative view for explaining the flow of actual clinical examination and abnormal frequency estimation. FIG. 13 is an illustrative view showing costs of main item categories. FIG. 14 is a graph of the results showing the number of items tested, the number of correct answers, and the number of incorrect answers for each gender when the inspection items are always constant. FIG. 15 is a graph showing a result of totaling the results of the male shown in FIG. FIG. 16 is a graph showing the results of totaling the results for the woman shown in FIG. 14 for each cumulative number of examinations. FIG. 17 is a graph of the results showing the number of items tested, the number of correct answers, and the number of incorrect answers for each gender when the test items are changed. FIG. 18 is a graph showing the results of counting the results of the male shown in FIG. 17 for each cumulative number of examinations. FIG. 19 is a graph showing the results of counting the results of the woman shown in FIG. 17 for each cumulative number of examinations. FIG. 20 is a graph for comparing the average number of incorrect answer items depending on whether or not an inspection item is selected. FIG. 21 is a graph showing changes in the inspection cost when an inspection item is selected in the case of a male. FIG. 22 is a graph showing the transition of the inspection cost when an inspection item is selected in the case of a woman. FIG. 23 is a graph for comparing the accuracy of estimation of abnormality frequency before and after filtering.

  Referring to FIG. 1, an abnormality frequency estimation system 10 of this embodiment includes an abnormality frequency estimation device 12 and an examination database 14. The abnormality frequency estimation device 12 and the inspection database 14 are connected so that commands and data can be transmitted and received. The abnormality frequency estimation device 12 is a general-purpose computer and includes components such as a CPU 12a and a RAM 12b.

  In this embodiment, the inspection database 14 is directly connected to the abnormality frequency estimation device 12, but may be connected via a network such as the Internet or a LAN. Or you may comprise the test | inspection database 14 with the hard disk drive (not shown) incorporated in the abnormality frequency estimation apparatus 12. FIG.

  As shown in FIG. 2, client information 16, inspection item information 18, and inspection result information 20 are stored in the inspection database 14. Although detailed description is omitted, in this embodiment, the examination database 14 stores information on 579 clients (male: 260, female: 319).

  As shown in FIG. 3A, the client information 16 includes information (data) about each client (user who receives a health check). Data about the client (client data) is data about the name, user ID, gender, and date of birth. However, the user ID is identification information of the client (user), and is, for example, an individual number or symbol (numerals or Roman letters or both) assigned to the client.

  As shown in FIG. 3B, the inspection item information 18 includes information (data) about each inspection item. The inspection item data is data on the item classification name, item name (inspection ID), reference value upper limit (male), reference value lower limit (male), reference value upper limit (female), and reference value lower limit (female).

  The item classification name is a name for each group (group) into which each inspection item is classified. In this example, “serum protein”, “protein fraction”, “hepatobiliary system”, “autoimmunity”, “iron metabolism”, “inorganic ion metabolism”, “lipid metabolism”, “lipoprotein”, “sugar” Metabolism, muscle muscle, kidney function, inflammation, enzyme, infection, blood in general, cellular immunity, tumor marker (serum), tumor marker (urine) "And" general urine ".

  The item name (inspection ID) is the name of the inspection item. For example, the inspection items are 134 items in total, and total protein (g / dl), albumin (g / dl), albumin fraction (%), etc. are described as item names. However, instead of the item name, identification information assigned by a number or symbol such as a user ID may be described.

For details on item classification and item names, please refer to the specifications for the Integrated Medical Infrastructure (CIMI) database developed by the Tokyo Women's Medical University IREIIMS (http: //www.twmu. ac.jp/TIIMS
/CIMI_Database_v1_0.pdf), the description is omitted here.

  The reference value upper limit (male) is the upper limit value of the reference value for men set in the inspection item. The reference value lower limit (male) is the lower limit value of the reference value for men set in the inspection item. The reference value upper limit (female) is an upper limit value of the reference value for the woman set in the inspection item. The reference value lower limit (female) is a lower limit value of the reference value for the woman set in the inspection item.

  Thus, a reference value is set for each test item in order to determine whether the result of the health check is normal or abnormal. As shown in FIG. 4, the reference value means a value that defines a range included in 95% of the normal distribution of inspection values. However, the normal distribution is a normal distribution of test values for healthy adults. The maximum value (upper limit) of the range included in 95% is the reference value upper limit, and the minimum value (lower limit) is the reference value lower limit. However, since the reference value upper limit and the reference value lower limit differ for some examination items between men and women, each case is stored (set).

FIG. 5 shows an example of reference values (reference value upper limit and reference value lower limit). Specifically, reference values for men and women corresponding to the test items of white blood cell count (x × 10 ³ ), red blood cell count (x × 10 ⁶ ), and creatinine (serum) (mg / dl) An upper limit and a reference value lower limit are respectively described. Although illustration is omitted, the same applies to other inspection items.

  Returning to FIG. 3C, the inspection result information 20 includes information (data) about each inspection result. The inspection result data is data on the user ID, item name (inspection ID), inspection date, inspection value, reference value upper limit, reference value lower limit, and determination result. The user ID and item name are as described above.

  The examination date is the date on which the clinical examination for the examination item indicated by the item name was performed. The inspection value is a value (real value, discrete value) of an inspection result (measurement result) for the inspection item indicated by the item name. The reference value upper limit is an upper limit of the reference value when a clinical test is performed on the test item indicated by the item name, and may vary depending on the gender of the client as described above. Similarly, the reference value lower limit is a reference value lower limit when a clinical test is performed on the test item indicated by the item name, and may be different depending on the sex of the client. The determination result is a normal or abnormal estimation (determination) result of a test result (test value) of a clinical test for the test item indicated by the item name.

  As described above, the reference value upper limit and the reference value lower limit of the inspection date (during inspection) are stored because the reference value may change depending on the inspection time. The reason why the reference value changes includes a technical variation factor and a physiological variation factor. As a technical variation factor, a change in a reference value or a measurement error due to a change in an instrument or a reagent used for measurement corresponds. In addition, the physiological variation factor includes a variation in a reference value or a seasonal variation depending on the age of data to be analyzed.

  In the abnormality frequency estimation apparatus 12 of this embodiment, the frequency (abnormal frequency) at which an abnormality occurs with respect to the test result of the test item that is not subjected to the clinical test is estimated with reference to the test database 14. In this embodiment, the abnormal frequency means the number of times the test result (test value) shows an abnormality among the test results of a plurality of consecutive (for example, three) health examinations (clinical tests). However, the abnormality frequency is estimated for each inspection item.

  As described above, there are a total of 134 examination items for clinical examination, and it is very expensive to examine all of these items every time. On the other hand, in general health examinations, there are about 10 to 13 examination items, and there are too few examination items to find signs of various diseases only from their abnormal states.

  Therefore, in this embodiment, while keeping the cost as low as possible, paying attention to the many-to-one relationship, the abnormal frequency for the test items that are not clinically tested is estimated, and further, based on the estimation result, It is determined (estimated) whether the test result for the test item for which the clinical test is not performed is normal or abnormal.

  Briefly, when estimating an abnormal frequency for a certain test item, test values for a predetermined number of times including the current time of a plurality of test items that have actually undergone the clinical test are used. However, the inspection value is not directly used, but is converted into the following value. Inspection values indicated by integer values and symbols are converted to real values. In addition, the inspection value corresponding to the different reference value in each period (each inspection time) is linearly converted. Furthermore, the test values (the test values after the above conversion) for the test results for the past predetermined number of times for the plurality of test items actually subjected to the clinical test this time, which are the basis of the abnormal frequency estimation. Similarly, the frequency (abnormal frequency) at which the test result (test value) for the past predetermined number of times for the plurality of test items that are not subjected to the clinical test this time is abnormal is calculated. A likelihood function is obtained from the average value thus calculated and the abnormality frequency. Then, the likelihood of the average value of the test values for the predetermined number of times including the current time described above is obtained, and the abnormality frequency is estimated for each of the plurality of test items that are not subjected to the clinical test. In this embodiment, a method based on Bayesian estimation is used to estimate the abnormality frequency, and thereby the abnormality frequency is identified.

  That is, in this embodiment, among all inspection values of all inspection items for all clients stored in the inspection database 14, a predetermined number of times before the previous time (in this embodiment, the past three times before the previous time) is used. Thus, an average value is calculated for each of the test items that have been clinically tested this time (referred to as “first test items” for convenience of explanation). Similarly, using the predetermined number of times before the previous time (in this example, the previous three times before the previous time), this time the clinical test is not performed (the abnormal frequency is estimated). For each of the “inspection items”), the abnormality frequency of the inspection result at the predetermined number of times is calculated. The relationship between the average value of each of the first inspection items calculated in this way and the abnormality frequency of each of the second inspection items is extracted. In this embodiment, a predetermined statistical model is created. Then, when the current clinical test is performed, the average value of the test values of three consecutive clinical tests including the current clinical test is calculated for each test item (first test item) for which the current clinical test was performed. This average value is applied to the created statistical model. Thereby, the abnormality frequency is estimated for each of the test items (second test items) that are not subjected to the clinical test this time.

  Here, an abnormality frequency calculation method and an abnormality frequency estimation method will be described. The abnormality frequency is calculated (estimated) based on the inspection value and the reference value recorded in the inspection database 14.

  As shown in FIG. 4, when a reference value is set corresponding to each inspection item and exceeds the upper limit (reference value upper limit) or lower than the lower limit (reference lower limit), the inspection concerned It is determined that the item inspection result is abnormal. However, in this embodiment, an inspection value that exceeds the upper limit of the reference value may be referred to as an “abnormally high value”, and an inspection value that is lower than the lower limit of the reference value may be referred to as an “abnormally low value”. In addition, as described above, since the reference value may differ between men and women, in this embodiment, it is divided into men, women, abnormally high values and abnormally low values, whether normal or abnormal. To be judged.

  In this embodiment, the test value in one clinical test is not used for estimation of the abnormal frequency, but the test value in consecutive clinical tests (three times) is used for estimation of the abnormal frequency. This is because, in general, the test value is known to be influenced by various factors such as seasonal variation, sleep time, time from meals, as well as variation related to the disease. For this reason, it is difficult to estimate other inspection values from one inspection value. This is the same for doctors, and is often determined based on test values of multiple clinical tests.

  Thus, as described above, in this embodiment, the average value of the past predetermined number of test values for each test item that has actually undergone the clinical test and the past predetermined number of times for each test item that does not perform the clinical test. Based on the abnormal frequency in the test results (test values), the abnormal frequency for each test item not undergoing the current clinical test is estimated from the average of multiple test values including the current clinical test. .

  FIG. 6A is a graph showing an inspection value for a certain inspection item in time series, and shows a result of determining whether the inspection value is abnormal based on the reference value. In FIG. 6A, when the inspection value exceeds the reference value, it is determined that the inspection result (inspection value) is abnormal, and when the inspection value is within the reference value, the inspection value is normal. It is determined that In FIG. 6A, the inspection values determined to be normal are indicated by white circles, and the inspection values determined to be abnormal are indicated by black circles. However, in FIG. 6A (the same applies to FIG. 6B), an abnormally high value or an abnormally low value is not distinguished.

  FIG. 6B is a table showing the abnormal frequency of the test value when the test value changes as shown in FIG. Specifically, FIG. 6 (B) corresponds to FIG. 6 (A), in which the judgment result of normality or abnormality of the test value is described corresponding to each number of clinical tests τ, and The numerical values (0, 1, 2, 3) of the abnormality frequency for the last three consecutive times ((τ−2) to τ times) are described.

  As shown in FIG. 6 (A) and FIG. 6 (B), since each inspection value from the first time (τ = 1) to the third time (τ = 3) exceeds the reference value, all are judged to be abnormal. Is done. That is, since three abnormalities have occurred in three clinical examinations, the abnormality frequency corresponding to the third time is “3”. Similarly, from the second (τ = 2) to the fourth (τ = 4), since the test values of the second test out of the three clinical tests are determined to be abnormal, the abnormality corresponding to the fourth test The frequency is “2”. Similarly, the abnormality frequency is calculated after the fifth time. In this embodiment, a case where the abnormality frequency is “0” is defined as “normal”, and a case where the abnormality frequency is “3” is defined as “high frequency abnormality”.

  In addition, in the first time and the second time, since the clinical examination is less than three times, the abnormality frequency is not calculated.

  Next, an abnormality frequency estimation method will be described. As described above, in order to estimate the abnormal frequency of test items that are not subjected to clinical tests, the average value of test values in the past three clinical tests and the values that do not perform clinical tests for each test item that has undergone clinical tests. The abnormal frequency of the past three inspection results (inspection values) of the inspection item is used. Hereinafter, the method for estimating the abnormality frequency will be described using equations. However, the variables used in each formula are defined as follows.

The variable i indicates the number of examination items for clinical examination, and the total number of examination items is indicated by N. The variable j indicates the type of client, and the total number of clients is indicated by H. Furthermore, the variable t indicates the number of clinical tests, and the cumulative number of tests of the client j is indicated by T _j .

X _j (t) obtained by vectorizing the inspection values x _{j, i} (t) of the client j and the inspection item i in accordance with the inspection item i is defined as Equation 1. However, X _j (t) represents all the test values obtained by the first (t-th) clinical test.

[Equation 1]
X _j (t) = {x _{j, i} (t) | i = 1, 2,..., N; t = 1, 2,..., T _j ; j = 1, 2,.
The boundary value between the inspection value (normal value) of the inspection item i determined to be normal and the inspection value (abnormal value) of the inspection item i determined to be abnormal, that is, the reference value is expressed as F _{i, low} (reference value lower limit ) And F _{i, high} (reference value upper limit), an abnormal vector B _j indicating whether each element of the inspection value vector X _j (t) is abnormal (abnormal value) or normal (normal value) (T) is defined as Equation 2.

[Equation 2]
B _j (t) = {b _{j, i} (t) | i = 1, 2,..., N; t = 1, 2,..., T _j ; j = 1, 2,.
Here, b _{j, i} (t) is “1” when the test value is abnormal (abnormal value), as shown in Equation 3, and when the test value is normal (normal value). “0”. However, when the inspection value is an abnormally high value exceeding _{Fi, high} (reference value upper limit) and when the inspection value is an abnormally low value falling below _{Fi, low} (reference value lower limit), b _{j, i} ( t) is “1”.

[Equation 3]

[Equation 4]

  Here, the processing (pre-processing) of the inspection value necessary before estimating the abnormality frequency will be described. As described above, test items for clinical tests include values whose test values are real numbers (real values) and values that are not real numbers (discrete values). For example, in urinalysis, (+) and (-) are used to represent the degree of negative and positive intensity. Specifically, (+) represents positive, and the greater the numerical value added to this, the stronger the positive tendency. In addition, (−) represents negative, and as with positive, the greater the numerical value added to this, the stronger the negative tendency. Furthermore, (+ −) represents a false positive, indicating that there is a suspicion of positive.

  In this way, values that are not real numbers are converted into numerical values by defining a table (conversion table) for conversion to real values and converting the values according to the conversion table, as shown in FIG. For example, when the test item is general urine (partial), if the test result is (3-), it is converted to a real value “0”, and if the test result is (+), the real value “ 4 ". Although the detailed description is omitted, the same applies to other inspection results.

  When the examination item is urine sediment, the examination result is expressed in the range of the number of sediments. Therefore, as shown in the conversion table of FIG. 7, a real value is defined corresponding to the range of the number of sediments, and by converting according to this conversion table, the range of the number of sediments is digitized. For example, when the number of sediments is 1 to 4 / HPF, it is converted to a real value “1”, and when the number of sediments is 30 to 39 / HPF, it is converted to a real value “5”. Although detailed description is omitted, the same applies to other test results (the number of sediments).

  In this way, a value that is not a real number is converted into a real value, and the inspection value is converted into a value that conforms to a reference value for each period (inspection time). That is, the past reference value is matched with the latest reference value. This is because each inspection item has a reference value upper limit and a reference value lower limit, but as described above, they may change depending on the inspection time.

  In this embodiment, it is necessary to consider the fluctuation of the reference value in order to focus on the relationship between the average value and the abnormality frequency by going back to the test values of the past clinical tests. For this reason, the test value is linearly converted before the abnormality frequency is estimated.

Specifically, in the case of a linear conversion the test values _{x i} of the test item i is the reference value _{F i} when the measured test value _{x i, low} and _{F i,} and _high, the latest reference value F Assuming that ′ _{i, low} and F ′ _{i, high} , the inspection value x ′ _i after the linear transformation is calculated according to Equation 5.

[Equation 5]

  In Equation 5, the ratio between the previous reference value and the current reference value is calculated, and the movement amount (difference) from the reference value lower limit is calculated.

  In the estimation of the abnormality frequency, a method based on Bayesian estimation is used. As described above, the average value and the abnormality frequency calculated from past inspection values (inspection results) are used as input values. The output of the estimation result is the abnormality frequency (“0 times abnormal / 3 times” to “3 times abnormal / 3 times”) of the inspection item i of the client j.

  Specifically, a method based on Bayesian estimation using a multidimensional likelihood function based on the abnormality frequency is introduced. The reason for introducing Bayesian estimation is that the prior distribution can be determined based on the prior knowledge of doctors and experts, and when identifying the frequency of abnormalities, it is more It is possible to estimate correctly (estimation accuracy is high).

  The other identification methods are kernel SVM, k-MM method, and J48 (decision tree). Although illustration and detailed explanation are omitted, estimation by Bayesian estimation has the highest accuracy rate according to experiments conducted in advance. It was.

  As described later, the prior distribution represents what is known about a certain variable as a probability distribution under a condition where there is no evidence. In this embodiment, the prior distribution is a probability distribution in which each class (0 to 3) of the abnormality frequency is estimated. In this embodiment, the prior distribution (probability distribution) is calculated from the test value, but it is also possible to adjust the probability distribution based on the doctor's knowledge. This is because, for example, according to the knowledge of doctors, there are individual circumstances, such as “if the search item i is to be estimated frequently to avoid risk”.

Here, when the abnormal frequency of a test item (test item to be estimated) that does not perform a clinical test was estimated from the test items that were actually tested in a general health examination, the clinical test was actually performed A set S of inspection items (inspection items to be inspected) is defined. For example, when the abnormal frequency of the test item “β microglobulin (urine)” is estimated from the test values of the three test items “blood glucose”, “β microglobulin (serum)”, and “ALP”, i = β microglobulin (urine), S = {blood glucose, β microglobulin (serum), ALP}.

[Equation 6]

[Equation 7]

  Although detailed description is omitted, when obtaining the abnormality frequency for each inspection item i that is not the inspection target, the inspection result (inspection value) of each inspection item i is determined to be normal or abnormal based on the reference value. At this time, “normal” or “abnormal” is stored (described) in the determination result of the data of the inspection result. Alternatively, when the inspection value is registered in the inspection database 14, normality or abnormality may be determined based on the reference value.

  Next, Bayesian estimation will be described. Bayesian estimation means a method of estimating the probability that an event that causes the fact will occur based on the observed fact. The basic formula for this Bayesian estimation is expressed by Equation 8.

[Equation 8]

  That is, based on the product of the probability P (X) of occurrence of event X and the likelihood (likelihood function) P (Y | X), the conditional probability of occurrence of event X under the occurrence of event Y (post facto) Probability) P (X | Y) is estimated.

As one of the characteristics of Bayesian estimation, the prior probability P (X) is determined as a constant in the conventional estimation statistics, but in Bayesian estimation, it is considered to be a function. This feature allows Bayesian estimation to be used in identifying spam emails where it is difficult to identify a population, and for producing “negative” and “positive” results in clinical tests.

That is, in this embodiment, the prior probability P (X) is a probability of “a certain abnormality frequency k”. As described above, since the abnormality frequency k is a discrete value (0 ≦ k ≦ M), the prior probability distribution is a discrete distribution. However, in this embodiment, M = 3. same as below. Of the inspection value data of the inspection item i for all clients stored in the database (different from the inspection database 14), the total number of inspection value data whose abnormality frequency is k is n _i (k). The prior probability P (X = k) is expressed as in Equation 9. However, u is an integer of 0 ≦ u ≦ M.

  Here, the database referred to when calculating the prior probability P (X) is a database in which test results of clinical tests of a large number of clients are accumulated, and although not shown, a network such as the Internet Is connected so as to be communicable with the abnormality frequency estimation device 12. In this database, as the number of clients increases, the inspection results also increase. In the database, the test results increase and decrease with the year and month, and the number of clients also increases and decreases. However, the same pre-processing as described above is performed on the inspection value used for calculating the prior probability P (X).

[Equation 9]

  In addition, the likelihood function P (Y | X) representing the likelihood of the average value of the test item i subjected to the clinical test at a certain abnormality frequency k of the test item i to be estimated is obtained by using the density of the average value vector. Calculated. However, since the average value vector is multidimensional, the density of the multidimensional distribution is estimated based on the inspection value.

The distribution density is obtained for each abnormality frequency k of the inspection item i to be estimated. In order to associate the inspection item i with the abnormality frequency k, the combination G _{i, k} of the client j and the inspection item i is defined as shown in Equation 10.

[Equation 10]

  In this embodiment, as a method for estimating the density of the distribution, a multi-dimensional kernel density estimation is adopted. The likelihood function P (Y | X = k) for performing the τ′-th estimation of the client j ′ of the inspection item i ′ is defined as shown in Equation 11. However, “′” is added to mean a client different from the client j. Further, FIG. 8 shows a conceptual diagram of a method for calculating (determining) the likelihood function P (Y | X = k). However, while the kernel is one-dimensional in FIG. 8, the actual kernel is multi-dimensional as described above.

[Equation 11]

[Equation 12]

  Here, N is the number of combinations of j and τ, D is the number of test items i that have actually undergone clinical tests, and h is the bandwidth shown in Formula 13. However, the bandwidth means a variable for setting the above-described kernel width. In this example, the bandwidth is based on Silverman's rule of thumb as shown in Equation 13.

[Equation 13]

  A posteriori probability P (X | Y) is calculated for each abnormality frequency k = X = 0, 1, 2, 3 using the Bayes estimation formula. Then, an expected value E (X | Y) is calculated based on the inspection value of the client j. Since the expected value E (X | Y) shown in Equation 14 is a real number, the abnormality frequency k has four stages of 0, 1, 2, and 3. Therefore, the calculation result of the expected value E (X | Y) is Rounded off and converted to a real number between 0 and 3. This is the estimation result of the abnormality frequency k.

[Formula 14]

  In this way, the abnormality frequency k is calculated and estimated. Next, overall processing in the system 10 will be described. First, in the system 10, the inspection item i is selected by comprehensively determining the past inspection values of the client j. Here, all of the test items i include a test item i (test target) that needs to be clinically tested (a test item i that becomes teacher data), and a test item i that estimates an abnormal frequency k (a target of estimation). are categorized.

  Here, a method (algorithm) for selecting the inspection item i used for estimation will be described. The test item i is classified into a type in which a clinical test is always performed every time a health examination is performed, and a type in which whether or not a clinical test is performed is determined by an algorithm (selection algorithm) for selecting (selecting) the test item i. The

  The items that must be inspected every time a health checkup is performed are the test item i that is performed in a general health checkup, the test item i that has a low unit cost, and the test item that is highly important in making a diagnosis. i. Further, as the objects of the selection algorithm, the inspection item i having a high inspection cost that is difficult to perform the inspection every time and the inspection item i having a low importance in performing the diagnosis.

  Selection algorithms are based on conventions and are defined by physicians and specialists. However, in this embodiment, the inspection item i (inspection item i to be inspected) to be subjected to the clinical examination in “when the abnormality has occurred twice or more consecutively in the past” is selected. As another example, the selected algorithm is defined as “if a specific test item i is abnormal, the other test item i is a subject of a clinical test”, and the test item i to be tested is selected. It is also possible. By this selection algorithm, the test item i is selected (selected) as the test item i that needs to be clinically tested and the test item i that is estimated without performing a clinical test.

  A clinical test is actually performed on the test item i selected as the test item i to be tested. The result of the clinical test (test result) is used when the abnormal frequency is estimated.

  Next, the respective inspection values of the inspection item i to be inspected and the inspection item i to be estimated are acquired from the inspection database 14. As described above, in order to estimate the abnormality frequency k based on the test value acquired from the test database 14, the preprocessing for converting the test value into a real value or linearly converting the reference value is performed. The average value of the detection values for each inspection item i to be inspected and the abnormal frequency of the inspection result (inspection value) for each inspection item i to be estimated are calculated using the inspection values for which inspection has been performed.

  Then, the abnormality frequency k of each inspection item i to be estimated is estimated by Bayesian estimation. At this time, an average value and an abnormality frequency calculated in advance are used. However, the average value is given in a form corresponding to the abnormality frequency (“0 times / 3 times” to “3 times / 3 times”) of the inspection item i to be estimated. Then, the estimated abnormality frequency is displayed (output) on, for example, a monitor (not shown).

  Further, for the inspection item i to be estimated, it is determined (estimated) whether the current inspection result (inspection value) is normal or abnormal. However, in this case, it must be performed based on the abnormal frequency estimated for the inspection item i.

  FIG. 9 shows a correspondence table for estimating the abnormal frequency estimation result and the normal or abnormal test result of the last clinical test. However, the numbers in parentheses indicate the abnormal frequency. In addition, as a result of the correspondence, a case where an abnormal frequency different from the estimated abnormal frequency appears is marked with an asterisk (*).

  In the correspondence table shown in FIG. 9, for example, “(normal) (normal)” as an inspection result means that the previous and previous inspection results were normal. Strictly speaking, the left parenthesis is the previous test result, and the right parenthesis is the previous test result. In FIG. 9, “(abnormal) (normal)” is omitted, because this is the same as the case of “(normal) (abnormal)”. Therefore, either of the inspection results in parentheses may be the previous result (or the previous result).

  As described above, when the test result of the previous time and the previous time is “(normal) (normal)” and the error frequency is estimated to be “0 times / 3 times”, this time, the previous time and the previous time It was estimated that there were no abnormalities in the three clinical tests. For this reason, it is determined (estimated) that the latest (current) test result is “normal”. In addition, when the test result of the previous time and the previous time is “(normal) (normal)”, if the abnormal frequency is estimated to be “1/3 times”, this time, the previous time and the previous time 3 times It is estimated that there was one abnormality in the clinical examination. For this reason, it is estimated that this test result is “abnormal”.

  In addition, if the previous and previous test results were “(normal) (normal)” and the abnormal frequency was estimated to be “2/3 times” or “3/3 times” This time, it is estimated that there were two or more abnormalities in the previous and previous three clinical examinations. However, in reality, twice of the past three times were normal, and the estimation results would be contradictory. Therefore, as in the case of “1/3 times”, contradiction is avoided by estimating only the current test result as “abnormal”.

  Although detailed description is omitted, the same applies to other cases. Further, although detailed description is omitted, for the examination item i in which the examination result is determined to be “abnormal” based on the estimation result of the abnormality frequency, a clinical examination is performed thereafter, and the examination value and The distinction between “normal” or “abnormal” is stored. On the other hand, for the inspection item i that is determined (estimated) that the inspection result is “normal” based on the estimation result of the abnormality frequency, the inspection value is not stored in the inspection database 14 and the inspection result is “normal”. It is remembered that there is.

  FIG. 10 is an illustrative view showing one example of a memory map 30 of the RAM 12b shown in FIG. As shown in FIG. 10, the RAM 12 b includes a program storage area 32 and a data storage area 34. The program storage area 32 stores a program (control program) for the overall processing of the abnormality frequency estimation device 12. The control program includes a main processing program 32a, an inspection item selection program 32b, a preprocessing program 32c, an abnormality frequency estimation program 32d, an abnormality frequency output program 32e, an inspection result estimation program 32f, an estimation result output program 32g, and the like.

  The main processing program 32a is a program for processing the main routine of the abnormality frequency estimation device 12. The examination item selection program 32b is a program that complies with the selection algorithm described above. The examination item i (the examination item i to be examined) that needs to be clinically tested and the examination item i (the examination to be estimated) that estimates the abnormal frequency k. Item i) is selected (selected). Specifically, when the inspection value data for the selected (target) client j is acquired, and the inspection value data of the previous time and the previous time is referred to for each inspection item i, both of them are abnormal The inspection item i is selected as the inspection item i to be inspected.

  As will be described later, it is not necessary to select an inspection item i (basic inspection item) for which a clinical inspection is always performed.

  The pre-processing program 32c is a program for converting test values that are not real numbers into test values acquired from the test database 14 and performing linear conversion on the acquired test values. The abnormality frequency estimation program 32d uses the average value calculated in advance based on the inspection value of the inspection item i to be inspected and the abnormality frequency calculated in advance based on the inspection value (inspection result) of the inspection item i to be estimated, This is a program for estimating the abnormality frequency of each inspection item i to be estimated by a method based on Bayesian estimation. However, each inspection value is preprocessed according to the preprocessing program 32c.

  The abnormality frequency output program 32e is a program for outputting the estimated abnormality frequency to a monitor (not shown) or a printer (not shown) connected to the abnormality frequency estimation device 12. However, data about a table describing the estimated abnormality frequency for each inspection item i may be output (transmitted) to an external computer or storage medium. The same applies to an estimation result output program 32g described later.

  The inspection result estimation program 32f is a program for estimating (determining) whether the inspection value of the inspection item i for which the abnormality frequency is estimated is normal or abnormal. The estimation result output program 32g calculates the result estimated according to the test result estimation program 32f, that is, the result (normal or abnormal) estimated for the test result (test value) of the test item i for which the abnormal frequency has been estimated. 12 is a program for outputting to a monitor (not shown) or a printer (not shown) connected to 12.

  Although not shown, the program storage area 32 also stores a program for transmitting / receiving commands and data to / from the examination database 14.

  The data storage area 34 stores (temporarily stores) data necessary for execution of the control program, and is provided with a counter (timer) and a flag necessary for execution of the control program. For example, the inspection ID for the inspection item i to be inspected selected according to the inspection item selection program 32b and the inspection ID of the inspection item i to be estimated are stored in an identifiable manner.

  FIG. 11 is a flowchart of the abnormality frequency estimation and abnormality determination process of the CPU 12a shown in FIG. Although illustration is omitted, the selection (selection) process of the examination item i is executed in advance, the examination item i to be examined is actually subjected to a clinical examination, and the examination value and the determination result (normal or abnormal) are examined. It is assumed that it is stored (recorded) in the database 14.

  As shown in FIG. 11, when the CPU 12a starts the abnormal frequency estimation and the inspection result estimation processing, the selected (target) client j (user ID) is input in step S1. In the next step S3, the inspection item i (inspection ID) to be estimated is input. However, in steps S1 and S3, an administrator of the abnormality frequency estimation system 10 inputs a user ID and each inspection item i.

  Subsequently, in step S5, the inspection value data of all inspection items i for the past three times (for example, last time, last time, three times before) for all users (clients) and the user ID input in step S1 are shown. The test value data of the test item i that has been clinically tested for the client j is read. Here, the CPU 12a reads the data of the inspection value from the inspection database 14 and stores it in a buffer area (not shown) of the RAM 12b. In the next step S7, pre-processing is performed on the read test value data.

  In this embodiment, the preprocessing is executed in the abnormal frequency estimation and the inspection result estimation processing. However, the preprocessing may be executed in advance separately from this processing. Or you may make it provide the test | inspection database 14 by which the pre-processing was performed.

  In this embodiment, the data of the last three inspection values are read. However, the data of the previous three previous inspection values may be used as long as they are continuous.

  In subsequent step S9, the inspection value of the inspection item to be estimated for the user ID input in step S1 is acquired. Here, the CPU 12a reads the inspection value data of the past three times (three times before, two times before and last time) for the inspection item i to be estimated from the data of the inspection value stored in the buffer area.

  In the next step S11, the inspection value of the inspection item to be inspected for the user ID input in step S1 is acquired. Here, the CPU 12a reads the inspection value data of the past three times (three times before, two times before and last time) for the inspection item i to be inspected from the inspection value data stored in the buffer area.

  In step S13, reference values for all inspection items are acquired. That is, the CPU 12a gives an instruction to read the inspection item information 18 (inspection item data) to the inspection database 14, and in response thereto, the inspection item data given from the inspection database 14 is stored in the buffer area of the RAM 12b. However, if the client j corresponding to the user ID is a male, the reference value upper limit (female) and reference value lower limit (female) data need not be read. Similarly, when the client j corresponding to the user ID is a female, the reference value upper limit (male) and reference value lower limit (male) data may not be read.

  Subsequently, in step S15, using the inspection value data read in step S11, the average value of the past inspection values of each inspection item i to be inspected is calculated according to Equation 6, and the inspection value read in step S9. Using this data, the abnormal frequency of the past inspection value (inspection result) of each inspection item i to be estimated is calculated according to Equation 7. At this time, the reference value (data of inspection item information) acquired in step S13 is referred to, and normality or abnormality of the inspection value (inspection result) is determined. However, normality or abnormality of the inspection value (inspection result) may be determined and registered in the inspection database 14 in advance. In such a case, the process of step S13 can be deleted.

  In step S17, a prior probability distribution, that is, a prior probability P (X) is calculated according to Equation 9. However, as described above, here, the inspection value is read from a database different from the inspection database 14 and pre-processed. Further, in step S19, the average value of the test values of each test item i to be inspected in the three consecutive clinical tests (this time, the previous time and the previous time) including the current time is calculated according to Equation 6. In the next step S21, a likelihood function (Equation 11) is obtained (defined) from the average value and the abnormality frequency calculated in Step S15, and in Step S23, the likelihood of the average value calculated in Step S19 is calculated according to Equation 11. calculate. In step S25, the posterior probability distribution is calculated according to Equation 8. Subsequently, in step S27, the abnormality frequency is estimated according to Equation 14.

  When the abnormality frequency is estimated, the abnormality frequency is output in step S29. Although illustration is omitted, data of a table or a graph indicating the estimated abnormality frequency is, for example, a monitor connected to the abnormality frequency estimation device 12, a printer, or other data communication connected to the abnormality frequency estimation device 12 Output to computer. The same applies to the case where the determination result is output in step S33 described later.

  Further, in step S31, from the estimation result of the abnormality frequency for three consecutive test values (inspection results) including this time and the previous and previous test results (judgment results), according to the correspondence table shown in FIG. Then, normality or abnormality is estimated (determined) for the inspection result of the inspection item i for which the abnormality frequency has been estimated. In step S33, the estimation (judgment) result in step S31 is output, and the abnormal frequency estimation and inspection result estimation processing is terminated.

  In addition, although detailed description was abbreviate | omitted, a process is performed about each test | inspection item i of estimation object in step S17-S29.

  Furthermore, in order to show that the abnormality frequency estimation system 10 of this embodiment is effective, a simulation was performed on the assumption that an abnormality frequency estimation method was combined with an actual health checkup. The table of FIG. 12 shows the flow of actual medical examination (clinical examination) and estimation of abnormal frequency.

  However, in the simulation, at least one clinical test is performed on all test items before the first clinical test, and the test results (test values) of the first and second clinical tests are used. Thus, in the third time, the average value of the past three times of the inspection value for the inspection target item and the abnormality frequency of the past three times of the inspection value (test result) for the estimation target item are calculated. Assume that a statistical model has been created.

  As shown in FIG. 12, in the first medical examination day (November 29, 2006), that is, the first clinical examination, all examination items i related to the disease to be detected are examined. In the second medical examination (March 6, 2007), after the doctor gave advice to the medical examination recipient based on the first clinical examination, all examination items i were Inspected.

  In the third medical examination (June 18, 2007), the doctor gives advice to the medical examination recipient based on the second clinical examination. Moreover, according to the selection algorithm, based on the past test values, the test item i that needs to be clinically tested and the test item i that estimates the abnormal frequency are selected (selected). Here, in addition to the test item i that is always subjected to the clinical test, the test item i that has been abnormal twice in the past is selected. Then, the inspection item i that needs to be clinically examined is inspected, and the abnormality frequency is estimated for the inspection item i that is not inspected.

  Further, in the fourth health examination (October 1, 2007), the doctor gives advice to the health examination recipient based on the third clinical examination. Moreover, according to the selection algorithm, based on the past test values, the test item i that needs to be clinically tested and the test item i that estimates the abnormal frequency are selected (selected). Then, the inspection item i that needs to be clinically examined is inspected, and the abnormality frequency is estimated for the inspection item i that is not inspected.

  According to such a flow, an actual clinical examination and an estimation of an abnormal frequency are performed. However, in the second and subsequent times, the examinee receives the previous examination result on the day when the medical examination is performed, and receives a health examination after receiving advice such as lifestyle habits from the doctor based on the examination result.

  Next, conditions for selecting (selecting) the inspection item i when performing analysis based on simulation will be described. Specifically, the inspection item i selected as the basic inspection item i (basic inspection item) and the conditions of the selection algorithm will be described.

  First, nine items used in general health examinations were defined as basic test items. Specifically, “AST”, “ALT”, “γ-GTP”, “neutral fat”, “HDL cholesterol”, “white blood cell count”, “red blood cell count”, “blood glucose” and “urine protein” are the basis. Selected as an inspection item. About these basic test items, a clinical test is always performed, and a test item i (teacher data) used for estimating an abnormal frequency is used. In this embodiment, in order to improve the accuracy of estimation, the inspection item i is added within a range where the inspection cost does not increase.

  When actually performing a medical examination, a plurality of test items i are set as one set in the blood test and the urine test, and the plurality of test items i are charged together. For example, the basic test items (9 items) also belong to three categories of “peripheral blood general test”, “general urine test”, and “biochemical test” as shown in FIG.

  Seven items in the peripheral blood general examination are “white blood cell count (WBC)”, “red blood cell count (RBC)”, “MCV”, “MCH”, “MCHC”, “hemoglobin”, and “hematocrit”. In addition, eight items of general urine tests are “urine protein”, “urine sugar”, “urine specific gravity”, “urine pH”, “urine urobilinogen qualitative”, “urine urobilin qualitative”, “urine ketone body” and “urine ketone body”. An occult blood reaction.

  Furthermore, 32 items of biochemical tests are “total protein (TP)”, “albumin”, “TTT”, “ZTT”, “CK”, “ALT (GPT)”, “AST (GOT)”, “LDH”. ”,“ ALP ”,“ γ-GTP ”,“ cholinesterase ”,“ LAP ”,“ serum amylase ”,“ acid phosphatase (ACP) ”,“ creatinine <serum> ”,“ uric acid <serum> ”,“ urea nitrogen ” <Serum>, “blood glucose”, “neutral fat (TG)”, “phospholipid”, “free fatty acid (NEFA)”, “total cholesterol”, “HDL cholesterol”, “sodium (Na)”, “potassium” (K) "," Klor (Cl) "," Magnesium (Mg) "," Calcium (Ca) "," Inorganic phosphorus (P) "," Serum iron (Fe) "," TIBC <Serum> "and" In total bilirubin That.

  In the simulation, among the 47 items described above, 46 items excluding the “urine ketone body” of the general urine test in which no abnormality was found in the test values stored in the test database 14 are used for estimation. Item i. As described above, the estimation result is output as an abnormal frequency (four stages of 0 to 3). However, whether or not there is an abnormality in the last health check, that is, whether the inspection result (test value) of the inspection item i for which the abnormality frequency has been estimated this time is normal or abnormal is the abnormal frequency estimation result and the previous and previous inspection results. From the above, it is estimated based on the table shown in FIG.

  Next, three evaluations of estimation rate, cost (test cost) and risk (possibility of missing a disease) performed in this example will be described.

(1) Evaluation of estimation rate In this example, evaluation based on an estimation rate indicating whether or not the abnormal frequency was correctly estimated was performed. However, the estimation rate means the ratio of the inspection item i that has correctly estimated the abnormality frequency among all the inspection items i that are the objects of estimation.

(2) Evaluation of cost (test cost) When performing a clinical test, a test cost is generated for each test item i. However, there is a cost that is generated (collectively) for a certain inspection item i and a cost that is generated for each inspection item i. Although the examination cost varies depending on the medical institution, in this embodiment, as one guideline, when the examination database 14 is recorded, verification (evaluation) is performed using the expense borne by the client.

  When only a general clinical test is performed, only the test item i described with reference to FIG. 13 may be tested, and the total cost of the test is 4,030 yen. This is almost the same as the cost for general medical examinations.

  However, when examining the progression of cancer in detail, it must undergo a tumor marker test. The inspection cost of the tumor marker is about 1,000 to 8,000 yen (list price) per inspection item. Depending on the selected algorithm (selection process), it may be expensive depending on the test item i that is determined to require clinical testing. Therefore, by performing the abnormal frequency estimation process of this embodiment, it is evaluated how much the inspection cost is reduced by suppressing (reducing) the inspection item i, and which inspection item i is the inspection cost. Evaluate whether there is a possibility of pushing up.

(3) Evaluation of risk (possibility of missing a disease) By performing the abnormal frequency estimation process of this embodiment, the test item i that is abnormal is judged to be normal, and it is intended to prevent the disease from being missed. The evaluation performed will be described.

  As described above, the estimation results are output in four stages from 0 (“0 times abnormal / 3 times during”) to 3 (“3 times abnormal / 3 times during”). For example, when the estimation result is 0 (“0 times abnormal / during 3 times”) in spite of the fact that all three consecutive times are abnormal, it is excluded from the inspection item i to be inspected, There is a risk of suffering from diseases related to this test item i. Therefore, for each inspection item i, an evaluation is performed with respect to a ratio at which an abnormal frequency lower than the actual abnormal frequency is estimated.

  Next, the result of the simulation will be described. In order to verify whether the selection process of the inspection item i in the system 10 is effective, a simulation was performed for each of the following two conditions. The first condition is that a certain inspection item i (the above 46 items) is always selected. The second condition is that the inspection item i is selected according to the selection algorithm of this embodiment, and the inspection item i is changed every time (the above-mentioned 46 items + α).

  FIG. 14 shows the result of a simulation (hereinafter referred to as “first simulation”) when the inspection item i is always constant. As shown in FIG. 14, the number of items inspected inspection item i (inspected item), the number of items in correct inspection item i (correct item), and the number of items in inspection item i that are incorrect (judged higher than actual). And the number of items of the inspection item i that are incorrect (judged lower than actual) is tabulated by gender.

  Moreover, the result of having totaled each result for every accumulation inspection number is shown by FIG. 15 and FIG. However, FIG. 15 shows the result of counting for men, and FIG. 16 shows the result of counting for women. In FIGS. 15 and 16, the horizontal axis represents the cumulative number of inspections, and the vertical axis represents the ratio of the inspection items i to be estimated. The same applies to FIGS. 18 and 19 described later.

  On the other hand, FIG. 17 shows the result of a simulation (hereinafter referred to as “second simulation”) when the inspection item i is changed each time according to the selection algorithm. As shown in FIG. 17, the number of detected items, the number of correct items, the number of incorrect items (determined higher than actual) and the items of incorrect items (determined lower than actual) and the items of incorrect items (determined lower than actual) Numbers are tabulated by gender.

  Moreover, the result of having totaled each result for every accumulation inspection number is shown by FIG. 18 and FIG. However, FIG. 18 shows the result of counting for men, and FIG. 19 shows the result of counting for women.

  In the case of the first simulation, the estimation rate is 79.4% on average for men and 80.3% on average for women. On the other hand, in the case of the second simulation, the estimation rate is 83.8% on average for men and 84.2% on average for women. Therefore, a result that the estimation rate was higher when the inspection item i was changed every time according to the selected algorithm was obtained.

  FIG. 20 shows the result of comparing the average number of incorrect (incorrect) items in the first simulation (without selection) and in the second simulation (with each selection by the selection algorithm). It is. FIG. 20 shows the change in the average number of incorrect answer items according to the cumulative number of inspections.

  In the case of the second simulation, the number of incorrect answer items was an average of 11.2 items, which was 28.4% lower than that in the case of the first simulation. In the case of the second simulation, among the incorrect answer items, the number of inspection items whose frequency of abnormality was estimated to be low was an average of 4.52 items, which was 41.8% lower than that in the case of the first simulation. Furthermore, in the case of the second simulation, the result of adding the average of the newly inspected inspection item and the correctly inspected inspection item is 64.7 items. Increased by 3 items.

  In the case of the first simulation, since the selected inspection item i is constant, the inspection cost is a fixed amount (4,030 yen) as described above. On the other hand, in the case of the second simulation, since the inspection item i is changed every time, the inspection cost varies depending on the selected inspection item i even for the same client j. 21 shows the transition of the inspection cost in the case shown in FIG. 18, and FIG. 22 shows the transition of the inspection cost in the case shown in FIG. As shown in FIG. 21 and FIG. 22, both male and female inspection costs were in the range of 23,000 to 27,000 yen.

  The cost when selecting an inspection item according to the selection algorithm is as shown in FIGS. 21 and 22, and is higher than when the selection is not performed according to the selection algorithm. It can be seen that the cost for performing all the inspections is about 120,000 yen, while the cost is reduced by about 80%.

  In the above-described embodiment, when the abnormality frequency is estimated by the method based on Bayesian estimation, the distribution of the abnormality frequency recorded in the examination database 14 is treated as a probability distribution (prior distribution). There is also a high probability that the abnormality frequency is zero with respect to the posterior distribution. As a result, there is a high possibility that the inspection item i that should be abnormal is determined to be normal. Therefore, in order to reduce the bias of the prior distribution, smoothing (filtering) is performed using mathematical expressions.

  In a certain inspection item i, the ratio of the number of data having an abnormality frequency of 0 to 3 is represented by P (X = 0), P (X = 1), P (X = 2), and P (X = 3), respectively. To do. The sum of these ratios is 1, as shown in Equation 15.

[Equation 15]

  As shown in Expression 16, Gaussian filtering was used for filtering. However, “′” means after filtering.

[Equation 16]

  The result of the filtering is shown in FIG. As shown in FIG. 23, when filtering is performed, the abnormality frequency tends to be estimated higher than when filtering is not performed, and the average number of incorrect (incorrect) items increases. It became.

  According to this embodiment, by using a method based on Bayesian estimation, a distribution of the probability of a certain abnormality frequency of each inspection item to be estimated is set as a prior probability, and the average of past inspection values of each inspection item to be inspected is averaged. The test item to be estimated from the average of multiple test values including the current time for each test item to be inspected using the relationship between the value and the abnormal frequency of the test value of the past multiple times of each test item to be estimated Therefore, it is not necessary to assume the relationship between the inspection item to be estimated and the inspection item to be inspected, and the estimation accuracy for estimating the abnormality frequency and the inspection result can be improved.

  In addition, according to this embodiment, by determining whether the test result of the estimation target is normal or abnormal from the estimation result of the abnormal frequency, it is possible to know the test item that requires the clinical test, The risk of missing an illness or non-illness can be reduced as much as possible.

  In this embodiment, the likelihood function is obtained using the inspection values for all the clients recorded in the inspection database. However, the present invention is not limited to this. You may make it use the test value about a specific some client. If there are enough user (client) test values (to create a statistical model) to find the likelihood function, use only the test values for the user who estimates the abnormal frequency It may be.

  In this embodiment, a plurality of processes including an inspection item selection process, a pre-process, an abnormality frequency estimation process, and an inspection result estimation process are executed by one computer. Can be applied to a system in which a plurality of computers are distributedly processed.

DESCRIPTION OF SYMBOLS 10 ... Abnormal frequency estimation system 12 ... Abnormal frequency estimation apparatus 14 ... Examination database

Claims (6)

A database that records the test values of a predetermined number of test items for clinical tests for each of a plurality of clients,
Of all the test values recorded in the database, the average value for the predetermined number of times for each of a plurality of first test items is calculated using the test values of the previous clinical test. First average value calculating means for
An abnormality frequency calculating means for calculating an abnormality frequency in the predetermined number of examinations for a second examination item other than the first examination item, using a test value of the previous clinical examination;
Extraction means for extracting a relationship between an average value vector having each average value calculated by the first average value calculation means as an element and an abnormal frequency of the second inspection item calculated by the abnormal frequency calculation means;
For the selected one client, a second average for calculating an average value of the test values for the predetermined number of tests including the test values of the current clinical test for each of the plurality of first test items recorded in the database From the average value for each of the plurality of first inspection items calculated by the second average value calculation means using the value extraction means and the relationship extracted by the extraction means, An abnormality frequency estimation device comprising abnormality frequency estimation means for estimating an abnormality frequency at the predetermined number of examinations including the current clinical examination.   The abnormality frequency estimation apparatus according to claim 1, wherein the abnormality frequency is a number of times determined to be abnormal in the predetermined number of clinical tests.   Based on the abnormality frequency of the second inspection item estimated by the abnormality frequency estimation means and the inspection value of the second inspection item before the previous time, the normality or abnormality of the inspection result of the second inspection item is estimated. The abnormality frequency estimation apparatus according to claim 1, further comprising a test result estimation unit that performs the inspection. For each of the multiple client comprises a database that was recorded test values for test items of a predetermined number of clinical examination, an abnormal frequency estimation program of the abnormality frequency estimating apparatus,
In the processor of the abnormal frequency estimation device,
Of all the test values recorded in the database, the average value for the predetermined number of times for each of a plurality of first test items is calculated using the test values of the previous clinical test. A first average value calculating step,
An abnormal frequency calculation step of calculating an abnormal frequency in the predetermined number of tests for a second test item other than the first test item, using test values of the previous clinical test.
An extraction step for extracting a relationship between an average value vector having each average value calculated in the first average value calculating step as an element and an abnormal frequency of the second inspection item calculated in the abnormal frequency calculating step;
For the selected one client, a second average for calculating an average value of the test values for the predetermined number of tests including the test values of the current clinical test for each of the plurality of first test items recorded in the database From the average value for each of the plurality of first inspection items calculated in the second average value calculation step using the value extracted in the value calculation step and the relationship extracted in the extraction step, An abnormal frequency estimation program for executing an abnormal frequency estimation step for estimating an abnormal frequency in the predetermined number of tests including the current clinical test. For each of the multiple client comprises a database that was recorded test values for test items of a predetermined number of clinical examination, an abnormal frequency estimation method of the abnormality frequency estimating apparatus,
The processor of the abnormal frequency estimation device is
(A) The average of the predetermined number of tests for each of a plurality of first test items using the test values of previous clinical tests among all test values of all test items recorded in the database Calculate the value,
(B) Using the test value of the clinical test before the previous time, calculating an abnormality frequency in the predetermined number of times for the second test item other than the first test item,
(C) extracting a relationship between an average value vector having each average value calculated in the step (a) as an element and an abnormal frequency of the second inspection item calculated in the step (b);
(D) For the selected one client, an average value of the test values for the predetermined number of times including the test values of the current clinical test for each of the plurality of first test items recorded in the database is calculated. And (e) using the relationship extracted in step (c), from the average value for each of the plurality of first inspection items calculated in step (d), for the second inspection item An abnormal frequency estimation method for estimating an abnormal frequency in the predetermined number of tests including the current clinical test. Database was recorded test values for test items of a predetermined number of laboratory tests for each of the multiple clients,
Of all the test values recorded in the database, the average value for the predetermined number of times for each of a plurality of first test items is calculated using the test values of the previous clinical test. First average value calculating means for
An abnormality frequency calculating means for calculating an abnormality frequency in the predetermined number of examinations for a second examination item other than the first examination item, using a test value of the previous clinical examination;
Extraction means for extracting a relationship between an average value vector having each average value calculated by the first average value calculation means as an element and an abnormal frequency of the second inspection item calculated by the abnormal frequency calculation means;
For the selected one client, a second average for calculating an average value of the test values for the predetermined number of tests including the test values of the current clinical test for each of the plurality of first test items recorded in the database From the average value for each of the plurality of first inspection items calculated by the second average value calculation means using the value extraction means and the relationship extracted by the extraction means, An abnormality frequency estimation system comprising abnormality frequency estimation means for estimating an abnormality frequency at the predetermined number of examinations including the current clinical examination.

Images