JP2006163465A - Medical treatment information analysis apparatus, method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To dynamically predict the result of unknown medical treatment data from known medical treatment data by storage base inference without requiring model construction. <P>SOLUTION: A storage base inference section 18 calculates and outputs the degree of influence for indicating the ratio of the contribution of each explanation variable to an object variable, the prediction value of the object variable value of inputted unknown medical treatment data, the degree of confidence of the prediction value, and the reliability scale of the prediction itself, based on the known medical treatment data 14 accumulated in a database 12. A degree-of-influence display section 20 displays the degree of influence outputted by the storage base inference section 18 corresponding to the object variable to be noticed. An alarm output section 22 sets prediction value determination conditions, a threshold to the degree of confidence in the prediction value, and a threshold to a reliability scale in advance, and compares the prediction value of the unknown medical treatment data outputted from the storage base inference section 18, the degree of confidence of the prediction value, and the scale of reliability for determination, thus outputting an alarm. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

The present invention relates to a medical information analysis apparatus, method, and program using memory-based reasoning, and in particular, a medical information analysis apparatus, method, and method for predicting results of unknown medical data from known medical data by memory-based reasoning. Regarding the program.

  In recent years, medical information processing systems such as electronic medical records, ordering systems, and medical accounting systems are rapidly spreading in various hospitals. With the spread of these medical information processing systems, not only accumulates medical information, but also automatically and efficiently extracts valuable information such as causal relationships and regularity, and grasps past trends. In addition, there is an increasing demand for medical information analysis that is useful for effective treatment and efficient hospital management by predicting the cost and variance of patients currently being treated using the grasped past trends.

  2. Description of the Related Art Conventionally, in a medical information analysis apparatus, a model is constructed in which regularity and conditions are extracted through analysis in advance on accumulated data including medical information, and a mechanism for performing prediction based on the model is realized in the medical information analysis apparatus. And according to the input conditions from the outside, the investigation object of medical information was predicted by the mechanism which performed the prediction, and the result was output.

At that time, analysis and model construction use statistical methods such as multivariate analysis and predictive classification mechanisms such as decision trees and neural networks.
JP 2000-155681 A JP 2003-323601 A

  However, in such a conventional medical information analyzer, a model is constructed using data accumulated at a certain point in time, and prediction is performed using the constructed model. Typically, the model will not fit into the current situation, at which point analysis specialists will need to reanalyze the data and rebuild the model, which is costly and this is a continuous process for the equipment. Was hindering automatic operation.

  Also, there are many cases where an algorithm that outputs only the prediction result and does not output the certainty of the prediction clearly is used, which makes it difficult to flexibly use the prediction result. Furthermore, in the method of constructing a model using accumulated data, it is usually assumed that accumulated data represents all the problems covered, but in reality, the data obtained initially is one of the problems. There is a possibility that it is biased in the department, or cases that did not appear in the past with the change of the world appear. However, it was impossible to judge the reliability of such a situation, that is, the prediction itself, from the model once constructed.

The present invention dynamically predicts the result of unknown case data from known case data that has results such as patient satisfaction and cost for conditions such as electronic medical records and medical accounting without the need for model construction. An object of the present invention is to provide a medical information analysis apparatus, method, and program.

The present invention provides a medical information analysis program executed by a computer. The medical information analysis program of the present invention is stored in a computer,
Based on known medical data stored in a database that stores known medical data, which is a set of records composed of explanatory variables and objective variables, the degree of influence that represents the proportion of each explanatory variable contributing to the objective variable was entered A memory-based reasoning step for calculating and outputting a predicted value of an objective variable value of unknown medical data and its reliability and a reliability measure of the prediction itself;
An impact display step for displaying the impact output by the memory-based reasoning step;
Prediction of unknown medical data output from the memory-based inference step by presetting a prediction value judgment condition indicating a range in which the prediction value can be regarded as equivalent to an expected result, a threshold value for the confidence value of the prediction value, and a threshold value for the reliability measure A warning output step for comparing and determining the value and its confidence and reliability measures and outputting a warning;
Is executed.

  Here, the warning output step warns that the unknown medical data is a new appearance case and the predicted value is a reference level when the reliability measure is equal to or less than the threshold value.

The warning output step is
If the predicted value does not satisfy the judgment condition and the certainty factor is less than or equal to the threshold, the predicted value is different from the expected result and warns that the certainty factor is low
When the predicted value does not satisfy the determination condition and the certainty level exceeds the threshold value, a warning is given that the predicted value is different from the expected result but the certainty level is high.

The medical information analysis program of the present invention further includes:
An actual result data input step for inputting the actual result data obtained by replacing the predicted result of the unknown medical data with the actual result obtained over time;
Each of the first case in which the actual result data is added to the known medical data, the second case in which the actual result data is added and the known medical data before the specified time is deleted, and the third case in which neither the actual result data is added nor deleted A prediction result evaluation step for evaluating and evaluating the prediction accuracy of unknown medical data;
A known medical data update step for updating the known medical data in the database according to the case with the highest prediction accuracy according to the evaluation result of the prediction result evaluation step;
It is provided with.

The known medical data update step
When the first case has the highest prediction accuracy, the actual result data is added to the known medical data,
When the second case has the highest prediction accuracy, the actual result data is added to the known medical data and the known medical data before the specified time is deleted.
When the prediction accuracy of the third case is the highest, the actual result data is discarded.

  The explanatory variables of medical data targeted by the present invention include electronic medical records, medical accounting, ordering, examination history, patient information, questionnaire results from patients, cost and profit information accumulated on the hospital side, The objective variable is patient satisfaction, cost and benefit, variance value or diagnostic group classification.

  The memory-based reasoning step calculates the degree of influence that represents the ratio of each explanatory variable of the medical data to the patient satisfaction, cost and benefit, variance value, or diagnosis group classification that are the objective variables of the medical data, and displays the degree of influence The data trend of which explanatory variables strongly influence the objective variable is displayed in the section.

In another form of the medical information analysis program of the present invention,
A medical data attribute classification step for classifying medical data based on specific explanatory variable values of medical data input from the outside,
A plurality of medical analysis steps including a storage-based reasoning unit, an impact display unit, and a warning output unit provided for each of the medical data classified in the medical data attribute classification step;
A final result output step for outputting the result of the medical analysis unit corresponding to the classification of the medical data attribute classification step;
It is provided with.

The present invention provides a medical information analysis method. The medical information analysis method of the present invention includes:
Based on known medical data stored in a database that stores known medical data, which is a set of records composed of explanatory variables and objective variables, the degree of influence that represents the proportion of each explanatory variable contributing to the objective variable was entered. A memory-based reasoning step for calculating and outputting a predicted value of an objective variable value of unknown medical data and its reliability and a reliability measure of the prediction itself;
An impact display step for displaying the impact output by the memory-based reasoning step;
Predictive value judgment conditions indicating a range in which the predictive value can be regarded as equivalent to an expected result, a threshold value for the predictive value certainty factor, and a threshold value for the reliability measure are preset, and unknown medical data output from the memory-based inference step A warning output step for comparing and determining the predicted value and its certainty factor and reliability measure, and outputting a warning;
It is provided with.

The present invention provides a medical information analysis apparatus. The medical information analyzer of the present invention is
A database of known medical data that is a set of records composed of explanatory variables and objective variables;
Based on the known medical data stored in the database, the degree of influence representing the ratio of each explanatory variable to the input objective variable, the predicted value of the objective variable value of unknown medical data whose objective variable value is unknown, and its certainty And a memory-based reasoning unit that calculates and outputs a reliability measure of the prediction itself;
An influence display section for displaying the influence output from the memory-based reasoning section;
Prediction of unknown medical data output from the memory-based reasoning unit by presetting a prediction value judgment condition indicating a range in which the prediction value can be regarded as equivalent to the expected result, a threshold value for the confidence value of the prediction value, and a threshold value for the reliability measure A warning output unit that compares and determines the value and its certainty factor and reliability measure, and outputs a warning;
It is provided with.

The details of the medical information analysis method and program according to the present invention are basically the same as those of the medical information analysis apparatus of the present invention.

  According to the present invention, based on the known medical data composed of explanatory variables (conditions) and objective variables (results) accumulated in the database, the degree of influence representing the ratio of each explanatory variable to the objective variable is calculated. By displaying, it is possible to know the trend of medical data by knowing which explanatory variable value has an influence on the target variable (result) of interest. Moreover, it becomes possible to predict the objective variables such as the final patient satisfaction, cost, profit, and variance of the patient currently being treated with high accuracy by inference based on storage of unknown medical data.

  Also, by evaluating the reliability measure obtained as a prediction result, it is warned that a new appearance case that can easily occur in medical data is a new appearance case without excluding this prediction result as an abnormality. Can be encouraged.

Furthermore, when the actual result is obtained for the predicted result over time, the prediction accuracy by the known medical data is evaluated, and the actual result data is added as the known medical data, or the old related data is deleted at the same time as the addition. By updating whether or not to add or delete, it is possible to update the known medical data stored in the database so as to improve the prediction accuracy.

  FIG. 1 is a block diagram of a functional configuration showing an embodiment of a medical information analysis apparatus according to the present invention. In FIG. 1, a database 12 is provided for the medical information analysis apparatus 10 of the present invention, and many known medical data 14 are stored in the database 12. Here, the known medical data 14 is a set of records composed of explanatory variables (conditions) and objective variables (results).

  The medical information analyzer 10 is provided with a storage base reasoning unit 18, an influence level display unit 20, and a warning output unit 22. The unknown medical data input unit 16 inputs unknown medical data that has an explanatory variable (condition) but an unknown objective variable (result) into the storage base inference unit 18 of the medical information analyzer 10.

  Based on the known medical data 14 accumulated in the database 12, the memory-based reasoning unit 18 of the medical information analysis apparatus 10 determines the degree of influence that represents the ratio of each explanatory variable to the objective variable, and the value of the input objective variable. Calculates and outputs the predicted value of the objective variable value of unknown unknown medical data, its reliability, and the reliability measure of the prediction itself.

  The influence level display unit 20 displays the degree of influence output by the storage base inference unit 18. Further, the warning output unit 22 is a storage-based reasoning unit in a state in which a predicted value determination condition indicating a range in which the predicted value can be regarded as equivalent to an expected result, a threshold for the predicted value certainty, and a threshold for the reliability measure are preset. The predicted value of the unknown medical data output from the device is compared with the certainty factor and the reliability measure, and a necessary warning is output.

  Here, the medical data which is processed by the medical information analyzer of the present invention will be described. Medical data targeted by the present invention means data included in electronic medical records, medical accounting, ordering, examination history, etc., meaning personal information such as patient gender and age, clinical plan such as clinical path and their deviation Variance value, medical treatment date and time, medical treatment, medication content, test content and results, findings, diagnostic results, diagnosis group classification (Diagnosis Procedure Combination: DPC), and final results such as discharge, complete cure, hospital transfer and death included.

  Also included are expenses paid by patients, medical care, and insurance points for testing. Furthermore, patient satisfaction obtained through questionnaires, presence / absence of referral letters from local clinics, referral rates, and cost and profit information related to medical care accumulated at the hospital are also included.

  The medical information analysis apparatus in FIG. 1 is realized by, for example, hardware resources of a computer as shown in FIG. In the computer of FIG. 2, the bus 101 of the CPU 100 includes a RAM 102, a hard disk controller (software) 104, a floppy disk driver (software) 110, a CD-ROM driver (software) 114, a mouse controller 118, a keyboard controller 122, a display controller 126, A communication board 130 is connected.

  The hard disk controller 104 is connected to the hard disk drive 106 and is loaded with a program for executing the medical information analysis processing of the present invention. When the computer is started, a necessary program is called from the hard disk drive 106 and expanded on the RAM 102. To execute.

  A floppy disk drive (hardware) 112 is connected to the floppy disk driver 110 and can read and write to the floppy disk (R). A CD drive (hardware) 116 is connected to the CD-ROM driver 114, and data and programs stored on the CD can be read.

  The mouse controller 118 transmits an input operation of the mouse 120 to the CPU 100. The keyboard controller 122 transmits the input operation of the keyboard 124 to the CPU 100. The display controller 126 displays the prediction result on the display unit 128. The communication board 130 uses a communication line 132 including wireless communication to communicate with a device in the network or an external device on the Internet.

  Next, the principle of operation of memory-based reasoning (hereinafter referred to as “MBR”) by the memory-based reasoning unit 18 of the present invention will be described.

  FIG. 3 is a conceptual diagram of MBR. FIG. 3A shows a two-dimensional space as a multi-dimensional space for simplicity of explanation. Each dimension represents an explanatory variable of medical data, and “+” and “−” mean coordinate points of an objective variable value. Yes. That is, the known medical data is represented as the coordinate points of the explanatory variables X and Y in the multidimensional space. FIG. 3B shows a prediction result of unknown medical data at the coordinate point 24.

  The basic principle of MBR prediction is that k known medical data close to the coordinate point 24 of unknown medical data, for example, K = 3, that is, medical data at coordinate points 26, 28, and 30 in the accumulated known medical data. Search is performed, and the predicted value is determined by the weighted sum.

  FIG. 4 is an example of known medical data, in which sex, age, examination A, examination B, and DPC name are set as explanatory variables, and patient satisfaction is set as an objective variable.

  FIG. 5 shows unknown medical data 15 that performs MBR prediction based on the known medical data in FIG. 4 and sets gender, age, examination A, examination B, and DPC name as explanatory variables. Satisfaction is undecided.

The MBR prediction by the storage base inference unit 18 is performed in the following procedure.
(1) The degree of influence of a certain objective variable is calculated on the accumulated known medical data 14.
(2) The degree of similarity between the unknown medical data 15 and the known medical data 14 is calculated.
(3) Using the similarity obtained in the above (2), k pieces most similar to the unknown medical data 15 are selected from the known medical data 14 and set as similar medical data.
(4) Using the obtained k similar medical data, the prediction result is determined.
(5) A certainty factor is calculated for the obtained prediction result.
(6) A reliability measure is calculated for the obtained prediction result.
Details of each will be described below.

(Influence calculation)
The degree of influence is obtained by calculating the strength of the degree of influence that each explanatory variable or each explanatory variable value has on the objective variable value from the known medical data. As this calculation method, an MIC (Mutual Information Content) method obtained for each explanatory variable, a CCF (Cross-Category Feature impulse) method obtained for each explanatory variable value, a new CCF (new Cross-Category Feature Implant) method, and the like. Is used. Each impact calculation is performed according to the following formula.

  In the case of the MIC method, the influence is calculated by the following equation.

ここで、cは各目的変数値を、iは説明変数の番号を、vは説明変数の値を、p(v、c)は、説明変数が値vをとり目的変数が値ｃをとる結合確率を表す。

Where c is the value of each objective variable, i is the number of the explanatory variable, v is the value of the explanatory variable, and p (v, c) is a combination where the explanatory variable has the value v and the objective variable has the value c. Represents a probability.

  That is, the similarity value of the MIC method is a value calculated for each explanatory variable according to the mutual information amount. When the bias of the coupling probability with the objective variable in a certain explanatory variable is large, If the influence value w at that time is close to 1 and the deviation of the coupling probability is small, the explanatory variable is regarded as not important for prediction, and the influence value w is set close to 0.

  In the case of the CCF method, the influence is calculated by the following equation.

ここで、cは各目的変数値を、ｉは説明変数の番号を、ｖは説明変数の値を、Ｐ（ｃ|ｖ）は、説明変数が値ｖをとる時に目的変数値ｃをとる確率を表す。

Here, c is the value of each objective variable, i is the number of the explanatory variable, v is the value of the explanatory variable, and P (c | v) is the probability of taking the objective variable value c when the explanatory variable takes the value v. Represents.

That is, the similarity value in the CCF method means that when the explanatory variable takes the value v, the probability of taking the value c of the objective variable is squared and summed for all classes. As a result, when the explanatory variable takes a certain value v, if the single objective variable value c is necessarily taken, the degree of influence is 1, and the probability of taking any objective variable value c by the explanatory variable value v is equal. Has a minimum value of 1 / _Nc . N _c represents the number of objective variable values.

  In the case of the new CCF method, the influence is calculated by the following equation.

ここで、ｃは各目的変数値を、ｉは説明変数の番号を、ｖは説明変数の値を、Ｎ_cは目的変数値の数を、Ｐ（ｃ）は目的変数値がｃとなる確率を、Ｐ（ｃ|ｖ）は、説明変数が値ｖをとる時に目的変数が値ｃをとる確率を表す。また、ｑ_v（ｃ）は、説明変数が値ｖをとる時の目的変数値ｃをとる確率の、もともと目的変数値がｃとなる確率に対する比を意味している。

Here, c is each objective variable value, i is the explanatory variable number, v is the explanatory variable value, N _c is the number of objective variable values, and P (c) is the probability that the objective variable value is c. , P (c | v) represents the probability that the objective variable takes the value c when the explanatory variable takes the value v. Further, q _v (c) means the ratio of the probability of taking the objective variable value c when the explanatory variable takes the value v to the probability that the objective variable value is originally c.

Compared to the CCF method, this new CCF method
(1) Consider distribution bias of objective variable values,
(2) When a certain explanatory variable value v does not contribute to the determination of the objective variable value, the influence degree is set to 0.
The two points are improved.

  That is, when the distribution of the objective variable value c when taking a certain explanatory variable value v matches the entire objective variable value distribution, the numerator is 0 and the influence is 0 (minimum value). Can eliminate the effect on the prediction. On the other hand, when only a single objective variable value c is taken when taking a certain explanatory variable value v, the influence value is 1.0 as in the CCF method.

  FIG. 6 shows an example of the influence of each explanatory variable value on the explanatory variable “age” when the new CCF method of equation (3) is used, and FIG. 7 shows each of the explanations when the MIC method of equation (1) is used. An example of the influence value of a variable is shown.

  In the MIC method, one influence value is given for each explanatory variable, and in the new CCF method, one influence value is given for each explanatory variable value. In the case of the CCF method, the degree of influence is calculated by the equation (2). Therefore, although the influence value is different from the calculation result by the new CCF method, the format is the same as in the new CCF method of FIG. One influence value is obtained for the variable value or the explanatory variable range.

In any case of the MIC method, the CCF method, and the new CCF method, if there is a missing value in the known medical data, the impact is calculated after deleting the corresponding record. Further, when the explanatory variable has a numerical attribute, the value v of the explanatory variable means not a single value but a numerical range.

(Calculation of similarity between unknown medical data and known medical data)
Similarity is a measure of similarity between medical data. The similarity between the unknown medical data 15 and each known medical data is obtained by the following equations (4) and (5).

ここでｗ_i（ｖ）は、ＭＩＣ法、ＣＣＦ法、ｎｅｗＣＣＦ法のいずれかで求めた未知医療データ１５のｉ番目の説明変数がｖの時の影響度を表す。即ち、類似度は医療データ間の影響度付き距離の逆数を意味しており、両医療データが似ているほど類似度は高くなる。

Here, w _i (v) represents the degree of influence when the i-th explanatory variable of the unknown medical data 15 obtained by any one of the MIC method, the CCF method, and the new CCF method is v. That is, the similarity means the reciprocal of the distance with influence between medical data, and the similarity is higher as the two medical data are more similar.

In addition, when the missing value is included in the known medical data, the single attribute distance of the corresponding attribute of the unknown medical data 15 is 1, and conversely, when the missing value is included in the unknown medical data 15, the corresponding attribute of the known medical data The similarity is calculated with the distance between the single attributes of 0 as 0.

(Select k items that are most similar to unknown medical data from known medical data)
Based on the equations (4) and (5), the similarity between each of the plurality of known medical data and the unknown medical data that is one data is obtained, k pieces having the highest similarity are selected from the known medical data, To do.

Here, k is determined by the following method.
(1) Designated by the user in advance.
(2) Considering a part of the known medical data as unknown medical data, repeating the prediction work with a plurality of k values, and automatically determining the k value with the highest predicted correct answer rate as the optimum value.

FIG. 8 shows similar medical data selected as k = 5 with respect to the unknown medical data 15 of FIG.

(Determination of prediction results using k similar medical data)
First, the case where the objective variable is a category value attribute will be described. In this case, the similarity total T _c for each objective variable value is obtained by the following formula (6) using similar medical data. Here, S _j means the similarity of the jth similar medical data among the selected k similar medical data to the unknown medical data 15.

次に（７）式に従い、得られた各目的変数値におけるＴ_cの内、最大のＴ_cを与える目的変数値ｃを、予測値ｃ_predictとして決定する。

According then (7), of the T _c of each target variable values obtained, the target variable value c which gives the maximum T _c, is determined as a prediction value c _predict.

次に目的変数が数値属性の場合を説明する。この場合には、（８）式を用いて予測値ｃ_predictが決定される。ここで、類似医療データ数をｋとし、ｊ番目の類似医療データと未知医療データ１５との間の類似度をＳ_j、ｊ番目の類似医療データの目的変数値をｃ_jとする。

Next, the case where the objective variable has a numerical attribute will be described. In this case, the predicted value c _predict is determined using equation (8). Here, the number of similar medical data is k, the similarity between the jth similar medical data and the unknown medical data 15 is S _j , and the objective variable value of the _jth similar medical data is c _j .

即ち、類似度による重み付き加算和により予測値を決定する。

That is, the predicted value is determined by the weighted addition sum based on the similarity.

(Calculate confidence for prediction results)
The certainty factor is a scale representing the probability of occurrence of a prediction result. It is obtained as follows.

If the target variable is the category value attribute First, with respect to the similarity sum of the target variable value, as a percentage of similarity total prediction result c _predict, is determined by the equation (9).

また目的変数が数値属性の場合には、同様に、（１０）式により求められる。ここで、σ_cは、目的変数の標準偏差を意味する。

Similarly, when the objective variable has a numerical attribute, it can be obtained by the equation (10). Here, σ _c means the standard deviation of the objective variable.

(Calculate reliability measure for prediction results)
Predictive reliability measures are used to determine whether unknown medical data has never appeared before or whether there is a high probability of outliers. Calculated based on frequency. Details of the calculation method are disclosed in Patent Document 2.

  FIG. 9 shows an example of a prediction result, its certainty factor, and a reliability measure for the unknown medical data 15 of FIG. 5 obtained from the similar data of FIG.

  As described above, in the MBR, it is possible to present the degree of influence of the known medical data 14, to predict the unknown medical data 15 using the known medical data 14, and to provide the certainty factor and the reliability scale.

  Next, the influence display unit 20 in FIG. 1 will be described. The influence level display unit 20 outputs, based on the similarity calculated by the storage base inference unit 18, the data base inference unit 18 outputs the data trend indicating which value of which explanatory variable strongly influences the target variable. This is a function for displaying the degree of influence in a graph.

  FIG. 10 is an example in which the influence degree display unit 20 displays the influence degree by the new CCF method or the CCF method shown in FIG. 6 as a graph, and shows the influence value and the number of cases in the explanatory variable “age”. .

  That is, the left bar graph is the number of cases, and the right bar graph is the influence value. It can be said that the greater the degree of influence of the right bar graph, the more the attribute value divided by age influences the objective variable. Here, when the graph is displayed in color, the objective variable value is represented by a color so that it can be determined at a glance which objective variable value has the strongest influence.

  On the other hand, due to its characteristics, the smaller the number of cases, the more likely it is to have a large value.Therefore, the number of cases is also displayed as shown in the left bar graph, and it is important to have explanatory variables that have as many cases as possible and have a high degree of influence. to decide.

  When the new CCF method or the CCF method is used as the influence degree calculation method, the influence degree is obtained for each explanatory variable value. In order to compare the influence degree for each explanatory variable at a glance, each explanation of FIG. Find the expected impact value of a variable. FIG. 11 is an example in which the influence degree expected value of each explanatory variable of FIG. 7 is displayed as a graph by the influence degree display unit 20 and is arranged in descending order of the expected influence degree value. From the display of the degree of influence in FIG. 11, the strength of the influence (expected value) on the objective variable can be known for each explanatory variable. Here, the expected impact value is obtained by the following equation.

ただし、iは説明変数Ａに含まれる説明変数値の番号、Ｉは説明変数Ａに含まれる説明変数値の種類、ｗ_iは、説明変数Ａのｉ番目の説明変数値の影響度値、Ｒ_iは説明変数Ａのｉ番目の説明変数値を持つレコード数を表す。

Where i is the number of the explanatory variable value included in the explanatory variable A, I is the type of the explanatory variable value included in the explanatory variable A, w _i is the influence value of the i-th explanatory variable value of the explanatory variable A, R _i represents the number of records having the i-th explanatory variable value of the explanatory variable A.

  In addition, even in the explanatory variable including the missing value, the missing value is regarded as one type of the explanatory variable value, and the influence value of the missing value is set to 0, so that it can be handled uniformly by the equation (11). In the case of a numerical explanatory variable, the area into which the numerical value is divided is treated as a value in the same way using equation (11). Similarly, FIG. 11 can be displayed even by the degree of influence by the MIC method in which a value is obtained for each explanatory variable.

  In the embodiment shown in FIG. 4 for such medical data, for example, patient satisfaction, cost and profit, variance value, and DPC are designated as objective variables in order, and the designated objective variable is strongly influenced. By displaying the influence degree of the data trend such as the explanatory variable and the explanatory variable value obtained by the storage base inference unit 18 as a graph by the influence degree display unit 20, the data tendency can be grasped.

  Next, the function of the warning output unit 22 in FIG. 1 will be described. The prediction reliability measure calculated by the stored data inference unit 18 is calculated from the distance between similar cases of unknown medical data on the basis of the average distance and frequency of similar cases, and the unknown medical data. Can be used to determine whether a new appearance has never occurred before.

  Especially in medical data, a new appearance case can easily occur from the following situation. That is, since there are a wide variety of cases, it is practically difficult to store all cases as data stored in one hospital. Therefore, there is a possibility of encountering a new case that has never been handled. In addition, medicines and medical equipment are also making progress every day, and there is a possibility that new medicines and equipment will always be used in the situation where medicines and equipment are developed every year.

  FIG. 12 shows an example of medical data as a new appearance example. Compared to the other medical data, the 201 medical data shows an example of using a new medicine “W” and a new equipment “Z” for a patient of “age 12” who has never visited the hospital until now. It is shown.

  When such a case of new appearance is encountered, a similar case of the new appearance case does not exist in the accumulated existing medical data, but the new case is classified and predicted using the stored case. Since there is a high possibility of outputting an incorrect result, it is necessary to devise an alarm.

  In the medical field where there is a high probability of encountering new cases due to the appearance of new cases, medicines, equipment, etc., it should always be determined whether there is a case similar to the current case in the accumulated data Therefore, effective use of the reliability measure is important.

  FIG. 13 shows the presence or absence of similar cases in the medical data accumulated in the database. In FIG. 13, each axis giving an explanatory variable represents the name of equipment used and the name of a medicine as explanatory variables of medical data. In practice, this means a multidimensional space with a plurality of explanatory variables, but here only a two-dimensional space is displayed for convenience.

  In FIG. 13, the unknown medical data at the coordinate point 32 to be analyzed exists in the known medical data in which similar medical data has been accumulated, and the inference result using these similar medical data is considered valid. On the other hand, the unknown medical data at the coordinate point 34 to be analyzed does not exist in the medical data in which similar medical data has been accumulated. That is, the unknown medical data to be analyzed at the coordinate point 34 is a case of new appearance that has never appeared in the past, and the inference result obtained from the accumulated medical data should be limited to a reference level.

  A criterion for determining whether or not medical data similar to unknown medical data to be analyzed exists in the accumulated known medical data is a prediction reliability measure. As shown in FIG. 13, in the unknown medical data to be analyzed at the coordinate point 32 where the similar medical data exists, the prediction reliability measure is high as “1.0” as shown in the prediction result 33, but the similar medical data In the unknown medical data to be analyzed at the coordinate point 34 where no data exists, the prediction reliability measure is low as “0.5” as shown in the prediction result 35.

The warning output unit 22 performs not only the reliability of the prediction result calculated by the storage-based inference unit 18 but also a reliability measure, and performs the following processing.
(1) If the reliability measure is less than or equal to the threshold value, warning is given that unknown medical data is a new appearance case and the predicted value is a reference level.
(2) When the predicted value does not satisfy the determination condition and the certainty factor is equal to or less than the threshold value, a warning is given that the predicted value is different from the expected result and the certainty factor is low.
(3) When the predicted value does not satisfy the determination condition and the certainty factor exceeds the threshold value, a warning is given that the predicted value is different from the expected result but the certainty factor is high.

  Here, the judgment condition of the predicted value means a range in which the predicted value can be regarded as equivalent to the expected result. Further, the user specifies in advance the objective variable value to be warned, its certainty value, each threshold value of the reliability scale, and the warning content to be issued when the threshold value is exceeded.

  FIG. 14 (A) designates patient satisfaction as an objective variable, inputs unknown medical data of a patient who is currently being treated and whose patient satisfaction is not confirmed, and predicts patient satisfaction by the memory-based reasoning unit 18. It is an example of the prediction result in the case.

The warning output unit 22 processes the prediction result as follows.
(1) If the patient satisfaction is “impossible” and the certainty is not less than the threshold value 0.8, a warning is given that “the patient cannot be satisfied with the treatment plan as it is”.
(2) If the reliability measure is less than or equal to the threshold of 0.7, a warning is given that “This patient has never appeared so far and should be observed carefully”.

  FIG. 14B shows an example of the prediction result when the benefit is specified as the objective variable, the unknown medical data of the patient who is currently being treated and the benefit is not confirmed, and the benefit is predicted by the memory-based reasoning unit 18. It is.

The warning output unit 22 processes the prediction result as follows.
(1) If the profit is negative (minus) and the certainty level is equal to or higher than the threshold value 0.8, a warning is given that “the treatment plan as it is is likely to be deficit”.
(2) If the reliability measure is less than or equal to the threshold of 0.7, a warning is given that “This patient has never appeared so far and should be observed carefully”.

  FIG. 14C shows an example of the prediction result when the variance is specified as the objective variable, the unknown medical data of the patient who is currently being treated and whose variance is not fixed is input, and the variance is predicted by the memory-based reasoning unit 18. It is.

The warning output unit 22 processes the prediction result as follows.
(1) When the variance is greater than or equal to the threshold value of 1 day and the certainty factor is greater than or equal to the threshold value of 0.8, a warning is given that “the treatment plan as it is is likely to be out of the clinical path”.
(2) If the reliability measure is less than or equal to the threshold of 0.7, a warning is given that “This patient has never appeared so far and should be observed carefully”.

  FIG. 15 is a flowchart of medical information analysis processing by the medical information analysis apparatus 10 of FIG. In this medical information analysis processing, known medical data is input from the database 12 in step S1, and a predicted value determination condition, a certainty factor threshold value, and a reliability measure threshold value are input according to user settings in step S2. Subsequently, unknown medical data is input in step S3, and in step S4, an unknown result as an objective variable of unknown data is predicted by the storage base inference unit 18. By this prediction, a predicted value, a certainty factor, and a reliability measure are generated and output in step S5.

  Subsequently, in step S6, it is checked whether or not the predicted value satisfies the determination condition input in step S2. If the determination condition is satisfied, the process proceeds to step S7, and it is checked whether the certainty factor is equal to or greater than a threshold value. If it is equal to or greater than the threshold value, the process proceeds to step S8, where an expected value as an expected result is output. In this case, no warning is given. When the certainty factor is less than the threshold value in step S7, “Would be expected prediction value is likely to be obtained but reliability is low” is output as a warning in step S9.

  If the predicted value does not satisfy the determination condition in step S6, the process proceeds to step S10 to check whether the certainty factor is equal to or greater than a threshold value. If the certainty factor is greater than or equal to the threshold value, the process proceeds to step S11, and a warning “highly likely to be different from the expected result” is issued as a warning. If the certainty factor is less than the threshold value in step S10, the process proceeds to step S12, and a warning “not unlike the expected result but the certainty factor is low” is warned.

  On the other hand, after the process of step S8, S9, S11, or S12 is completed, the reliability measure is compared with the threshold value in step S13. If the reliability measure is equal to or greater than the threshold value, the process is terminated. If there is, it is warned in step S14 that it is a “new appearance case and the predicted value is a reference level”.

  In the medical information analysis process of FIG. 15, following the warning display process by the influence level display unit 20 provided in the medical information analysis apparatus 10 of FIG. As a process different from the warning process, the user causes the data to be displayed as a warning by displaying the influence level as necessary.

  FIG. 16 is a block diagram showing another embodiment of the medical information analysis apparatus according to the present invention. In this embodiment, the analysis result of unknown medical data obtained from the medical information analysis apparatus 10 of FIG. Then, when the actual result is obtained over time, the known medical data in the database 12 is obtained by using the actual result data in which the objective variable (result) of the unknown prediction data is replaced with the actually obtained result. It is characterized in that it is updated.

  In FIG. 16, in this embodiment, in addition to the medical information analysis apparatus 10 and the database 12, an actual result data input unit 40, a prediction result evaluation unit 42, and a known medical data update unit 44 are provided.

  The database 12 stores, in addition to the known medical data 14, unknown medical data 15 for which a prediction result has already been obtained, a predicted value as a prediction result, its certainty factor 36, and a reliability measure 38, respectively. The medical information analysis apparatus 10 is the same as that shown in FIG. 1 and includes a storage base inference unit 18, an influence level display unit 20, and a warning output unit 22.

  The actual result data input unit 40 replaces the objective function of the unknown medical data 15 with the actual result based on the actual result obtained over time after the prediction result of the unknown medical data 15 is obtained. Data is input to the prediction result evaluation unit 42.

The prediction result evaluation unit 42 obtains and evaluates the prediction accuracy of the unknown medical data 15 in a state where the known medical data 14 in the database 12 divided into the following three cases is updated.
(1) In the first case, the actual result data is added to the known medical data 14.
(2) In the second case, the actual result data is added to the known medical data 14, and the known medical data before the designated time is deleted.
(3) In the third case, neither actual result data is added nor deleted from the known medical data 14.

  In the first case, the second case, and the third case, the unknown medical data 15 is input in a state where the known medical data 14 is updated, and the medical information analyzer 10 obtains a prediction result. The known medical data update unit 44 updates the known medical data 14 in the database 12 according to the case with the highest prediction accuracy based on the evaluation result of the prediction result evaluation unit 42.

  That is, when the prediction accuracy of the first case is the highest, the actual result data is added to the known medical data 14, and when the prediction accuracy of the second case is the highest, the actual result data is added to the known prediction data 14 and at the specified time. Previously known medical data, that is, related data is deleted, and when the prediction system of the third case is the highest, the actual result data is discarded without being added to or deleted from the known medical data 14.

  Here, a cross-validation method is used as a prediction accuracy evaluation method in the prediction result evaluation unit 42. FIG. 17 is an explanatory diagram of the cross-validation method. In this example, the cross-validation is shown when the known medical data 14 is divided into five. When the known medical data 14 is divided into five parts, one place of the known medical data 14 is regarded as unknown medical data 46, and the remaining four places are regarded as known medical data 48 and input to the storage-based reasoning unit 18, and the prediction result Ask for confidence.

  Subsequently, for the known medical data 14 divided into five, the combination is repeated five times, and the prediction result of the known medical data 14 used at this time is measured by matching the actual result data with the obtained prediction result. .

  FIG. 18 is a flowchart of known medical data update processing according to the embodiment of FIG. In FIG. 18, known medical data and unknown medical data are input in step S <b> 1, and predicted values, certainty factors, and reliability scale prediction results for unknown medical data are input in step S <b> 2. Subsequently, in step S3, the threshold value of the reliability scale and the data investigation time are set. Subsequently, in step S4, actual result data is input. Regardless of the prediction result and the actual result data, if the reliability measure of the prediction result is equal to or greater than the threshold, the target medical data as the actual result data is converted to the known medical data. Add.

  When the reliability measure is less than the threshold value in step S5, the process proceeds to step S6, and the prediction accuracy of the known medical data is evaluated for cases A, B, and C. Case A is the first case, and the target medical data is added to the known medical data. Case B is the second case, in which the target medical data is added to the known medical data and the known medical data before the designated time is deleted. Case C is the third case, and neither addition nor deletion of the target medical data is assumed.

  Based on the prediction accuracy of the prediction results obtained for cases A, B, and C in step 6, when the prediction accuracy of case A is the highest in step S7, the known medical data of the target medical data in step S9 Add to If the prediction accuracy of case B is the highest in step S8, the target medical data is added to the known medical data in step S10, and at the same time, the known medical data before the specified period is deleted. On the other hand, when the prediction accuracy of case C is the highest, the process proceeds from step S7 and step S8 to step S11, and the target medical data is discarded without being included in the known medical data.

  FIG. 19 is a block diagram showing another embodiment of the medical information analysis apparatus according to the present invention. In this embodiment, medical data is classified according to its attributes, and medical information analysis is performed for each classified medical data. It is characterized in that processing is executed.

  In FIG. 19, in this embodiment, a data attribute classification unit 50 is provided. The data attribute classification unit 50 divides the known medical data 14 in the database 12 on the basis of the explanatory variable designated in advance, and the corresponding medical information. Input to the analysis units 10-1, 10-2, and 10-3. The unknown medical data input unit 16 inputs unknown medical data whose target variable is unknown. The unknown medical data is also input to the corresponding medical information analysis unit 10-1 based on the explanatory variable value in the data classification unit 50. 10-3.

The attribute classification by the data attribute classification unit 50 includes, for example, (1) each disease name of the disease name explanation variable (2) each DPC name of the DPC explanation variable (3) sex (4) outpatient or inpatient The designation of the explanatory variable serving as a reference is specified by the user in advance, or is automatically determined through accuracy evaluation by the cross-validation method of data division and prediction for each explanatory variable value.

  The data attribute classification unit 50 classifies the known medical data 14 of the database 12 and the unknown medical data from the unknown medical data input unit 16 using the specified explanatory variable value as a reference value, and analyzes the target medical information Input to any of the units 10-1 to 10-3 to perform prediction based on inference based on memory.

  The configuration of the medical information analysis units 10-1 to 10-3 is the same as that of the medical information analysis device 10 of FIG. 1 and includes a storage base reasoning unit 18, an influence level display unit 20, and a warning output unit 22, respectively.

  The final result output unit 52 includes a prediction result, a warning, and an analysis result of any one of the medical information analysis units 10-1 to 10-3 corresponding to the reference value as the explanatory variable value specified by the data attribute classification unit 50. The warning display is output and displayed as the final result.

  The present invention also provides a medical information analysis program, which has contents according to the flowcharts of FIGS.

  In addition, this invention is not limited to said embodiment, The appropriate deformation | transformation which does not impair the objective and advantage is included. Further, the present invention is not limited by the numerical values shown in the above embodiments.

Here, the features of the present invention are summarized as follows.
(Appendix)
(Appendix 1)
On the computer,
Based on known medical data stored in a database that stores known medical data, which is a set of records composed of explanatory variables and objective variables, the degree of influence that represents the proportion of each explanatory variable contributing to the objective variable was entered. A memory-based reasoning step for calculating and outputting a predicted value of an objective variable value of unknown medical data and its reliability and a reliability measure of the prediction itself;
An impact display step for displaying the impact output by the memory-based reasoning step;
A prediction value determination condition indicating a range in which the prediction value can be regarded as equivalent to an expected result, a threshold value for the prediction value certainty factor, and a threshold value for the reliability measure are preset, and the unknown value output from the storage-based inference step A warning output step for comparing and determining the predicted value of the medical data and its certainty factor and reliability measure, and outputting a warning;
Medical information analysis program characterized by causing (1)

(Appendix 2)
In the medical information analysis program according to attachment 1, the warning output step includes:
A medical information analysis program that warns that the unknown medical data is a new appearance case and the predicted value is a reference level when the reliability measure is equal to or less than a threshold value. (2)

(Appendix 3)
In the medical information analysis program according to attachment 1, the warning output step includes:
When the predicted value does not satisfy the determination condition and the certainty factor is equal to or less than the threshold value, the predicted value warns that the certainty factor is low unlike the expected result,
A medical information analysis program that warns that a predicted value is different from an expected result but has a high certainty factor when the predicted value does not satisfy a determination condition and the certainty factor exceeds a threshold value. (3)

(Appendix 4)
In the medical information analysis program described in Appendix 1,
An actual result data input step for inputting actual result data obtained by replacing the prediction result of the unknown medical data with an actual result obtained over time;
The first case in which the actual result data is added to the known case data, the second case in which the actual result data is added and the known medical data before the specified time is deleted, and the third case in which neither the actual result data is added nor deleted A prediction result evaluation step for obtaining and evaluating the prediction accuracy of the unknown medical data for each of
A known medical data update step for updating the known medical data in the database according to the case with the highest prediction accuracy according to the evaluation result of the prediction result evaluation step;
A medical information analysis program characterized by comprising (4)

(Appendix 5)
In the medical information analysis program according to attachment 4, the known medical data update step includes:
When the prediction accuracy of the first case is the highest, the actual result data is added to the known medical data,
When the second case has the highest prediction accuracy, the actual result data is added to the known medical data and the known medical data before the specified time is deleted,
The medical information analysis program, wherein the actual result data is discarded when the prediction accuracy of the third case is the highest. (5)

(Appendix 6)
In the medical information analysis program according to attachment 1, the explanatory variables of the medical data include electronic medical records, medical accounting, ordering, examination history, patient information, patient questionnaire results, costs accumulated on the hospital side, A medical information analysis program including benefit information, wherein the objective variable of the medical data is patient satisfaction, cost and benefit, variance value, or diagnosis group classification. (6)

(Appendix 7)
In the medical information analysis program according to attachment 6, in the memory-based reasoning step, each of the explanatory variables of the medical data includes the patient satisfaction, the cost and the profit, the variance value, or the diagnosis group classification as the objective variables of the medical data. A medical information analysis program characterized in that an influence degree representing a contribution ratio is calculated, and a data tendency indicating which variable strongly influences a target variable is displayed in the influence degree display step. (7)

(Appendix 8)
In the medical information analysis program described in Appendix 1,
A medical data attribute classification step for classifying the medical information data based on a specific explanatory variable of the medical information data input from the outer step;
A plurality of medical analysis steps comprising a storage-based reasoning step, an impact display step and a warning output step provided for each of the medical information data classified in the medical data attribute classification step;
A final result output step for outputting a result of a medical analysis step corresponding to the classification of the medical data attribute classification step;
A medical information analysis program characterized by comprising (8)

(Appendix 9)
Based on known medical data stored in a database that stores known medical data, which is a set of records composed of explanatory variables and objective variables, the degree of influence that represents the proportion of each explanatory variable contributing to the objective variable was entered. A memory-based reasoning step for calculating and outputting a predicted value of an objective variable value of unknown medical data and its reliability and a reliability measure of the prediction itself;
An impact display step for displaying the impact output by the memory-based reasoning step;
A prediction value determination condition indicating a range in which the prediction value can be regarded as equivalent to an expected result, a threshold value for the prediction value certainty factor, and a threshold value for the reliability measure are preset, and the unknown value output from the storage-based inference step A warning output step for comparing and determining the predicted value of the medical data and its certainty factor and reliability measure, and outputting a warning;
A medical information analysis method characterized by comprising: (9)

(Appendix 10)
In the medical information analysis method according to attachment 9, the warning output step includes:
When the reliability measure is equal to or lower than a threshold value, the medical information analysis method is characterized by warning that the unknown medical data is a new appearance case and the predicted value is a reference level.

(Appendix 11)
In the medical information analysis method according to attachment 9, the warning output step includes:
When the predicted value does not satisfy the determination condition and the certainty factor is equal to or less than the threshold value, the predicted value warns that the certainty factor is low unlike the expected result,
A medical information analysis method that warns that a predicted value is different from an expected result but has a high certainty factor when the predicted value does not satisfy a determination condition and the certainty factor exceeds a threshold value.

(Appendix 12)
In the medical information analysis method according to appendix 9,
An actual result data input step for inputting actual result data obtained by replacing the prediction result of the unknown medical data with an actual result obtained with the passage of time;
A first case where the actual result data is added to the known medical data, a second case where the actual result data is added and the known medical data before the specified time is deleted, and a third case where the actual result data is neither added nor deleted A prediction result evaluation step for obtaining and evaluating the prediction accuracy of the unknown medical data for each of
A known medical data update step for updating the known medical data in the database according to the case with the highest prediction accuracy according to the evaluation result of the prediction result evaluation step;
A medical information analysis method characterized by comprising:

(Appendix 13)
In the medical information analysis method according to appendix 9,
A medical data attribute classification step for classifying the medical information data based on a specific explanatory variable of the medical information data input from the outside;
A plurality of medical analysis steps including a storage-based reasoning unit, an impact display unit, and a warning output unit provided for each of the medical information data classified in the medical data attribute classification step;
A final result output step of outputting a result of the medical analysis unit corresponding to the classification of the medical data attribute classification step;
A medical information analysis method characterized by comprising:

(Appendix 14)
Based on known medical data stored in a database that stores known medical data, which is a set of records composed of explanatory variables and objective variables, the degree of influence that represents the proportion of each explanatory variable contributing to the objective variable was entered. A memory-based reasoning unit that calculates and outputs a predictive value of an objective variable value of unknown medical data and its reliability and a reliability measure of the prediction itself;
An influence display section for displaying the influence output from the storage base inference section;
A prediction value determination condition indicating a range in which the prediction value can be regarded as equivalent to an expected result, a threshold value for the prediction value certainty factor, and a threshold value for the reliability measure are preset, and the unknown value output from the storage-based reasoning unit A warning output unit that outputs a warning by comparing and determining the predicted value of medical data and its certainty and reliability scale;
A medical information analysis apparatus comprising: (10)

(Appendix 15)
The medical information analyzer according to appendix 14, wherein the warning output unit includes:
When the reliability measure is less than or equal to a threshold value, the medical information analysis apparatus warns that the unknown medical data is a new appearance case and the predicted value is a reference level.

(Appendix 16)
The medical information analyzer according to appendix 14, wherein the warning output unit includes:
When the predicted value does not satisfy the determination condition and the certainty factor is equal to or less than the threshold value, the predicted value warns that the certainty factor is low unlike the expected result,
When the predicted value does not satisfy the determination condition and the certainty factor exceeds a threshold value, the medical information analysis device warns that the predicted value is different from the expected result but the certainty factor is high.

(Appendix 17)
In the medical information analyzer according to appendix 14,
An actual result data input unit for inputting actual result data obtained by replacing the prediction result of the unknown medical data with an actual result obtained over time;
A first case where the actual result data is added to the known medical data, a second case where the actual result data is added and the known medical data before the specified time is deleted, and a third case where the actual result data is neither added nor deleted A prediction result evaluation unit that calculates and evaluates the prediction accuracy of the unknown medical data for each of
A known medical data update unit that updates the known medical data in the database according to the highest prediction accuracy according to the evaluation result of the prediction result evaluation unit;
A medical information analysis apparatus comprising:

(Appendix 18)
In the medical information analyzer according to appendix 14,
A medical data attribute classification unit for classifying the medical information data based on a specific explanatory variable of the medical information data input from the outside;
A plurality of medical analysis units including a storage-based reasoning unit, an impact display unit, and a warning output unit provided for each of the medical information data classified by the medical data attribute classification unit;
A final result output unit that outputs a result of the medical analysis unit corresponding to the classification of the medical data attribute classification unit;
A medical information analysis apparatus comprising:

The block diagram of the function structure of the medical information analyzer by this invention Explanatory drawing of the hardware environment of the computer to which the medical information analyzer of FIG. 1 is applied Explanatory drawing of the medical data in which the target variable exists in the multi-dimensional explanatory variable space showing the inference principle by the memory-based reasoning unit of FIG. Explanatory drawing of known medical data accumulated in database of FIG. Explanatory diagram of unknown medical data in Fig. 1 whose objective variable is unknown Explanatory drawing of an example of the degree of influence on age by the newCCF method Explanatory drawing of an example of the influence level of the MIC method, or an example of the expected impact level by the new CCF method or the CCF method Explanatory drawing of the example of the reciprocal degree of similarity of similar medical data when k = 5 with respect to the unknown medical data of FIG. Explanatory drawing of the example of the predicted value of the unknown medical data in FIG. 5 and its certainty factor and reliability measure Explanatory drawing of the influence graph made into a graph by the influence display part of FIG. Explanatory drawing of the influence degree expected value graph made into a graph by the influence degree display part of FIG. Illustration of examples of known medical data including newly emerging medical data Explanatory diagram of the distribution state of target variables in multivariable space for memory-based inference of medical data with similar medical data and medical data without similar medical data Explanatory diagram of examples of prediction results when the objective variables are patient satisfaction, profit, and variance Flowchart of medical information analysis processing according to the embodiment of FIG. Block diagram of another embodiment of the present invention for updating known medical data based on actual result data Explanatory drawing of the cross-validation method which processes a known medical data by dividing into 5 by the prediction result evaluation part of FIG. Flowchart of known medical data update processing according to the embodiment of FIG. A block diagram of another embodiment of the present invention in which known medical data and unknown medical data are divided based on specific explanatory variables to predict an objective variable of unknown medical data based on a memory base

Explanation of symbols

10: Medical information analysis devices 10-1 to 10-3: Medical information analysis unit 12: Database 14: Known medical data 15: Unknown medical data 16: Unknown medical data input unit 18: Memory-based reasoning unit 20: Influence degree display unit 22: Warning output units 24, 34: Coordinate points 26, 28.30, 32 of unknown medical data 36: Coordinate points 36 of similar medical data 36: Predicted value and its certainty 38: Reliability scale 40: Actual result data input unit 42 : Prediction result evaluation unit 44: Known medical data update unit 46: Deemed unknown medical data 48: Deemed known medical data 50: Data attribute classification unit 52: Final result output unit

Claims (10)

On the computer,
Based on known medical data stored in a database that stores known medical data, which is a set of records composed of explanatory variables and objective variables, the degree of influence that represents the proportion of each explanatory variable contributing to the objective variable was entered. A memory-based reasoning step for calculating and outputting a predicted value of an objective variable value of unknown medical data and its reliability and a reliability measure of the prediction itself;
An impact display step for displaying the impact output by the memory-based reasoning step;
A prediction value determination condition indicating a range in which the prediction value can be regarded as equivalent to an expected result, a threshold value for the prediction value certainty factor, and a threshold value for the reliability measure are preset, and the unknown value output from the storage-based inference step A warning output step for comparing and determining the predicted value of the medical data and its certainty factor and reliability measure, and outputting a warning;
Medical information analysis program characterized by causing
The medical information analysis program according to claim 1, wherein the warning output step includes:
A medical information analysis program that warns that the unknown medical data is a new appearance case and the predicted value is a reference level when the reliability measure is equal to or less than a threshold value.
The medical information analysis program according to claim 1, wherein the warning output step includes:
When the predicted value does not satisfy the determination condition and the certainty factor is equal to or less than the threshold value, the predicted value warns that the certainty factor is low unlike the expected result,
A medical information analysis program that warns that a predicted value is different from an expected result but has a high certainty factor when the predicted value does not satisfy a determination condition and the certainty factor exceeds a threshold value.
The medical information analysis program according to claim 1, further comprising:
An actual result data input step for inputting actual result data obtained by replacing the prediction result of the unknown medical data with an actual result obtained with the passage of time;
A first case where the actual result data is added to the known medical data, a second case where the actual result data is added and the known medical data before the specified time is deleted, and a third case where the actual result data is neither added nor deleted A prediction result evaluation step for obtaining and evaluating the prediction accuracy of the unknown medical data for each of
A known medical data update step for updating the known medical data in the database according to the case with the highest prediction accuracy according to the evaluation result of the prediction result evaluation step;
A medical information analysis program characterized by comprising
The medical information analysis program according to claim 4, wherein the known medical data update program includes:
When the prediction accuracy of the first case is the highest, the actual result data is added to the known medical data,
When the second case has the highest prediction accuracy, the actual result data is added to the known medical data and the known medical data before the specified time is deleted,
The medical information analysis program, wherein the actual result data is discarded when the prediction accuracy of the third case is the highest.
The medical information analysis program according to claim 1, wherein the explanatory variables of the medical data include electronic medical records, medical accounting, ordering, examination history, patient information, patient questionnaire results, costs accumulated on the hospital side, A medical information analysis program including profit information, wherein the objective variable of the medical data is patient satisfaction, cost and profit, variance value, or diagnosis group classification
7. The medical information analysis program according to claim 6, wherein the memory-based reasoning step includes each explanatory variable of medical data for patient satisfaction, cost and profit, variance value, or diagnosis group classification as objective variables of the medical data. A medical information analysis program characterized in that an influence degree representing a contribution ratio is calculated and a data tendency indicating which variables strongly influence the objective variable is displayed on the influence degree display section.
In the medical information analysis program according to claim 1,
A medical data attribute classification step for classifying the medical information data based on a specific explanatory variable of the medical information data input from the outside;
A plurality of medical analysis steps including a storage-based reasoning unit, an impact display unit, and a warning output unit provided for each of the medical information data classified in the medical data attribute classification step;
A final result output step of outputting a result of the medical analysis unit corresponding to the classification of the medical data attribute classification step;
A medical information analysis program characterized by comprising
Based on known medical data stored in a database that stores known medical data, which is a set of records composed of explanatory variables and objective variables, the degree of influence that represents the proportion of each explanatory variable contributing to the objective variable was entered. A memory-based reasoning step for calculating and outputting a predicted value of an objective variable value of unknown medical data and its reliability and a reliability measure of the prediction itself;
An impact display step for displaying the impact output by the memory-based reasoning step;
A prediction value determination condition indicating a range in which the prediction value can be regarded as equivalent to an expected result, a threshold value for the prediction value certainty factor, and a threshold value for the reliability measure are preset, and the unknown value output from the storage-based inference step A warning output step for comparing and determining the predicted value of the medical data and its certainty factor and reliability measure, and outputting a warning;
A medical information analysis method characterized by comprising:
A database of known medical data that is a set of records composed of explanatory variables and objective variables;
Based on the known medical data accumulated in the database, the degree of influence representing the contribution ratio of each explanatory variable to the objective variable, the predicted value of the objective variable value of the input unknown medical data, its reliability, and the reliability of the prediction itself A memory-based reasoning unit that calculates and outputs a sex measure;
An influence display section for displaying the influence output from the storage base inference section;
A prediction value determination condition indicating a range in which the prediction value can be regarded as equivalent to an expected result, a threshold value for the prediction value certainty factor, and a threshold value for the reliability measure are preset, and the unknown value output from the storage-based reasoning unit A warning output unit that outputs a warning by comparing and determining the predicted value of medical data and its certainty and reliability scale;
A medical information analysis apparatus comprising:

Images