CN106909792A - Hospital's Indexes Abnormality pattern automatic testing method - Google Patents

Abstract

The invention provides a kind of hospital's Indexes Abnormality pattern automatic testing method, including：Step 1, obtains the historical data of index system；Step 2, the historical data is split, to obtain being segmented achievement data；Step 3, extracts the characteristic value of the segmentation achievement data；Step 4, by characteristic value standardization；Step 5, according to standardization after the characteristic value calculate it is described segmentation achievement data exceptional value；Step 6, historical data obtains exceptional value threshold value according to the historical data；Step 7, abnormal patterns are determined according to the exceptional value and exceptional value threshold value.The present invention can help the abnormal patterns of the users such as hospital automatic detection index system in management, reduce the requirement to manager's managerial experiences, detect the abnormal patterns of more science, and with instantaneity, detect at any time abnormal.

Description

Hospital's Indexes Abnormality pattern automatic testing method

Technical field

The present invention relates to Indexes Abnormality detection technique field, more particularly to a kind of hospital's Indexes Abnormality pattern automatic detection side Method.

Background technology

Constantly perfect and perfect along with what hospital information was set up, hospital generates substantial amounts of data, and the data have The features such as being worth high, dimension is big.Complexity just because of data is high, when some indexs of hospital occur abnormal, it is difficult to find, And the abnormality detection of some important indicators is to Decision-making of Hospital Management important.Therefore how doctor is timely and effectively detected Institute services the abnormal patterns of figureofmerit, it is ensured that normally operation has great importance for hospital.Hospital services figureofmerit refers to extremely Hospital services figureofmerit deviates the situation of its normal mode.With the constantly improve of hospital information system, data volume is continuously increased, Hospital's critical services abnormal influence to the normal operation of hospital of figureofmerit is increasing.Thus how accurately and rapidly to detect Indexes Abnormality, and rational response is made, it is to ensure hospital's normally one of precondition of operation.

Hospital services figureofmerit mainly has two aspects, and both macro and micro hereinafter referred to as services figureofmerit.Hospital's macroscopic view clothes Business amount is in hospital actual occupancy bed day, discharge number etc., macro services including out-patient department person-time, emergency call salving person-time, patient What amount was generally determined by social demand.Microcosmic amount refers to all departments, the volume of services of each inside of section office, as hospital orders Medicine, hygienic material, the specimen amount of inspection, piece amount, reasonable arrangement personnel, equipment bed etc. are taken the photograph by dept. of radiology.Hospital services amount The abnormal patterns detection technique of index (Number of Outpatients, measures in hospital, amount for surgical etc.) contributes to hospital administrators to find hospital's phase in time The abnormal conditions of index are closed, search problem the reason for occurring, solve problem in time, it is ensured that hospital's work in every runs well.

At present, the mode of the existing detection abnormal patterns of hospital, mainly rule of thumb judge index is hospital administrators No exception, this method for detecting abnormality there are problems that a lot：(1) can not note abnormalities in time, often through after a while Data summarization (when such as monthly magazine is reported) manager just note abnormalities, cause to solve a problem promptly, time lag；(2) need Manager has certain managerial experiences, and the requirement to manager is higher, and reproducibility is poor；(3) due to the experience of manager Difference, causes the abnormal patterns for detecting with more empirical, personal subjectivity, lacks scientific, reasonability.

The content of the invention

The invention provides a kind of hospital's Indexes Abnormality pattern automatic testing method, to solve prior art in judge index , it is necessary to managerial experiences higher when system is abnormal, and cannot pinpoint the problems in time, the big problem of time-lag effect.

To solve the above problems, as one aspect of the present invention, there is provided a kind of hospital's Indexes Abnormality pattern is examined automatically Survey method, including：Step 1, obtains the historical data of index system；Step 2, the historical data is split, to obtain Segmentation achievement data；Step 3, extracts the characteristic value of the segmentation achievement data；Step 4, by characteristic value standardization；Step 5, according to standardization after the characteristic value calculate it is described segmentation achievement data exceptional value；Step 6, historical data is according to Historical data obtains exceptional value threshold value；Step 7, abnormal patterns are determined according to the exceptional value and exceptional value threshold value.

Preferably, the step 2 includes：Begun stepping through from the second point of the historical data, if the point is simultaneously than it Front and rear 2 points big are simultaneously smaller than its front and rear 2 points, then using this as the historical data an extreme point；With described The historical data is divided into multistage to obtain being segmented achievement data by extreme point for separation.

Preferably, the step 3 includes：The characteristic value includes height, length, average, standard deviation, wherein height is should Last number of segment data and the difference of the first number, length are that the segment data includes how many data points, average segment data In arithmetic mean of instantaneous value, standard deviation is the standard deviation of the segment data.

Preferably, the step 4 includes：If c₁=<c₁₁,c₁₂, ,c_1p>Be one group of characteristic value, then it is with following formula that the group is special Each characteristic value in value indicative is normalized between 0 to 1：

Wherein, c_maxAnd c_minRespectively c₁=<c₁₁,c₁₂, ,c1_p>In maximum and minimum value.

Preferably, the step 5 includes：

Step 51, be defined on any two points in four dimensional feature spaces apart from computing formula, wherein, four dimensional feature spaces are just It is characteristic value space, each in four dimensional feature spaces is put and includes four dimensions, i.e., four characteristic values defined above, if Point p (x_p,y_p,z_p,k_p), q (x_q,y_q,z_q,k_q) it is four dimensional feature spacesIn any two points, So, the distance of object p, q is：

Step 52, defines the k of any point p in four dimensional feature spaces^thApart from k-dist (p), wherein, the distance is exactly four The distance apart from p in dimensional feature space sorts from small to large, the corresponding point of the big distance of kth；Wherein, in four dimensional feature spaces Point be the points that four characteristic values determine, the k at this is that the distance apart from p in four dimensional feature spaces sorts from small to large, The corresponding point of the big distance of kth, dist represents distance；

Step 53, according to the k of step 52^thThe k average distances of distance definition p, wherein, the k average distances of the p refer to p 1-dist (p), the arithmetic average of 2-dist (p) ..., k-dist (p) distances is designated as k-MD (P)；

Step 54, the k average distances according to p define the k- coefficient of variation, comprise the following steps：

Step 54a, calculates the proper subspace of each characteristic value of four dimensional feature spaces, totally four proper subspaces；

Step 54b, calculates ks average distance k-MD1 (p), k-MD2 (p), k-MD3 of the p in four proper subspaces respectively (p), k-MD4 (p)；

Step 54c, after k-MD (p), k-MD1 (p), k-MD2 (p), k-MD3 (p), k-MD4 (p) are standardized, obtains k- MDO(p)、k-MDO1(p)、k-MDO2(p)、k-MDO3(p)、k-MDO4(p)；

Step 54d, the k- coefficient of variation is calculated according to following formula：

The k- coefficient of variation=k-MDO (p)+max { k-MD1 (p), k-MD2 (p), k-MD3 (p), k-MD4 (p) }.

The present invention can help the abnormal patterns of the users such as hospital automatic detection index system in management, reduce to management The requirement of person's managerial experiences, detects the abnormal patterns of more science, and with instantaneity, detect at any time abnormal.

Brief description of the drawings

Fig. 1 schematically shows flow chart of the invention.

Specific embodiment

Embodiments of the invention are described in detail below in conjunction with accompanying drawing, but the present invention can be defined by the claims Multitude of different ways with covering is implemented.

The present situation of abnormal patterns is detected for hospital, the present invention proposes a kind of automatically exception based on hospital services amount Detection technique, because each index of hospital is the automatic abnormality detection technology of time series data, i.e. time series, the technological sciences Define abnormal patterns, and the method for proposing automatic detection abnormal patterns, realize the function of automatic detection abnormal patterns.

The invention provides a kind of hospital's Indexes Abnormality pattern automatic testing method, for example, being that a kind of hospital services amount refers to Mark abnormal module automatic testing method, comprises the following steps：

Step 1, obtains the history number of index system (Number of Outpatients, amount of being in hospital such as in hospital services figureofmerit system) According to；

Step 2, the historical data is split, to obtain being segmented achievement data.

Step 3, extracts the characteristic value of the segmentation achievement data.Preferably, the characteristic value includes height, length, equal Value, standard deviation.Wherein, be highly the segment data last number and the first number difference, length includes for the segment data How many data points, the arithmetic mean of instantaneous value in average segment data, standard deviation is the standard deviation of the segment data.

Specifically, if segmentation achievement data is X₁=<x_i1,x_i2, ,x_ip>, volume of services achievement data is X=<x₁,x₂, ,x_n>, then can be by following various calculating aforementioned four characteristic values：

Height sph=x (i_p)-x(i₁)；Length spl=i_p-i₁+1；AverageStandard deviation

Step 4, by characteristic value standardization；Specifically, can be standardized by following manner：

If c₁=<c₁₁,c₁₂, ,c_1p>It is one group of characteristic value, then with following formula by each the characteristic value mark in this group of characteristic value Standardization is between 0 to 1：

Wherein, c_maxAnd c_minRespectively c₁=<c₁₁,c₁₂, ,c_1p>In maximum and minimum value.

Step 5, according to standardization after the characteristic value calculate it is described segmentation achievement data exceptional value；

Step 6, historical data obtains exceptional value threshold value according to the historical data.At this time, it may be necessary to mark in the historical data Remember that it is abnormity point which point, the characteristic value of characteristic value and normal point according to known abnormity point relatively draws the threshold of exceptional value Value, wherein on, the principle of threshold value makes a distinction normal point and abnormity point for that can try one's best.

Step 7, abnormal patterns are determined according to the exceptional value and exceptional value threshold value, and the exceptional value of output is more than exceptional value threshold The point of value, as abnormity point.If for example, the outlier threshold of characteristic value Plays difference is 10, the mould is then thought more than the threshold value The standard deviation exception of formula.In one embodiment, first, each segmentation achievement data can calculate an exceptional value, and Abnormal patterns are segmentation achievement data of the exceptional value more than exceptional value threshold value for being segmented achievement data, complete each segmentation index The comparing of data and exceptional value threshold value, the segmentation achievement data section more than exceptional value threshold value is abnormal module.

By adopting the above-described technical solution, the present invention can help the users such as hospital automatic detection index system in management Abnormal patterns, reduce the requirement to manager's managerial experiences, detect the abnormal patterns of more science, and with instant Property, detect at any time abnormal.

More feature ground, the present invention can also can such as be determined using other feature spaces in addition to aforementioned four characteristic value The kurtosis of adopted pattern, degree of bias of pattern etc., and recognize corresponding abnormal patterns；Additionally, can also be using others in the present invention Abnormal value calculating method, can also be applied to other field by the abnormal patterns detection technique.

Preferably, the step 2 includes：Begun stepping through from the second point of the historical data, if the point is simultaneously than it Front and rear 2 points big are simultaneously smaller than its front and rear 2 points, then using this as the historical data an extreme point；With described The historical data is divided into multistage to obtain being segmented achievement data by extreme point for separation.For example, providing n in one piece of data Extreme point, n+1 sections is divided into this n extreme point as end points by the segment data.

Preferably, the step 5 includes：

Step 51, be defined on any two points in four dimensional feature spaces apart from computing formula；Wherein, four dimensional feature spaces are just It is characteristic value space, each in four dimensional feature spaces is put and includes four dimensions, i.e., four characteristic values defined above, if Point p (x_p,y_p,z_p,k_p), q (x_q,y_q,z_q,k_q) it is four dimensional feature spacesIn any two points, So, the distance of object p, q is：

Step 52, defines the k of any point p in four dimensional feature spaces^thApart from k-dist (p), wherein, the distance is exactly four The distance apart from p in dimensional feature space sorts from small to large, the corresponding point of the big distance of kth；Wherein, in four dimensional feature spaces Point be the points that four characteristic values determine, the k at this is that the distance apart from p in four dimensional feature spaces sorts from small to large, The corresponding point of the big distance of kth, dist represents distance；

Step 53, according to the k of step 52^thThe k average distances of distance definition p, wherein, the k average distances of the p refer to p 1-dist (p), the arithmetic average of 2-dist (p) ..., k-dist (p) distances is designated as k-MD (P)；

Step 54, the k average distances according to p define the k- coefficient of variation, comprise the following steps：

Step 54a, calculates the proper subspace of each characteristic value of four dimensional feature spaces, totally four proper subspaces；

Step 54b, calculates ks average distance k-MD1 (p), k-MD2 (p), k-MD3 of the p in four proper subspaces respectively (p), k-MD4 (p)；

Step 54c, after k-MD (p), k-MD1 (p), k-MD2 (p), k-MD3 (p), k-MD4 (p) are standardized, obtains k- MDO(p)、k-MDO1(p)、k-MDO2(p)、k-MDO3(p)、k-MDO4(p)；

Step 54d, the k- coefficient of variation is calculated according to following formula：

The k- coefficient of variation=k-MDO (p)+max { k-MD1 (p), k-MD2 (p), k-MD3 (p), k-MD4 (p) }.

The preferred embodiments of the present invention are the foregoing is only, is not intended to limit the invention, for the skill of this area For art personnel, the present invention can have various modifications and variations.It is all within the spirit and principles in the present invention, made any repair Change, equivalent, improvement etc., should be included within the scope of the present invention.

Claims (5)

1. a kind of hospital's Indexes Abnormality pattern automatic testing method, it is characterised in that including：

Step 1, obtains the historical data of index system；

Step 2, the historical data is split, to obtain being segmented achievement data；

Step 3, extracts the characteristic value of the segmentation achievement data；

Step 4, by characteristic value standardization；

Step 5, according to standardization after the characteristic value calculate it is described segmentation achievement data exceptional value；

Step 6, historical data obtains exceptional value threshold value according to the historical data；

Step 7, abnormal patterns are determined according to the exceptional value and exceptional value threshold value.

2. method according to claim 1, it is characterised in that the step 2 includes：

Begun stepping through from the second point of the historical data, if the point simultaneously than its front and rear 2 points greatly or simultaneously than its front and rear two Point it is small, then using this as the historical data an extreme point；

The historical data is divided into multistage with the extreme point as separation to obtain being segmented achievement data.

3. method according to claim 1, it is characterised in that the step 3 includes：

The characteristic value includes height, length, average, standard deviation, wherein last number and first of height for the segment data The difference of number, length is that the segment data includes how many data points, the arithmetic mean of instantaneous value in average segment data, and standard deviation is should The standard deviation of segment data.

4. method according to claim 1, it is characterised in that the step 4 includes：

If c₁=<c₁₁,c₁₂,,c_1p>It is one group of characteristic value, then each characteristic value in this group of characteristic value is normalized into 0 with following formula To between 1：

$norm\left(c_{1i}\right)=\frac{c_{1i}-c_{\min }}{c_{max}-c_{\min }}$

Wherein, c_maxAnd c_minRespectively c₁=<c₁₁,c₁₂,,c_1p>In maximum and minimum value.

5. method according to claim 3, it is characterised in that the step 5 includes：

Step 51, be defined on any two points in four dimensional feature spaces apart from computing formula, wherein, four dimensional feature spaces are exactly special Value indicative space, in four dimensional feature spaces each point include four dimensions, i.e., four characteristic values defined above, set up an office p (x_p,y_p,z_p,k_p), q (x_q,y_q,z_q,k_q) it is four dimensional feature spacesIn any two points, that , the distance of object p, q is：

Step 52, defines the k of any point p in four dimensional feature spaces^thApart from k-dist (p), wherein, the distance is exactly four-dimensional special The distance apart from p levied in space sorts from small to large, the corresponding point of the big distance of kth；Wherein, the point in four dimensional feature spaces The point that four characteristic values determine is, the k at this is that the distance apart from p in four dimensional feature spaces sorts from small to large, and kth is big The corresponding point of distance, dist represents distance；

Step 53, according to the k of step 52^thThe k average distances of distance definition p, wherein, the k average distances of the p refer to the 1- of p The arithmetic average of dist (p), 2-dist (p) ..., k-dist (p) distance, is designated as k-MD (P)；

Step 54, the k average distances according to p define the k- coefficient of variation, comprise the following steps：

Step 54a, calculates the proper subspace of each characteristic value of four dimensional feature spaces, totally four proper subspaces；

Step 54b, calculates k average distance k-MD1 (p), k-MD2 (p), k-MD3s (p) of the p in four proper subspaces respectively, k-MD4(p)；

Step 54c, after k-MD (p), k-MD1 (p), k-MD2 (p), k-MD3 (p), k-MD4 (p) are standardized, obtains k-MDO (p)、k-MDO1(p)、k-MDO2(p)、k-MDO3(p)、k-MDO4(p)；

Step 54d, the k- coefficient of variation is calculated according to following formula：

The k- coefficient of variation=k-MDO (p)+max { k-MD1 (p), k-MD2 (p), k-MD3 (p), k-MD4 (p) }.