CN108831563A - A kind of decision-making technique detected for differentiating adverse drug reaction Modulation recognition - Google Patents
A kind of decision-making technique detected for differentiating adverse drug reaction Modulation recognition Download PDFInfo
- Publication number
- CN108831563A CN108831563A CN201810275119.8A CN201810275119A CN108831563A CN 108831563 A CN108831563 A CN 108831563A CN 201810275119 A CN201810275119 A CN 201810275119A CN 108831563 A CN108831563 A CN 108831563A
- Authority
- CN
- China
- Prior art keywords
- detection
- decision
- total sample
- data
- adr
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H70/00—ICT specially adapted for the handling or processing of medical references
- G16H70/40—ICT specially adapted for the handling or processing of medical references relating to drugs, e.g. their side effects or intended usage
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H20/00—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance
- G16H20/10—ICT specially adapted for therapies or health-improving plans, e.g. for handling prescriptions, for steering therapy or for monitoring patient compliance relating to drugs or medications, e.g. for ensuring correct administration to patients
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Epidemiology (AREA)
- General Health & Medical Sciences (AREA)
- Public Health (AREA)
- Primary Health Care (AREA)
- Medical Informatics (AREA)
- Data Mining & Analysis (AREA)
- Medicinal Chemistry (AREA)
- Chemical & Material Sciences (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Pharmacology & Pharmacy (AREA)
- Toxicology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Computational Biology (AREA)
- Investigating Or Analysing Biological Materials (AREA)
Abstract
The invention discloses a kind of for differentiating the decision-making technique of adverse drug reaction Modulation recognition detection, the decision-making technique is based on Chinese adverse drug reaction (ADR) data reporting, whether the problem of whether carrying out classification and Detection when studying ADR signal detection, i.e. differentiation Chinese medicine data carry out independent signal detection problem.It devises the discriminant criterion based on java standard library and constructs the decision tree for classification foundation, to compare the otherness of extraction ADR signal from total sample and subsample and decision is carried out with tri- kinds of signal detecting methods of PRR, MHRA and IC respectively, the suggestion for whether carrying out classification and Detection to conceptual data finally provided.The present invention, which provides one kind for Chinese adverse drug reaction signal detection data classification decision, can refer to method.
Description
Technical field
The invention belongs to signal detection fields, and in particular to a kind of for differentiating the detection of adverse drug reaction Modulation recognition
Decision-making technique.
Background technique
At present in the world it is main using including PRR, ROR, MHRA and IC etc. multi-signals detection method to ADR monitoring data
It is detected, but the ADR signal detection technique in China is still not perfect, main problem has：
(1) China not yet establishes the signal detecting method for meeting China's ADR quality of data, using the method for mainstream in the world
Such as：PRR, ROR, MHRA, IC etc..These methods are applied to China ADR data, the consistency of testing result by domestic multidigit scholar
It is poor, a variety of detection methods and deposit and bring difficulty to signal detection；
(2) all drugs, such as Chinese medicine, Western medicine and biological products are contained in China ADR data bank, when signal detection
Do not detected according to classifying drugs.Since data volume is huge, national drug adverse reaction monitoring center carries out a total data
Signal detection take around a large amount of signals for taking a long time, and filtering out also need expert carry out manual analysis,
It was found that the efficiency of signal is relatively low.There has been no the correlative studys that signal detection is carried out after whether can classifying for ADR data at present.
(3) Chinese medicinal ingredients are complicated, have the difference of essence with Western medicine.In China, if by adverse reaction of tcm report into
The detection of row independent signal is always a controversial problem.
Summary of the invention
The technical issues of solution：The present invention solves above-mentioned deficiency in the prior art, from four kinds of ADR to international mainstream
Signal detecting method starting in China's ADR data using the analysis of result, wherein three kinds of methods are used for tactics research for selection,
And the decision tree for being used for discriminant classification is constructed, selected section data are as subsample from total sample, with three kinds of ADR signal detections
Method carries out signal detection to total sample and subsample respectively, finally whether carries out classification and Detection to total sample adverse reaction data
The result of decision is provided, classification and Detection is carried out for ADR data and provides with reference to method.
Technical solution：A kind of decision-making technique detected for differentiating adverse drug reaction Modulation recognition, the decision-making technique
Include the following steps：
1) acquisition and processing of data：Wherein the acquisition of data includes the acquisition of original ADR data and the acquisition of java standard library；
The processing of data be from deleting drug not to be covered and its adverse reaction data in java standard library in the original ADR data, and
Data of the ADR frequency less than 3 are deleted, total sample is obtained, and gross sample is originally separated into subsample and remaining sample；
2) signal detection of data：Signal detection is carried out to total sample and subsample by ADR signal detecting method；
3) validity after judging total sample and subsample before separation using the decision index system based on java standard library；
Using recall ratio R, precision ratio P and variance rate D as decision index system, it is as follows to execute decision process：
301) four fold table based on java standard library is designed：With drug in java standard library-adverse reaction combination to based on the gross sample
The testing result of this and subsample is labeled, if the combination appears in java standard library, is labeled as 1, is otherwise labeled as 0；
Element in the four fold table is respectively a (a1, a0)、b(b1, b0)、c(c1, c0) and d (d1, d0)；Wherein, a is represented
Total sample and subsample detection are the ADR number of combinations of positive signal；B represents total sample detection as positive signal, and subsample is examined
It is out the ADR number of combinations of negative signal；C represents total sample detection as negative signal, and subsample detection is the ADR group of positive signal
Close number；D represents total sample and subsample detection is the ADR number of combinations of negative signal；a1、b1、c1And d1Respectively indicate a, b, c
With the ADR number of combinations appeared in d in java standard library；a0、b0、c0And d0It respectively indicates in a, b, c and d and does not appear in java standard library
ADR number of combinations；And a=a1+a0, b=b1+b0, c=c1+c0, d=d1+d0；
302) recall ratio compares：If total sample and subsample recall ratio are respectively R1And R2,
If R1> R2, and R1And R2Difference be greater than the first preset threshold value, then without carrying out classification inspection to total sample
It surveys；If R2> R1, and R1And R2Difference be greater than the first preset threshold value, then need to total sample carry out classification and Detection；If R1
And R2Difference less than the first preset threshold value, then carry out step 303)；
303) precision ratio compares：If total sample and subsample precision ratio are respectively P1And P2,
If P1> P2, and P1And P2Difference be greater than the first preset threshold value, then without carrying out classification inspection to total sample
It surveys；If P2> P1, and P1And P2Difference be greater than the first preset threshold value, then need to total sample carry out classification and Detection；If P1
And P2Difference less than the first preset threshold value, then carry out step 304)；
304) variance rate compares：If the variance rate is D,
If D is greater than the second preset threshold value, without carrying out classification and Detection to total sample；It otherwise is needs to total sample
Carry out classification and Detection.
Preferably, the original ADR data are obtained from national drug adverse reaction monitoring center；The java standard library is to pass through
Network acquires each phase adverse drug reaction communication of the specification of related drugs, State Food and Drug Administration's publication
And library known to pharmacovigilance news flash and the ADR of various regulation files foundation.
Preferably, the decision-making technique further includes the verifying of class test result, is " to need to total sample in the result of decision
Carry out classification and Detection " in the case where, the remaining sample and total sample are tested, if decision index system R, P value are less than the
One preset threshold value and when the value of D is less than the second preset threshold value, then receive " to need to carry out classification inspection to total sample
Survey " the result for " needing to carry out classification and Detection to total sample " as a result, otherwise do not receive.
Preferably, the gross sample is originally divided into the total data comprising three Chinese medicine, Western medicine and biological products classifications.
Preferably, the subsample is the data of Chinese medicine classification；The residue sample is Western medicine and biological products classification
Data.
Preferably, ADR signal detecting method is PRR, MHRA, ROR based on asymmetric measuring principle in the step 2)
With IC method.
Preferably, first preset threshold value is 2%；Second preset threshold value is 10%.
Beneficial effect：1. whether the present invention based on Chinese ADR data reporting, studies ADR signal detection process to total
Sample data carries out the problem of classification and Detection, devises the Testing index based on java standard library, constructs determining for classification and Detection
Plan tree.
2. the present invention is extracted from total sample and subsample by tri- kinds of signal detecting methods of PRR, MHRA, IC to compare
The otherness of ADR signal determines most suitable ADR signal detecting method, to improve the timeliness and validity of signal detection.
3. the present invention, which provides one kind for Chinese adverse drug reaction signal detection data classification decision, can refer to method.
Detailed description of the invention
Fig. 1 is decision flow diagram；
Fig. 2 is the decision tree for discriminating whether classification；
In figure, R1、R2Respectively represent the recall ratio of total sample and subsample, P1、P2Respectively represent total sample and subsample
Precision ratio, b1Indicate that total sample detection is positive signal and subsample detection is the ADR number of combinations of negative signal, c1Indicate increment
This detection is positive signal and total sample detection is the number of combinations of negative signal, b1、c1For known signal.
Specific embodiment
The following examples can make those skilled in the art that the present invention be more fully understood, but not limit this in any way
Invention.
Embodiment 1
The acquisition of 1ADR monitoring data and processing
1.1 data processing
(1) initial data summarizes：2 years ADR data reporting records are obtained altogether from national drug adverse reaction monitoring center
1823144, wherein multiple adverse reaction relationships are corresponded to the presence of a kind of 608710 (accounting for 33.4%) drugs of record, it will be this kind of
Data split into one-to-one relationship.Delete nomenclature of drug or entitled " unknown " record of adverse reaction in data.By place
The data amount obtained after reason 2221942 records, drug classification include " Chinese medicine ", " Western medicine " and " biological products ", wherein " in
The data volume of medicine " classification is 317417, accounts for 14.29%；The data volume of " Western medicine " classification is 1874904, accounts for 84.38%；" biology
The data volume of product " classification is 29621, accounts for 1.33%.According to adopted drug name and adverse drug reaction title to overall number
According to being summarized, the frequency that 139281 drugs-adverse reactions combination and its occurred is obtained, remembers that this data acquisition system is Datal.
The data set contains 6174 kinds of drugs and 2458 kinds of adverse reactions.
(2) java standard library is established：In order to judge the validity of data classification testing result, need to establish one with known bad
Reaction normal library is as reference.Specification, the State Food and Drug Administration's publication of related drugs are acquired by network
Each phase adverse drug reaction communication and pharmacovigilance news flash and various regulation files etc. establish library known to ADR, claim
Be java standard library.In java standard library, altogether comprising 53774 drugs-adverse reaction combination, remember that this data acquisition system is Data2.The number
2401 kinds of drugs and 2460 kinds of adverse reactions are contained according to collection.
(3) in order to keep the consistency with drug in java standard library, do not include in deletion java standard library Data2 in Data1
Drug obtains data set Data3.Due to the ADR frequency be 1,2 when be accidentalia probability it is larger and in signal screening it is usual
It is required that the frequency is greater than equal to 3, therefore the data by the ADR frequency less than 3 are deleted, and obtained data set contains 39782 notes
Record, shares 1692 kinds of drugs and 877 kinds of adverse reactions.Remember that this data set is total sample.
(4) " Chinese medicine " data separating is obtained into data set Data4 from total sample.The data set contains 4697 notes
Record, shares 326 kinds of drugs and 283 kinds of adverse reactions.Remember that this data set is subsample, which is the proper subclass of total sample.
(5) drug in two sample sets-adverse reaction combination is labeled with java standard library：It appears in java standard library
Combination is labeled as " 1 ", is otherwise labeled as " 0 ".The field of two datasets includes：Drug classification, adopted name, adverse reaction name
Claim, the frequency and whether known etc..
1.2 two sample data statistical analysis
It is related to 1692 kinds of drug kinds comprising 1,972,008 ADR reports altogether in total sample, 39782 drugs-are bad anti-
It should combine with 877 kinds of adverse reactions, the quantity that adverse reaction occurs for the every kind of drug that is averaged is 1,165.49, average each ADR group
Closing the frequency occurred is 49.57, and the frequency that average every kind of adverse reaction occurs is 2248.58.
Comprising 199115 ADR reports in subsample, the 10.1% of total sample is accounted for；It is related to 326 kinds of drugs, 4697 medicines
Product-adverse reaction combination and 283 kinds of adverse reactions, the quantity that adverse reaction occurs for average every kind of drug is 610.78, than totality
Sample reduces 554.71, and the frequency that average each ADR combination occurs is 42.39, reduces 7.18 than population sample, average every kind is not
The frequency that good reaction occurs is 703.59, reduces about 1545 than population sample.Statistics indicate that ADR occurs in subsample (Chinese medicine)
Average level to be significantly lower than total sample.
It is compareed with java standard library, the known drug-adverse reaction number of combinations for including in total sample is 13555, ratio
It is 34.07%；Known drug-adverse reaction number of combinations in subsample is 830, and ratio 17.67% is fewer than total sample
16.4%, i.e., known drug-adverse reaction combination will be far below total sample in Chinese medicine data.
In terms of the quality of data, in total sample altogether comprising serious adverse reaction number of reports be 88083, accounting 4.47%,
Serious adverse reaction number of reports is 10007 in subsample, accounts for 5.03%, therefore subsample has relatively high data than total sample
Quality.
Total frequency occupies preceding ten adverse reaction in 1. two samples of table
| Adverse reaction | Total sample report number (ratio) | Adverse reaction | Subsample number of reports (ratio) |
| Fash | 281399 (14.27%) | Fash | 32161 (16.15%) |
| Nausea | 226368 (11.48%) | Itch | 22736 (11.42%) |
| Itch | 173111 (8.78%) | Nausea | 14339 (7.20%) |
| Vomiting | 141098 (7.16%) | It is dizzy | 9577 (4.81%) |
| It is dizzy | 83278 (4.22%) | Shiver with cold | 9206 (4.62%) |
| Headache | 56507 (2.87%) | Palpitaition | 8728 (4.38%) |
| Abdominal pain | 52291 (2.65%) | Vomiting | 8661 (4.35%) |
| Diarrhea | 49839 (2.53%) | Anaphylactoid reaction | 7182 (3.61%) |
| Anaphylactoid reaction | 49261 (2.50%) | It is uncomfortable in chest | 6251 (3.14%) |
| Shiver with cold | 39066 (1.98%) | Fever | 5847 (2.94%) |
| It is total | 1152218 (58.44%) | 124688 (62.62%) |
Table 1 is the adverse reaction situation that before the frequency occupies ten occur in two samples, and the sum of proportion is more than
60% or so of each sample size.The adverse reaction title that the two arranges preceding ten has 7 kinds of ADR identical, the difference is that having in total sample
" headache ", " abdominal pain " and " diarrhea ", and have " palpitaition ", " uncomfortable in chest " and " fever " in subsample.Arrange the adverse reaction of front three
Be both " fash ", " nausea " and " itch ", but the ratio of " nausea " in total sample is higher than subsample by 4.28%, and
" itch " then lower than subsample 2.64%.The ratio of sample size shared by each adverse reaction difference, maximum difference is " to dislike
The heart ", subsample fewer than total sample 3.36%.
The above-mentioned analysis to two sample datas, which is compared, to be found out, the two is believed in having a certain difference property of many aspects
It can't determine when number detection with the presence or absence of otherness, need further comparative studies.
3 detection method applied analyses and selection
3.1 signal detecting method
Currently, the ADR signal detecting method in China uses four kinds of main stream approach based on asymmetric measuring principle：PRR,
MHRA, ROR and IC, their calculation formula are all based on what following classical four fold table carried out.
The classical four fold table of table 2.
(1) PRR method
95%CI=eln(PRR)±196SE(ln PRR)Formula 3
The critical value for generating signal is 95%CI lower limit > 1, i.e.,：
(2) ROR method
Report odds ratio method (reporting odds ratio, ROR) by Dutch pharmacovigilance center (Lareb) laboratory
It proposes first, its calculation formula is：
95%CI=eln(ROR)±1.96SE(ln ROR)Formula 7
The critical value for generating signal is 95%CI lower limit > 1, i.e.,：
(3) MHRA method
MHRA is British Drug and health product management board (Medicines and Healthcare products
Regulatory Agency, MHRA) use comprehensive standard method, that is, combine PRR value, absolute number of reports and Pearson x2Value is come
The strength of association of assessment signal, referred to as MHRA method, the critical value of signal judgement：PRR >=2, A >=3, x2≥4。
(4) IC method
2002 Uppsala monitoring center of the World Health Organization (UMC) Bate etc. establish a set of new adverse drug
Reaction signal detection method, the referred to as progressive neural network model in Bayesian Decision credibility interval (Bayesian Confidence
Propagation Neural Network, BCPNN).Since the process employs the informational contents in informatics
(information component, IC), so being referred to as IC method at present.Its calculation formula is as follows：
N=a+b+c+d formula 9
R=(N+2)2/ ((a+b+1) (a+c+1)) formula 10
E=log2((a+1) r/ (N+r)) formula 11
v1=(N-a+r-1)/((a+1) (1+N+r)) formula 12
v2=(N-a-b+1)/((a+b+1) (3+N)) formula 13
v3=(N-a-c+1)/((a+c+1) (3+N)) formula 14
V=(log102)-2(v1+v2+v3) formula 15
Signal judgment criteria is：IC > 0.
The selection of 3.2 signal detecting methods
It is respectively applied to total sample and subsample using the signal detecting method of above-mentioned four kinds of mainstreams, finds positive signal
As a result it is compared as follows：
The semaphore that each method detects in 3. two samples of table
From table 3 it is observed that it is consistent that each detection method, which applies the generated number of signals relationship on two samples,
's：Relationship of successively decreasing is presented by PRR, ROR, MHRA, IC.Positive signal is indicated with " 1 ", and " 0 " indicates non-positive signal, four kinds of methods
Testing result related coefficient is shown in Table 4.
Related coefficient in the total sample of table 4. between each detection method
Related coefficient in 5. subsample of table between each detection method
By table 4 and table 5 it is found that the related coefficient of PRR method and ROR method is close to 1, i.e. the result of the two is almost the same；
The related coefficient of MHRA and PRR, ROR are more than 82%；Related coefficient between IC and other three kinds of methods is lower.It simultaneously can also be with
Find out, compared with total sample, the related coefficient of the several method result of subsample is declined, and especially IC method decline is close
8%.Therefore, the detection method whether the present patent application selects tri- kinds of methods of PRR, MHRA and IC to classify for data.
The design of 4 decision-making techniques
The design of 4.1 decision processes
The superiority and inferiority for judging the signal detecting result of two samples before and after data separating, needs to solve the problems, such as following：
What the process of decision is？Signal detecting method applies how the validity on two samples before and after data separating is evaluated？Into
Which index is row decision need？By Chinese medicine data after separating in total sample, carrying out signal detecting result to remaining data can be produced
Which type of raw influence？The flow chart of a decision is provided first, as shown in Figure 1.
Chinese medicine sample is isolated from total sample as subsample, remaining data is to utilize detection as the sample of verifying
The method provided in method applied analysis and selection carries out signal detection to total sample and subsample respectively, respectively obtains two knots
Fruit set constructs decision-making technique, judges the superiority and inferiority of two result sets respectively, obtain decision conclusions.Using remaining sample to result
It is verified, judges whether acceptance decision conclusion.
4.2 four fold tables of the design based on java standard library
Java standard library provides objective foundation for categorised decision.In order to compare the validity for separating former and later two data sets,
Our selection criteria libraries are as the standard examined.With drug in java standard library-adverse reaction combination to based on total sample and subsample
Testing result be labeled, if the combination appears in java standard library, be labeled as " 1 ", be otherwise labeled as " 0 ".Utilize two
A sample carries out signal detection respectively and is compared with java standard library, may make up following four fold table, is shown in Table 6.
Two pattern detection results of the table 6. based on java standard library compare four fold table
| Subsample+ | Subsample- | |
| Total sample+ | a(a1, a0) | b(b1, b0) |
| Total sample- | c(c1, c0) | d(d1, d0) |
In table 6, a represents total sample and subsample detection is the ADR number of combinations of positive signal；B represents total sample inspection
It is out positive signal, subsample detection is the ADR number of combinations of negative signal；C represents total sample detection as negative signal, subsample
Detection is the ADR number of combinations of positive signal；D represents total sample and subsample detection is the ADR number of combinations of negative signal；a1、
b1、c1And d1The ADR number of combinations appeared in java standard library in a, b, c and d is respectively indicated, it can be by b1Regard missing inspection in subsample as
The signal number of java standard library, by c1Regard the signal number of the java standard library of missing inspection in total sample, b as1、c1In combination be two samples
The difference signal of detection can be used as a foundation of categorised decision；a0、b0、c0And d0It respectively indicates in a, b, c and d and does not appear in
ADR number of combinations in java standard library；And a=a1+a0, b=b1+b0, c=c1+c0, d=d1+d0。
The design of 4.3 decision index systems
(1) decision index system R- recall ratio
Recall ratio refers to known signal (the i.e. a from java standard library1+b1+c1+d1) in detect the ratio of signal, be a kind of right
The measurement of known signal coverage rate.With recall ratio R1The ability that total sample detects known signal is described, as shown in formula 17：
With recall ratio R2Description subsample detects the ability of known signal, as shown in formula 18：
The ability that the two detects known ADR signal can be distinguished by the comparison of the recall ratio to two samples, is also embodied
The difference of the sensitivity of the detecting signal of two samples under conditions of on the basis of java standard library.Therefore recall ratio is a pass
Key index should be used as the primary foundation of categorised decision.
(2) decision index system P- precision ratio
The defect of formula 17 and formula 18 is to work as a1Much larger than b1And c1When, even if b1With c1Differ larger, but the two
Otherness is not significant.Therefore, in R1With R2When being not much different, it is also necessary to define precision ratio.Precision ratio refer to based on some sample into
Ratio shared by known signal in the result that row signal detection obtains is a kind of measurement to known signal Detection accuracy.With
Precision ratio P1The ability that total sample detects known signal is described, is indicated with formula 19：
With precision ratio P2Description subsample detects the ratio of known signal, is indicated with formula 20：
(3) decision index system D- variance rate
Variance rate represents the Diversity measure of two pattern detection results, is indicated with formula 21：
The design of 4.4 decision trees
Using tri- indexs of total sample and R, P, D of subsample, decision is carried out to whether Chinese medicine data carry out classification and Detection,
Implementation procedure is as follows：
1) the primary foundation of judgement is to consider to carry out covering of the result of signal detection to known signal based on two class samples
Rate, i.e., the signal recall ratio of two samples (see formula 17 and 18).Select recall ratio high to carry out the sample of signal detection, such as
Total sample recall ratio R1Height, then without carrying out classification and Detection to total sample；If subsample recall ratio R2Height is then needed to gross sample
This progress classification and Detection；If be not much different, further decision (given threshold 2%) is needed.
2) when two sample recall ratios are not much different, it need to consider the precision of signal detection, i.e. the signal precision ratio of the two
(see formula 19 and 20).If population sample precision ratio P1Height, then without carrying out classification and Detection to total sample；If subsample is looked into
Quasi- rate P2Height then needs to carry out classification and Detection to total sample；If be not much different, needing further decision, (given threshold is
2%).
3) when two sample recall ratios and precision ratio are not much different (i.e. the difference of the two is within 2%), then directly compare
Compared with the variance rate of the two, if variance rate is greater than given threshold value (given threshold 10%), the result of decision is without to total
Sample carries out classification results；Otherwise result is to need to carry out classification and Detection to total sample.
4) in the case where the result of decision is to need to carry out classification and Detection to total sample, with the remaining data after separation Chinese medicine
(including Western medicine and biological products) are tested on above-mentioned indices with total sample, if test result is in decision index system public affairs
In 19,20 and 21 3 indexs of formula within the acceptable range, then receive to need to carry out total sample classification and Detection as a result,
Otherwise do not receive the result for needing to carry out total sample classification and Detection.According to the above analysis, whether building classifies inspection for data
The decision tree of survey such as Fig. 2.
5 classification and Detection interpretations of result and decision
The experimental procedure of categorised decision is as follows：
(1) each drug in total sample and subsample-adverse reaction combination b, c, d (being shown in Table lattice 3) are calculated separately, and is divided
Not Ji Suan three kinds of detection methods value；
It (2) will not be that the data of " Chinese medicine " remove to form remaining data in total sample, as a result for the ease of comparing
Verify sample.Treated, and always sample and subsample contain drug-adverse reaction combination of 4697 Chinese medicines and have one by one
Corresponding relationship；
(3) signal is extracted from total sample and subsample using three kinds of detection method signal judgment criterias and carry out with java standard library
Compare；
(4) categorised decision is carried out using the detection method in the 4th part.
Total sample and subsample are tested respectively using ADR signal detecting method PRR, MHRA and IC, the results are shown in Table 7
It is shown：
7. 3 kinds of detection method testing result tables of comparisons of table
Table 8. differentiates the decision-making foundation table of Chinese medicine data classification detection
Using in table 7 data and categorised decision tree establish decision-making foundation table (table 8).According to the decision tree constructed in 4.3,
Decision process is as follows：
(1) PRR method：Due to R1-R2=0.72% < 2%, P1-P2=0.16% < 2% and D=7.16% < 10%,
Differentiate according to decision tree, conclusion is " needing to carry out classification and Detection to total sample ".
(2) MHRA method：Due to R1-R2=0.16% < 2%, P1-P2<=0.972% and D=3.26% < 10%,
Differentiate according to decision tree, conclusion is " needing to carry out classification and Detection to total sample ".
(3) IC method：Due to R1-R2=1.44% < 2%, P2-P1<=4.06% differentiates that conclusion is according to decision tree
" needing to carry out classification and Detection to total sample ".
The decision path of PRR, MHRA are consistent, and IC method be since the recall ratio of two samples is very close, and
The precision ratio of Chinese medicine data is significantly better than total sample, and therefore, conclusion is also " needing to carry out classification and Detection to total sample ".Due to three
The result of decision of kind method is to need to carry out classification and Detection to total sample, it is therefore desirable to remaining data (Western medicine and biological system
Product) above-mentioned test is carried out in the same way, each achievement data of detection is as follows：
The decision-making foundation table of the verifying remaining data classification and Detection of table 9.
As can be seen from Table 9, when being verified using remaining sample, three kinds of detection methods are equal in the difference of the index of R, P, D
Within defined threshold range, that is to say, that by Chinese medicine data after being separated in population sample, examined to the signal of remaining data
It is smaller to survey influence, therefore, the result of decision is that " needing to carry out classification and Detection to total sample " is acceptable.
Claims (7)
1. a kind of for differentiating the decision-making technique of adverse drug reaction Modulation recognition detection, which is characterized in that the decision-making technique
Include the following steps：
1) acquisition and processing of data：Wherein the acquisition of data includes the acquisition of original ADR data and the acquisition of java standard library；Data
Processing be and to delete from deleting drug not to be covered and its adverse reaction data in java standard library in the original ADR data
Data of the ADR frequency less than 3 obtain total sample, and gross sample is originally separated into subsample and remaining sample；
2) signal detection of data：Signal detection is carried out to total sample and subsample by ADR signal detecting method；
3) validity after judging total sample and subsample before separation using the decision index system based on java standard library；Using
It is as follows to execute decision process as decision index system by recall ratio R, precision ratio P and variance rate D：
301) four fold table based on java standard library is designed：With the combination of drug in java standard library-adverse reaction to based on total sample and
The testing result of subsample is labeled, if the combination appears in java standard library, is labeled as 1, is otherwise labeled as 0；
Element in the four fold table is respectively a (a1, a0)、b(b1, b0)、c(c1, c0) and d (d1, d0)；Wherein, a represents gross sample
It originally is the ADR number of combinations of positive signal with subsample detection；B represents total sample detection as positive signal, and subsample detection is
The ADR number of combinations of negative signal；C represents total sample detection as negative signal, and subsample detection is that the ADR of positive signal is combined
Number；D represents total sample and subsample detection is the ADR number of combinations of negative signal；a1、b1、c1And d1Respectively indicate a, b, c and d
In appear in ADR number of combinations in java standard library；a0、b0、c0And d0It respectively indicates in a, b, c and d and does not appear in java standard library
ADR number of combinations；And a=a1+a0, b=b1+b0, c=c1+c0, d=d1+d0；
302) recall ratio compares：If total sample and subsample recall ratio are respectively R1And R2,
If R1> R2, and R1And R2Difference be greater than the first preset threshold value, then without to total sample carry out classification and Detection；If R2
> R1, and R1And R2Difference be greater than the first preset threshold value, then need to total sample carry out classification and Detection；If R1And R2Difference
Value then carries out step 303) less than the first preset threshold value；
303) precision ratio compares：If total sample and subsample precision ratio are respectively P1And P2,
If P1> P2, and P1And P2Difference be greater than the first preset threshold value, then without to total sample carry out classification and Detection；If P2
>P1, and P1And P2Difference be greater than the first preset threshold value, then need to total sample carry out classification and Detection；If P1And P2Difference
Value then carries out step 304) less than the first preset threshold value；
304) variance rate compares：If the variance rate is D,
If D is greater than the second preset threshold value, without carrying out classification and Detection to total sample；Otherwise total sample is carried out for needs
Classification and Detection.
2. it is according to claim 1 a kind of for differentiating the decision-making technique of adverse drug reaction Modulation recognition detection, it is special
Sign is：The original ADR data are obtained from national drug adverse reaction monitoring center；The java standard library is acquired by network
Each phase adverse drug reaction communication and drug that the specification of related drugs, State Food and Drug Administration issue
Guard against library known to the ADR that news flash and various regulation files are established.
3. it is according to claim 1 a kind of for differentiating the decision-making technique of adverse drug reaction Modulation recognition detection, it is special
Sign is：The decision-making technique further includes the verifying of class test result, is " to need to classify to total sample in the result of decision
In the case where detection ", the remaining sample and total sample are tested, if decision index system R, P value are set in advance less than first
When determining the value of threshold value and D less than the second preset threshold value, then receive the knot of " needing to carry out classification and Detection to total sample "
Otherwise fruit does not receive the result of " needing to carry out classification and Detection to total sample ".
4. it is according to claim 1 a kind of for differentiating the decision-making technique of adverse drug reaction Modulation recognition detection, it is special
Sign is：The gross sample is originally divided into the total data comprising three Chinese medicine, Western medicine and biological products classifications.
5. it is according to claim 1 a kind of for differentiating the decision-making technique of adverse drug reaction Modulation recognition detection, it is special
Sign is：The subsample is the data of Chinese medicine classification；The residue sample is the data of Western medicine and biological products classification.
6. it is according to claim 1 a kind of for differentiating the decision-making technique of adverse drug reaction Modulation recognition detection, it is special
Sign is：ADR signal detecting method is PRR, MHRA, ROR and IC method based on asymmetric measuring principle in the step 2).
7. it is according to claim 1 a kind of for differentiating the decision-making technique of adverse drug reaction Modulation recognition detection, it is special
Sign is：First preset threshold value is 2%；Second preset threshold value is 10%.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810275119.8A CN108831563B (en) | 2018-03-29 | 2018-03-29 | Decision method for distinguishing classification detection of adverse drug reaction signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810275119.8A CN108831563B (en) | 2018-03-29 | 2018-03-29 | Decision method for distinguishing classification detection of adverse drug reaction signals |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108831563A true CN108831563A (en) | 2018-11-16 |
| CN108831563B CN108831563B (en) | 2021-09-28 |
Family
ID=64154308
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810275119.8A Active CN108831563B (en) | 2018-03-29 | 2018-03-29 | Decision method for distinguishing classification detection of adverse drug reaction signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108831563B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110767319A (en) * | 2019-09-30 | 2020-02-07 | 南京邮电大学 | Method for detecting adverse reaction signals of combined medication |
| CN110879822A (en) * | 2019-11-15 | 2020-03-13 | 南京邮电大学 | Drug adverse reaction signal detection method based on association rule analysis |
| CN112133450A (en) * | 2019-10-15 | 2020-12-25 | 南京邮电大学 | Method for eliminating adverse drug reaction data shielding effect based on decision tree layering |
| WO2023124802A1 (en) * | 2021-12-29 | 2023-07-06 | 上海太美数字科技有限公司 | Pharmacovigilance signal generation method and apparatus, device, and medium |
| TWI812056B (en) * | 2022-03-10 | 2023-08-11 | 宏碁股份有限公司 | Method and electronic device of checking drug interaction |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105760698A (en) * | 2016-03-18 | 2016-07-13 | 华中科技大学同济医学院附属协和医院 | Adverse drug reaction early warning and analyzing system and method |
-
2018
- 2018-03-29 CN CN201810275119.8A patent/CN108831563B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105760698A (en) * | 2016-03-18 | 2016-07-13 | 华中科技大学同济医学院附属协和医院 | Adverse drug reaction early warning and analyzing system and method |
Non-Patent Citations (2)
| Title |
|---|
| 候永芳等: "信号检测在药品不良反应监测***中的应用", 《中国药物警戒》 * |
| 陈友生等: "常用药品不良反应信号检测方法研究进展", 《中国药物依赖性杂志》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110767319A (en) * | 2019-09-30 | 2020-02-07 | 南京邮电大学 | Method for detecting adverse reaction signals of combined medication |
| CN112133450A (en) * | 2019-10-15 | 2020-12-25 | 南京邮电大学 | Method for eliminating adverse drug reaction data shielding effect based on decision tree layering |
| CN110879822A (en) * | 2019-11-15 | 2020-03-13 | 南京邮电大学 | Drug adverse reaction signal detection method based on association rule analysis |
| CN110879822B (en) * | 2019-11-15 | 2022-11-15 | 南京邮电大学 | Drug adverse reaction signal detection method based on association rule analysis |
| WO2023124802A1 (en) * | 2021-12-29 | 2023-07-06 | 上海太美数字科技有限公司 | Pharmacovigilance signal generation method and apparatus, device, and medium |
| TWI812056B (en) * | 2022-03-10 | 2023-08-11 | 宏碁股份有限公司 | Method and electronic device of checking drug interaction |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108831563B (en) | 2021-09-28 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108831563A (en) | A kind of decision-making technique detected for differentiating adverse drug reaction Modulation recognition | |
| CN104970789B (en) | Electrocardiogram sorting technique and system | |
| CN106248559B (en) | A kind of five sorting technique of leucocyte based on deep learning | |
| CN108833409A (en) | webshell detection method and device based on deep learning and semi-supervised learning | |
| US20140221247A1 (en) | System, method, and article for detecting abnormal cells using multi-dimensional analysis | |
| CN110023759A (en) | For using system, method and the product of multidimensional analysis detection abnormal cell | |
| CN109872776A (en) | A kind of screening technique and its application based on weighted gene coexpression network analysis to gastric cancer potential source biomolecule marker | |
| CN106228190A (en) | Decision tree method of discrimination for resident's exception water | |
| CN110322930A (en) | Metabolism group operator logo object recognition methods based on horizontal relationship | |
| CN110532298A (en) | More attribute railway accident reason weight analysis methods | |
| Bonham-Carter | A Numerical Method of Classification Using Qualitative and Semi-Quantative Data, as Applied to the Facies Analysis of Limestones | |
| CN113749619A (en) | Mental fatigue assessment method based on K-TRCA | |
| CN111986819B (en) | Adverse drug reaction monitoring method and device, electronic equipment and readable storage medium | |
| CN113539473A (en) | Method and system for diagnosing brucellosis only by using blood routine test data | |
| CN109033747A (en) | It is a kind of to disturb integrator gene selection and the recognition methods of tomour specific gene subset based on PLS more | |
| CN112599250A (en) | Postoperative data analysis method and device based on deep neural network | |
| Charrad et al. | NbClust Package. An examination of indices for determining the number of clusters | |
| WO2021011698A1 (en) | Artificial intelligence for early cancer detection | |
| WO2020201443A1 (en) | Cytometry data analysis | |
| CN107122394A (en) | Abnormal deviation data examination method and device | |
| CN108804494A (en) | A kind of method that data shadowing effect is minimized in adverse drug reaction signal detection | |
| CN109146232A (en) | Stock market's investor sentiment of principal component and the coefficient of variation acts on behalf of index selection method | |
| CN110689961B (en) | Gastric cancer disease risk detection device based on big data analysis technology | |
| WO2022056478A2 (en) | Automated classification of immunophenotypes represented in flow cytometry data | |
| CN114169901A (en) | Medical insurance abnormity detection method and system based on behavior sequence classification |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |