Introduction to standards:
GB/T 27415-2013 Evaluation of detection limit and quantification limit of analytical methods
GB/T27415-2013
Standard compression package decompression password: www.bzxz.net
This standard specifies the detection limit and quantification limit evaluation procedures when the relative standard deviation (RSD) is Z%.
This standard is applicable to the evaluation of detection limit and quantification limit between laboratories or within laboratories when calibration errors are ignored.
This standard was drafted in accordance with the rules given in GB/T1.1-2009.
This standard was proposed and managed by the National Technical Committee for Certification and Accreditation of Standardization (SAC/TC261).
The drafting units of this standard are: Liaoning Entry-Exit Inspection and Quarantine Bureau, Shandong Entry-Exit Inspection and Quarantine Bureau, Beijing University of Technology, China National Accreditation Service for Conformity Assessment, China Center for Disease Control and Prevention, Dehong Prefecture Quality and Technical Supervision Comprehensive Testing Center, Guangdong Entry-Exit Inspection and Quarantine Bureau, Shanxi Entry-Exit Inspection and Quarantine Bureau, Shanghai Entry-Exit Inspection and Quarantine Bureau, China National Institute of Metrology, National Dangerous Chemicals Quality Supervision and Inspection Center, National Electrical Safety Quality Supervision and Inspection Center.
The main drafters of this standard are: Wang Douwen, Chen Shishan, Xie Tianfa, Niu Xingrong, Sun Hairong, Yang Jiaolan, Shi Changyan, Deng Yun, Zhong Zhiguang, Zhao Fabao, Chen Junshui, Wang Jing, Sun Xingquan, Wang Dong, Ji Hongtao, Wang Ni.
The following documents are indispensable for the application of this document. For all dated references, only the dated version applies to this document. For all undated references, the latest version (including all amendments) applies to this document.
GB/T3359 Statistical processing and interpretation of data Determination of statistical tolerance intervals
GB/T22554 Linear calibration based on standard samples
GB/T27407 Laboratory quality control Evaluation and analysis of the performance of measurement systems using statistical quality assurance and control chart techniques
JJF1001 General metrological terms and definitions
Some standard content:
National Standard of the People's Republic of China
GB/T27415-—2013
Estimate of detection and quantitation limit for analytical methods method2013-09-06 Issued
General Administration of Quality Supervision, Inspection and Quarantine of the People's Republic of China Standardization Administration of China
2013-12-01 Implementation
This standard was drafted in accordance with the rules given in GB/T1.1-2009
This standard was proposed and managed by the National Technical Committee for Certification and Accreditation (SAC/TC261). GB/T27415-2013
The drafting units of this standard are: Liaoning Entry-Exit Inspection and Quarantine Bureau, Shandong Entry-Exit Inspection and Quarantine Bureau, Beijing University of Technology, China National Accreditation Service for Conformity Assessment, China Center for Disease Control and Prevention, Dehong Prefecture Quality and Technical Supervision Comprehensive Testing Center, Guangdong Entry-Exit Inspection and Quarantine Bureau, Shanxi Entry-Exit Inspection and Quarantine Bureau, Shanghai Entry-Exit Inspection and Quarantine Bureau, China National Institute of Metrology, National Dangerous Chemicals Quality Supervision and Inspection Center, National Electrical Appliance Safety Quality Supervision and Inspection Center. The main drafters of this standard: Wang Douwen, Chen Shishan, Xie Tianfa, Niu Xingrong, Sun Hairong, Yang Jiaolan, Shi Changyan, Deng Yun, Zhong Zhiguang, Zhao Fabao, Chen Junshui, Wang Jing, Sun Xingquan, Wang Dong, Ji Hongtao, Wang Ni. GB/T27415—2013
The evaluation procedure of this standard adopts the "mathematical model method", which is different from the commonly used "single-point calibration method". The "single-point calibration method" and the "mathematical model method" are the main means of evaluating the detection limit and quantification limit at present. The differences between the two are as follows a) Standard deviation processing
The "single-point calibration method" regards the standard deviation value as a constant; the "mathematical model method" believes that the standard deviation value varies with concentration and uses multi-point experimental data for fitting. b)
Two types of error rate (and β) control
“Single point calibration method” only gives the level of ; “Mathematical model method” considers the levels of and β at the same time. Fitting method
“Single point calibration method” uses ordinary least squares fitting (OLS); “Mathematical model method” uses weighted least squares fitting (WLS). Model test
“Single point calibration method” does not perform statistical tests on the model fit; “Mathematical model method” performs significance tests on the model fit. Bias correction
The "single-point calibration method" does not consider bias correction: "Mathematical model method\Establish bias correction regression model. Interval calculation
The "single-point calibration method" is calculated according to the "confidence interval"; the "mathematical model method" is calculated according to the "statistical tolerance interval". g)
Definition scope
The "single-point calibration method" is only applicable to intra-laboratory research. The "mathematical model method" is applicable to both inter-laboratory and intra-laboratory research. 1 Scope
Evaluation of detection limit and quantification limit of analytical method This standard specifies the evaluation procedure of detection limit and quantification limit when the relative standard deviation (RSD) is Z%. GB/T27415-2013
This standard is applicable to the evaluation of detection limit and quantification limit between laboratories or within laboratories when calibration errors are ignored. 0
Normative references||tt| |The following documents are indispensable for the application of this document. For any dated referenced document, only the dated version applies to this document. For any undated referenced document, its latest version (including all amendments) applies to this document. GB/T3359 Statistical processing and interpretation of data Determination of statistical tolerance intervals GB/T22554 Laboratory quality control for linear calibration based on standard samples Evaluation and analysis of the performance of measurement systems using statistical quality assurance and control chart techniques GB/T27407
JJF1001 General metrological terms and definitions
Terms and definitions
The terms, definitions and symbols defined in GB/T3359 and JJF1001 and the following apply to this document. 3.1
Interlaboratory detection limit
Interlaboratory Detection EstimateIDE
The minimum concentration that can be detected with a high probability, that is, at a 90% confidence level, the proportion of laboratories that detect samples with a concentration of IDE is 95%, and the proportion of laboratories that do not detect samples with a concentration of 0 is 99%. 3.2
InterlaboratoryQuantitationEstimateThe minimum concentration corresponding to the interlaboratory limit of quantitation
RSD equals Z%.
Interlaboratory Critical Limit
Interlaboratory Critical LimitAt a 90% confidence level, the proportion of laboratories that correctly do not detect samples with a concentration of 0 is 99%. 3.4
experimental standard deviation Experimental standard deviation
Standard deviationstandarddeviation
Measurement of the same measured quantity by n times, which characterizes the dispersion of the measurement results. Note 1: The experimental standard deviation of a single measured value in n measurements can be calculated by using the Bessel formula: ()
GB/T27415—2013
Where:
Measured value of the nth measurement:
Number of measurements;
The arithmetic mean of a group of measured values obtained by
Measurements Note 2: The experimental standard deviation of the arithmetic mean of 1 measurement5 (1) is: s)-s/Vn
[JJF1001—2011, Definition 5.17]
Measurement modelmeasurementmodel,model of measurementmodel
The mathematical relationship between all known quantities involved in the measurement. Note: The general form of the measurement model is the square (Y, X) derived from the input quantity Xi, X. [JJF1001—2011
Definition 531
, where the output quantity I in the measurement model is the measured quantity, and its value is determined by the bias correction regression model
recoverymodel of bias correction model R
measurement concentration true liquid concentration (concentration indication) regression line model manual bias correction 3.7
measurement bias
measurement bias
measurement bias
estimate of systematic measurement error
LJJF100
2011.Definition 5.
intermediate precision
condition of mcasurement
intermediate precision Condition refers to a set of measurement conditions for repeated measurements of the same or similar measured objects in the same place and with the same measurement procedure over a longer period of time. It may also include other conditions that are suitable for changes. Note 1: Changes may include new standards, measurement standards, operators and measurement systems. Note 2: The description of the conditions should include the new and changed conditions and the actual extent of the changes. Note 3: In chemistry, the term "inter-sequence precision measurement conditions" is sometimes used to refer to "inter-sequence precision measurement conditions". Note 4: The standard deviation of inter-sequence precision is 5 and the symbol for the inter-sequence precision limit is R [JJF1001-2011, definition 5.11]
referencematerial
reference material
Standard material
Material with specific characteristics that are sufficiently uniform and stable, and whose characteristics are proven to be suitable for the intended use in measurement or in the inspection of nominal characteristics.
Note 1: Both assigned and unassigned standard materials can be used for measurement precision control, and only assigned standard materials can be used for calibration or measurement accuracy control. Note 2: In a specific measurement, a given standard material can only be used for one of the two purposes, calibration or quality assurance. [JJF1001—2011, definition 8.14]
Censoreddata
Censoreddata
Data that does not report a value but reports "not detected" or "less than". 2
Statistical tolerance intervalstatistical tolerance interval determined by a random sample and containing at least a specified proportion of the sampling population with a specified probability. Note that the confidence level of the interval established in this way is the probability that it contains at least a specified proportion of the sampling population when used repeatedly. [GB/T3359—2009, definition 3.1.1]4 General
GB/T27415—2013
4.1 After removing outliers, the collaborative test shall retain the true concentration, prepared concentration or concentration indication T), and its measurement data shall come from at least 6 independent laboratories.
4.2 Identify and fit the standard deviation model (SD model) of constant or straight line, establish the relationship between the SD model and T, and perform statistical tests and graphical analysis.
4.3 If the standard deviation (s) varies with T, weighted least squares (WLS) shall be used to fit the bias-corrected regression model (model R) and test its degree of fit. The SD model is used to estimate the standard deviation of blank sample measurements [s(O)], and at a 90% confidence level, gives the inter-laboratory critical value (ICL), the measured value (YC) at the ICL concentration, and the direct or alternative calculated value of the inter-laboratory detection limit (IDE). 4.4 According to the selected SD model, using RSD = 10%, the lowest concentration of the inter-laboratory quantification limit (IQE) IQE1% is obtained. If there is no IQE%, IQE% or IQE% can be calculated
4.5IDE and IQE are important for data use and provide a basis for the selection and use of detection methods. Design of IDE and IQE
Analyte selection
Select analytes with trace or near-trace concentrations. 5.1.2 The maximum T should exceed the expected value of IDE or IQE by more than 2 times. 5.1.3 Model R should cover sample levels ranging from 0 to the maximum T, which is helpful for the statistical significance test of SD model and model R. 5.2 Concentration Design
5.2.1 Choose at least 5 T of IDE (initial estimate IDE, can be 3s, s is the s at non-zero concentration trace level), one of the following schemes can be selected
a).IDE/4.IDE/2IDE,2XIDE4XIDE.0,IDE/2.IDE(3/2)XIDE..2XIDE.(5/2)XIDE.b)
other schemes, blank, at least one approximate 2×IDE., at least one non-zero concentration lower than IDE. e
At least 7 T of IQE are selected (initial estimate IQE. 10s can be taken), and one of the following schemes can be selected: 5.2.2
0IQE./4,QE./2,IQE..2×IQE..4×IQE.8×IQE..a)
b)0,IQE./2IQE,(3/2)XIQE.2XIQE.(5/2)XIQE.3XIQE.Other schemes, blank, at least one approximately 2×IQE, at least one non-zero concentration lower than IQE. 5.3 Daily error sources that can be covered
5.3.1 The error sources included in IDE and IQE involve but are not limited to: inherent errors of the instrument, certain transfer errors, as well as laboratories, operators, samples, instruments, reagents and environments. 5.3.2 It is assumed that the various variations of the data collected by no less than 6 participating laboratories are normally distributed (see GB/T27407). 5.3.3 Each laboratory shall measure the randomly distributed standard materials under the conditions of period precision 3
GB/T27415—2013
5.4 Avoidable sources of error
The sources of error that IDE and IQE should reasonably avoid include but are not limited to: sample crystal changes, equipment changes in the method, and transcription errors.
5.5 Missing data
5.5.1 After removing outliers, each laboratory shall submit method blanks and uncorrected measurement data, which shall be submitted to the organizer of the collaborative test to decide whether to correct them.
5.5.2 Among the retained data under any T, the proportion reported as "not detected" or "less than" should be less than 10%. 5.5.3 If there are too many missing data, the organizer should inform the participating laboratories to select the "uncensored" data. If this cannot be remedied, a higher T standard substance measurement should be added to the IDE study. 6 Model fitting and testing
6.1SD model
Constant model
Model fitting is shown in formula (1):
Where:
Standard deviation
Fitting constant:
Error item.
6.1.1.2 Ordinary most common (OLS) combination entered in 6.2 is shown in Table
Table 1 Model R coefficient estimation OLS and WLS calculation OLS ||tt ||Slope=6=Sv/Sm
Intercept--
5m-(T-)
Sr-ZT,-()
Slope=b=Sury/Sarr
Intercept=a—6T
6.1.2 Straight-line model
Model fitting is shown in formula (2):
Where:
—Fitting constant:
True concentration, prepared concentration or concentration indication, S=g+AT+E
6.1.2.2 OLS regression of beauty on T, see formula (3):Sa=g+hT
Where:
Standard deviation estimate of the measured value at the first level: T value of the first level.
GB/T27415—2013bzxZ.net
(2 )
(3)
6.1.2.3 Test the P value of "slope is zero". If the P value is <0.05, it means that the slope is significant and there is an obvious trend in the plot of s against T. Select the linear model.
6.1.2.4 Calculate r as shown in formula (4):
r=s-(g+hT)
Where:
Residual value:
Standard deviation of the measured value at the first level.
.(4)
6.1.2.5 If the plot of T is randomly distributed, select the linear model (see Appendix A): If there is an obvious trend in the plot, consider other optional models (see Appendix B). 6.1.2.6 If the linear model is selected, proceed to the weighted least squares (WLS) fitting in 6.2, see Table 1, The weight calculation is shown in formula (5) t
Where:
Weight.
6.2 Model R
The fitting of model R is shown in formula (6):
Y-a+bT+e
Where:
Measurement result:
Estimated trimming value:
Estimated slope value,
If the goodness of fit meets the following conditions, the model R can pass the test (see GB/T22554): 6.2.2
The ratio test of lack of fit and experimental error has a p value>0.05. a)
There is no obvious trend in the residual plot
6.2.3If the goodness of fit does not meet the above conditions, the organizer should decide whether to continue to use part of the data or provide more data for analysis.
7 Calculation of IDE and IQE
7.1 Calculation of IDE
7.1.1 The calculation of YC and ICL is given by equation (7) and equation (8), respectively, where k1 is given in Table 2. GB/T27415—2013
In the equation:
In the equation:
Corresponding to the measured value of ICL:
YC=klxs(0)+a
90% confidence level 99% quantile tolerance interval upper limit adjustment factor; standard deviation estimate of the measurement of the blank sample. ICL_YC-a
Inter-laboratory critical value
Table 290% confidence level statistical tolerance interval upper limit adjustment factor number of retained observations,
99% quantile,1
The IDE calculation of the constant model is given by equation (9), where 2 is given in Table 2. IDE=LC+2×3(0)
Where:
Inter-laboratory detection limit:
-90% confidence level 95% quantile tolerance interval upper limit adjustment factor. 95% quantile, #2
.(8)
........(9)
7.1.3 The IDE calculation of the linear model is given by formula (10), in which the IDE estimated by each iteration is substituted into the regression formula to give a new IDE, until the difference between the consecutive IDEs obtained by the iteration is less than 1%. IDE=[1×i(0)+k2x(g+hx IDE.))6
.........C10
Wherein:
GB/T27415—2013
The first alternative value of IDE, the initial estimated value IDE, can be 2LC or LC+starting 2×(0) According to Table 3 and formula (11), the adjusted value of IDE is given as: 7.1.4
IDEIDEXa
Bias correction adjustment value of IDE:
Bias correction adjustment factor.
Table 3 Bias correction adjustment factor a2 for IDE and IOE
Note: When n>10=1+4(-1)
Calculate YD according to formula (12):
Where:
Measured value at IDE concentration.
7.2IQE calculation
YD-a+6XIDE
...(11)
7.2.1First calculate IQE10%. If IQE% does not exist, calculate IQE2%. If IQE20% does not exist, calculate IQE%. 7.2.2 To find an appropriate 2%, calculate 2 using formula (13): Z'_100×h
Where:
Ratio of Zh to b.
7.2.3 Obtain IQE2% under different models, where 2 is approximately 10, 20 or 30. 7.2.4 The IQE calculation of the constant model is shown in formula (14): IQEzy
Where:
The inter-laboratory quantification limit when the relative standard deviation (RSD) is Z%. 7.2.5 The IQE calculation of the linear model is shown in formula (15): IQE%
7.2.6 According to Table 3 and formula (16), the adjusted value of 1QE is given: IQE#=IQE×a
Where:
IQE network
Bias correction adjustment value of IQE.
(12)
+###(13)
(14)
(15)
(16)
GB/T27415—2013
8 Report
Content of the report
Identification of personnel and analytical methods, selection of analytes, matrix, sample characteristics and methods used. All abnormal phenomena found in the study.
Description of data screening and the results retained, outlier removal, missing values and percentage of data used after comparison with the original data set. Statistical analysis of the data and calculation of ICL, IDE and IQE. The selected SD model with reasons.
Estimated coefficients of the SD model and model R.
Content of confirmation
Whether the data are copied and reported correctly. Whether the analysis of the data is correct.
Appropriateness of the use of the analytical results, including the likelihood of assumptions required for the calculations. Completeness assessment of the final report of IDE and IQE calculations. Review content
Whether the report is accompanied by a statement of the review and results. 8
A.1 Overview
Appendix A
(Informative Appendix)
Example of linear model calculation for IDE and IQE
GB/T274152013
In this example, 10 laboratories participated in the IDE and IQE study of the test method. There are 5 T (ug/L) under the IDE study: 0.0.0.25, 0.501.0 and 2.0, and there are 7 1 (g/L) under IQE Research: 0.0 .0.5, 1.0.2.0.4.0.8.0 and 12.0. This example is also applicable to the evaluation of detection limit and quantification limit within the laboratory A.2 IDE linear model statistics
SD model identification and fitting
Table A1 gives the measurement results and statistics of R10 at the factory T level liquid concentration.
IDE measurement report and statistical calculation
Inter-laboratory
This pair of tests
3.9, 2.22#3#8196, 0 .
7.34,6.41
1361066
15-10419-737:27
12.88.18.31.16.47-16.06.12.56.14721.13.96.17.37 Note 1: The data in the table are not eliminated by slow determination
Unit: g/L
Note 2: This example gives the blank measurement value of the load height and the abnormal regression slope to distinguish the measurement The difference between the value and the work. In fact, the distance and slope should be close to 0 and 1 respectively
A.2.1.2 The OLS regression of s on T in Table A,1 has: intercept -1.089 slope h=0.957. The node value of the "slope is zero" test = 0.0128 <0.05, and the choice of constant model should be rejected. In addition, Figure A,1 shows that 3 increases with T, so it is recommended to use the linear model
8=1.089+0.9577
T under fitting05, the choice of constant model should be rejected. In addition, Figure A, 1 shows that 3 increases with T, so it is recommended to use the linear model
8=1.089+0.9577
T under the fitting05, the choice of constant model should be rejected. In addition, Figure A, 1 shows that 3 increases with T, so it is recommended to use the linear model
8=1.089+0.9577
T under the fitting
Tip: This standard content only shows part of the intercepted content of the complete standard. If you need the complete standard, please go to the top to download the complete standard document for free.