1、Designation: D 6091 07An American National StandardStandard Practice for99 %/95 % Interlaboratory Detection Estimate (IDE) forAnalytical Methods with Negligible Calibration Error1This standard is issued under the fixed designation D 6091; the number immediately following the designation indicates th
2、e year oforiginal adoption or, in the case of revision, the year of last revision. A number in parentheses indicates the year of last reapproval. Asuperscript epsilon (e) indicates an editorial change since the last revision or reapproval.1. Scope1.1 This practice establishes a standard for computin
3、g a99 %/95 % Interlaboratory Detection Estimate (IDE) and pro-vides guidance concerning the appropriate use and application.The calculations involved in this practice can be performedwith DQCALC, Microsoft Excel-based software availablefrom ASTM.21.2 The IDE is computed to be the lowest concentratio
4、n atwhich there is 90 % confidence that a single measurement froma laboratory selected from the population of qualified labora-tories represented in an interlaboratory study will have a truedetection probability of at least 95 % and a true nondetectionprobability of at least 99 % (when measuring a b
5、lank sample).1.3 The fundamental assumption of the collaborative studyis that the media tested, the concentrations tested, and theprotocol followed in the study provide a representative and fairevaluation of the scope and applicability of the test method aswritten.When properly applied, the IDE proc
6、edure ensures thatthe 99 %/95 % IDE has the following properties:1.3.1 Routinely Achievable IDE ValueMost laboratoriesare able to attain the IDE detection performance in routineanalyses, using a standard measurement system, at reasonablecost. This property is needed for a detection limit to bepracti
7、cally feasible. Representative laboratories must be in-cluded in the data to calculate the IDE.1.3.2 Routine Sources of Error Accounted forThe IDEshould realistically include sources of bias and variation whichare common to the measurement process. These sourcesinclude, but are not limited to: intri
8、nsic instrument noise, sometypical amount of carryover error, plus differences in labora-tories, analysts, sample preparation, and instruments.1.3.3 Avoidable Sources of Error ExcludedThe IDEshould realistically exclude avoidable sources of bias andvariation, that is, those which can reasonably be a
9、voided inroutine field measurements. Avoidable sources would include,but are not limited to: modifications to the sample, measure-ment procedure, or measurement equipment of the validatedmethod, and gross and easily discernible transcription errors(provided there was a way to detect and either corre
10、ct oreliminate them).1.3.4 Low Probability of False DetectionThe IDE is atrue concentration consistent with a measured concentrationthreshold (critical measured value) that will provide a highprobability, 99 %, of true nondetection (a low probability offalse detection, a = 1 %). Thus, when measuring
11、 a blanksample, the probability of not detecting the analyte would be99 %. To be useful, this must be demonstrated for the particularmatrix being used, and not just for reagent water.1.3.5 Low Probability of False NondetectionThe IDEshould be a true concentration at which there is a highprobability,
12、 at least 95 %, of true detection (a low probabilityof false nondetection, b = 5 %, at the IDE), with a simulta-neous low probability of false detection (see 1.3.4).Thus, whenmeasuring a sample at the IDE, the probability of detectionwould be at least 95 %. To be useful, this must be demonstratedfor
13、 the particular matrix being used, and not just for reagentwater.NOTE 1The referenced probabilities, a and b, are key parameters forrisk-based assessment of a detection limit.1.4 The IDE applies to measurement methods for whichcalibration error is minor relative to other sources, such aswhen the dom
14、inant source of variation is one of the following(with comment):1.4.1 Sample Preparation, and calibration standards do nothave to go through sample preparation.1.4.2 Differences in Analysts, and analysts have little oppor-tunity to affect calibration results (such as with automatedcalibration).1.4.3
15、 Differences in Laboratories, for whatever reasons,perhaps difficult to identify and eliminate.1This practice is under the jurisdiction of ASTM Committee D19 on Water andis the direct responsibility of Subcommittee D19.02 on General Specifications,Technical Resources, and Statistical Methods.Current
16、 edition approved March 1, 2007. Published April 2007. Originallyapproved in 1997. Last previous edition approved in 2003 as D 6091 03.2Available fromASTM International Headquarters. OrderAdjunct No.ADJDQ-CALC. Original adjunct produced in 2007.1Copyright ASTM International, 100 Barr Harbor Drive, P
17、O Box C700, West Conshohocken, PA 19428-2959, United States.1.4.4 Differences in Instruments (measurement equipment),which could take the form of differences in manufacturer,model, hardware, electronics, sampling rate, chemical process-ing rate, integration time, software algorithms, internal signal
18、processing and thresholds, effective sample volume, and con-tamination level.1.5 Alternative Data Quality ObjectivesOther valuesfora, b, confidence, etc. may be chosen for calculating an IDE;however, this procedure addresses only the 99 %/95 % IDE.2. Referenced Documents2.1 ASTM Standards:D 2777 Pra
19、ctice for Determination of Precision and Bias ofApplicable Test Methods of Committee D19 on Water2.2 ASTM Adjuncts:DQCALC Microsoft Excel-based software for the Inter-laboratory Quantitation Estimate (IQE)23. Terminology3.1 Definitions:3.1.1 99 %/95 % Interlaboratory Detection Estimate (99 %/95 % ID
20、E, also denoted LD for Limit of Detection in accor-dance with Currie (1)3The lowest concentration at whichthere is 90 % confidence that a single measurement from alaboratory selected from the population of qualified laborato-ries represented in an interlaboratory study will have a truedetection prob
21、ability of at least 95 % and a true nondetectionprobability of at least 99 %.3.2 Definitions of Terms Specific to This Standard:3.2.1 Censored MeasurementA measurement that is notreported numerically nor is reported missing but as a nondetector a less-than, for example, “less than 0.1 ppb.” The form
22、ermeans that an algorithm in the measurement system deter-mined that the measurement should not be reported numeri-cally for one of two reasons: either it was considered notsufficiently precise or accurate, or the identification of theanalyte was suspect. A reported less-than may have the samemeanin
23、g, but it also implies (perhaps erroneously) that anyconcentration greater than or equal to the accompanying value(for example, 0.1 ppb) can be measured and will be reportednumerically.3.2.2 Detection Limit (DL) or Limit of Detection (LD)Anumerical value, expressed in physical units or proportion,in
24、tended to represent the lowest level of reliable detection (alevel which can be discriminated from zero with high prob-ability while simultaneously allowing high probability ofnondetection when blank samples are measured.NOTE 2In some cases, the discrimination may be from a value otherthan zero, suc
25、h as a background level. Note also that a DL also dependson other characteristics of the measurement and detection process, such asdescribed in 1.3.2. The IDE is an example of a DL.3.2.3 Probability of False DetectionThe false positiveprobability, denoted a, that a single measurement of a blanksampl
26、e will result in a detection. (See Fig. 1.) This probabilityis often referred to as the Type 1 error probability and dependson the analyte, measurement system, analytical method, ma-trix, analyst, and measurement (recovery) threshold (measure-ment critical value) used to decide whether detection has
27、occurred. This definition can be generalized to refer to un-wanted detection from a single measurement of a sample at anynonzero concentration of the analyte rather than a blanksample, provided that the nonzero concentration is less than thedetection limit or IDE.3.2.4 Probability of False Nondetect
28、ionThe false nega-tive probability, denoted b or b (T), that a single measurementof a sample containing a nonzero concentration, T,ofananalyte of interest will result in a nondetection. This is thecomplement of the probability of true detection. (See Fig. 1.)This probability function is often referr
29、ed to as the Type 2 errorprobability function, and it depends explicitly on the concen-tration (T). It depends implicitly on the analyte, measurementsystem, analytical method, matrix, analyst, and critical valuefor detection.3.2.5 Probability of True DetectionThe probability, de-noted 1-b or 1-b (T)
30、, that a single measurement of a samplecontaining a nonzero concentration, T, of an analyte of interestwill result in a detection. (See Fig. 1.) This probability is often3The boldface numbers in parentheses refer to the list of references at the end ofthis standard.FIG. 1 Simplest Case of Reliable D
31、etectionD 6091 072referred to as statistical power or the power of detection, and itdepends explicitly on the concentration (T). It depends implic-itly on the analyte, measurement system, analytical method,matrix, analyst, and critical value for detection.3.2.6 Probability of True NondetectionThe tr
32、ue negativeprobability, denoted 1-a, that a single measurement of a blanksample will result in a nondetection. This is the complement ofthe probability of false detection. (See Fig. 1.) This probabilityalso depends on the analyte, measurement system, analyticalmethod, matrix, analyst, and response t
33、hreshold. The probabil-ity of true nondetection can be similarly generalized: it canapply to a single measurement of a sample at any nonzeroconcentration less than the detection limit or IDE.3.2.7 100(1-g)%Confidence Statistical Tolerance Limitfor 100(1-d) % of a Population (also known as a One-Side
34、dStatistical Tolerance Interval)Astatistically determined limitthat will, with 100(1-g) % confidence, exceed (or fall below)100(1-d) % of the population (the 100(1-d) % quantile). SeeHahn and Meeker (2) for further explanation and tables ofvalues.4. Summary of Practice4.1 Every ASTM D-19 test method
35、 is evaluated to deter-mine precision and bias by conducting a collaborative study inaccordance with Practice D 2777. That study, or a similarcollaborative study, can also be used to evaluate the lowestconcentration level of reliable detection for a test method,referred to herein as the Interlaborat
36、ory Detection Estimate.Such a study must include concentrations suitable for modelingthe uncertainty of mean recovery of interlaboratory measure-ment (preferably without extrapolation). It must also beplanned and conducted to allow the known, routine sources ofmeasurement variability to be observed
37、at typical levels ofinfluence. After it is conducted, outlying laboratories andindividual measurements should be eliminated using an ac-cepted, scientifically based procedure for outlier removal, suchas found in Practice D 2777. The IDE computations must bebased on retained data from at least six in
38、dependent laborato-ries at each concentration level.4.2 Retained data are analyzed to identify and fit one ofthree proposed interlaboratory standard deviation (ILSD) mod-els which describe the relationship between the interlaboratorystandard deviation of measurements and the true concentration.The i
39、dentification process involves evaluating the models inorder, from simplest to most complex: constant, straight-line,or exponential (all with respect to true concentration, T).Evaluation includes statistical significance and residual analy-sis.4.3 The chosen model is used to predict interlaboratorym
40、easurement standard deviation at any true concentrationwithin the study concentration range. If interlaboratory stan-dard deviation is not constant, the predictions are used togenerate weights for fitting the mean recovery relationship (thestraight-line relationship between measured concentration an
41、dtrue concentration), using weighted least squares (otherwise,ordinary least squares is used). The mean recovery curve isevaluated for statistical significance and lack of fit and usingresidual analysis. An ILSD model prediction is also used toestimate the interlaboratory standard deviation of measu
42、re-ments of blanks. This estimate is used to compute YC,ameasurement critical value for detection (see 6.4.1). The YC isthe value that with approximately 90 % confidence will not beexceeded by 99 % of all measurements of blanks made byqualified laboratories as represented in the study. The LCcompute
43、d from YC is the true concentration with expectedmeasurement equal to YC (see 6.4.2). The model is also used topredict interlaboratory standard deviation at nonzero concen-trations. The IDE is directly or iteratively computed to be thetrue concentration that with approximately 90 % confidencewill pr
44、oduce measurements that will exceed YC at least 95 %of the time and simultaneously not exceed more than 1 % ofthe time when blank samples are measured.5. Significance and Use5.1 Appropriate application of this practice should result inan IDE achievable by most laboratories properly using the testmet
45、hod studied. This IDE provides the basis for any prospec-tive use of the test method by qualified laboratories for reliabledetection of low-level concentrations of the same analyte as theone studied in this practice and same media (matrix).5.2 The IDE values may be used to compare the detectionpower
46、 of different methods for analysis of the same analyte inthe same matrix.5.3 The IDE provides high probability (approximately95 %) that result values of the method studied which exceedthe IDE represent presence of analyte in the sample and highprobability (approximately 99 %) that blank samples will
47、 notresult in a detection.5.4 The IDE procedure should be used to establish theinterlaboratory detection capability for any application of amethod where interlaboratory detection is important to datause. The intent of IDE is not to set reporting limits.6. Procedure6.1 The procedure described as foll
48、ows has stages describedin the following sections: IDE Study Plan, Design and Protocol(6.2); Conduct the IDE Study, Screen the Data, and Choose aModel (6.3); and Compute the IDE (6.4). A flowchart of theprocedure is shown in Fig. 2.6.2 IDE Study Plan, Design, and Protocol:6.2.1 Choose Analyte, Matri
49、x, and MethodAt least oneanalyte of interest is selected, typically one for which there isinterest in trace levels of concentration, such as toxic materialsthat are controlled and regulated. For each analyte, an approxi-mate maximum true concentration is selected based on thefollowing considerations:6.2.1.1 The anticipated IDE should be exceeded by a factorof 2 or more,6.2.1.2 A single model (ideally a straight-line model in trueconcentration, T) should describe mean recovery from zero tothat maximum concentration,6.2.1.3 Asingle model in true concentration should des
