Accred Qual Assur (1998) 3 : 150154 Q Springer-Verlag 1998

SHORT CONTRIBUTION

Rdiger Kaus

Detection limits and quantitation limits in the view of international harmonization and the consequences for analytical laboratories

Received: 3 November 1997 Accepted: 25 January 1998 Presented at: 2nd EURACHEM Workshop on Measurement Uncertainty in Chemical Analysis, Berlin, 2930 September 1997 R. Kaus (Y) Fachhochschule Niederrhein, Fachbereich Chemie, Adlerstrasse 32, D-47798 Krefeld Tel.: c49-2151-822161 e-mail: kaus@kr.fh.niederrhein.de

Abstract In the literature and in daily work there exist many different models to describe detection limits and quantitation limits in the chemical measurement process (CMP). It is still an open question whether the evaluation of detection and quantitation limits should be an one-off process carried out during the validation of a CMP, or whether it should be a continuous process. Because of these reasons

many laboratories have difficulties to interprete such values. It is necessary that the official standardization bodies should do some efforts in the international harmonization of the special standards concerning detection and quantitation limits. Key words Chemical measurement process 7 Capability of detection 7 Detection limit 7 Quantitation limit 7 International harmonization

Introduction
For these reasons, Currie et al. published in 1995 the IUPA recommendation Nomenclature in evaluation of analytical methods including detection and quantification capabilities [4]. DIN 32645 is mostly in accordance with these recommendations [5]. The practical procedure of evaluating performance characteristics is described in the IUPAC recommendations only in a general form. The description in DIN 32645 seems to merit discussion, as there are quantitative differences from the formulae published by Funk et al. [6] in a study sponsored by the Umweltbundesamt. The difficulties for commercial laboratories which try to calculate detection and quantitation limits using the formulae in the IUPAC recommendations or DIN 32645 are not described in the literature in all their aspects. I have found only one comment on how often the validation process should be done [7]. But it is still an open question whether the evaluation of detection and quantitation limits should be a one-off process (maybe once a year or more often) carried out during the validation or revalidation of a CMP, or whether it should be a continuous process.

If the calculations are done at each calibration process as a daily routine, one will get values for quantitation limits which are higher than some calibration points. If these values are passed on to customers, the latterare likely to be dissatisfied. In Table 1, several calculation formulae are compared. The measured data in Fig. 1 are taken from DIN 32645.

A personal note
When I started my job as an external quality assurance manager in a commercial analytical laboratory, one of my first tasks was the calculation of detection limits and quantitation limits directly from the daily calibration procedures, for example from AAS or GC measurements. The analysts were at first very surprised by the results obtained, and when we revealed values to some of our customers they suspected incompetence. We had to make many efforts to convince them that these were realistic statistical values and that if our competitors were able to provide better data, these were likely to be from one very careful and costly validation process

151

Table 1 Some alternative definitions of the detection limit and the quantitation limit From DIN 32645 [5] Symbol xNG Name Detection limit Estimated value x(NG) Formula From blank xNG p sL mtf; a b 1 1 "mc n 1 1 x ps 7t " c c m n Q
x0 f; a

2 x

From calibration data xNG xEG xBG Detection limit Estimated value bei Schnellschtzung x(EG) Quantitation limit (IUPAC) Estimated value x(BG) From Funk [6] Symbol XN xP XB Name Capability of detection Testing value to secure the lower range limit Limit of determination Formula XNp(tactb)7s0 XNpxP If 0.5 x1~xP~x1

IUPAC: detection limit Funk: capcability of detection 1 1 (xBGPx) 2 c c xBGpk7sx 07tf; a m n Qx k free to choose, kp3 B 33%

"

XBpt7s0 / VBrel, max7100% VBrel, max free to choose (33%)

From IUPAC [2] Symbol xD Name Detection limit Formula Scpt1Pa, n s0 xDp(dab,n s0 /A) (K/I);2t1Pa,n s0 /A7(K/I) where Kp1cr(B, A) (sB / s0) [t1Pa,n (s /A)] and Ip1P[t1Pa,n (s /A] 2 (if the calibration function is estimated)

Coleman [8]/Hubaux-Vos model [9] Symbol xD Name Detection limit s 1 1 (xDPx) 2 x 2 1/2 ctnP2, 1Pb 1 c c xD p 7 tnP2, 1Pa 1 c c b n Sxx n Sxx Explicit PJc(J 2P4 HK) 1/2 xDpDLH-V p 2H Recurcive formula Formula
1/2

4 2

Ap

b s7tnP2, 1Pb

Bp 1 n

tnP2, 1Pa 1 x2 1c c tnP2, 1Pb n Sxx

1/2

Fp B 2P 1 c

247S

xx

Gp2 AB7Sxx HpA 27S P1 xx JpGc2 x KpFPx 2

once a year, from literature, or from the data sheets of the producer of the equipment. Furthermore, it was necessary economically to publish only the values for detection limits and these only to the customers who demanded them.

Which values of detection limits and quantitation limits are correct?

The various calculation procedures cited in literature give quantitative differences, as some parameters, e.g.

152

confidence levels, could be chosen arbitrarily (1-a or 1-b could be 0.05 or 0.01). To obtain quantitation limits, a confidence range must be chosen: 1. Recommendation of DIN 32645: kp3 (B33%) 2. Recommendation of IUPAC: 1/kp0.1 (B10%) Thus, one can obtain true values within the laws of mathematics and statistics which differ from each other. Such values, including the choosen parameters, should be absolutely correct. But these could be understood only by experts, and such results need interpretation.

strate the difficulties or they did not realize they existed.

Detection limits from blanks problems with the normal distribution

Calculating detection limits from blanks has the advantage of giving very low values, but ignores the fact that these data are very seldom from the same statistical universe as the calibration data and measured data, and often are not normally distributed. Automatic and digitalized measurement equipment in particular do not give values under a certain limit, but give the value of zero only. On the EXCEL sheet in Fig. 4, this is reported for the determination of naphthalene with GC/ MS. If the calculating is performed using statistical programs, which take only normal distributions in account, the result will not be correct.

Quantitation limits and working range

There is another problem which seems not to be discussed in the literature. This is that to do calibration measurements a working range must be chosen, and normally this is the range within which the expected concentrations of the samples fall. Calculation of the quantitation limit from these calibration data often gives a value which is higher than some of the calibration data (in the procedure suggested by Funk [6] the quantitation limit is always higher than the first calibration point). The difficulty is to decide on the lowest concentration that can be published. This problem becomes apparent on calculating the quantitation limit of the measured data published in DIN 32645. This is shown by the EXCEL sheet given in Fig. 3. The data have poor precision. Either the authors of DIN 32645 chose these data in order to demonFig. 1 The values are calculated from the formulae from Funks book [6] in an EXCEL sheet

Conclusions
1. The IUPAC recommendations should be illustrated by calculation examples from common practice. 2. The confidence limits (1-a; 1-b) used in the calculation of the detection limit should be standardized in an ISO guide. 3. The calculation of the detection limit from blanks should not be recommended. It would be better to determine standardized reference samples which have, for example, 0.10.5 times the concentration of the lowest calibration sample. (here I remember a recommendation to publish the performance charac-

Fig. 2 The values are calculated from the formulae from DIN 32645 [5] in an EXCEL sheet

153

Fig. 3 EXCEL sheet with graphical representation of the process data

teristics of the validation in every norm of the analytical determinations) 4. It there is agreement that there should be a quantitation limit, the recommendation of DIN 32645 (33%) would be preferred instead of the UPAC recommendation (10%) to be practicable in a commercial laboratory.

The calculations performed with the EXCEL sheets may be obtained from the author.

154

Fig. 4 Example of the difficulties in calculating the detection limit from blanks

References
1. Kaiser H (1965) Z Anal Chem 209 : 1 2. Freiser H, Nancollas GH (eds) (1987) IUPAC Orange Book: Compendium of analytical nomenclature, 2nd edn. Blackwell Scientific, Oxford 3. Danzer K (1996) In: Gnzler H (ed) Accreditation and quality assurance in analytical chemistry. Springer, Berlin Heidelberg New York 4. Currie LA (1995) Pure Appl Chem 67 : 16991723 5. DIN 32645 (1994) Chemische Analytik Nachweis-, Erfassungs- und Bestimmungsgrenzen. Beuth, Berlin 6. Funk W, Dammann V, Donnevert G (1992) Qualittssicherung in der Analytischen Chemie. VCH, Weinheim 7. Wegscheider W (1996) In: Gnzler H (ed) Accreditation and quality assurance in analytical chemistry. Springer, Berlin Heidelberg New York 8. Coleman D (1997) Statistically valid detection limits and quantitation limits. Short course at the PITTCON conference 9. Hubaux A, Vos G (1970) Anal Chem 42 : 849855