International interlaboratory study of forensic ethanol standards

Transcription

1 International interlaboratory study of forensic ethanol standards Bernard King and Richard Lawn LGC, Queens Road, Teddington, Middlesex, UK TW11 0LY. Received 18th January 1999, Accepted 10th May 1999 In March, 1996, CITAC (Cooperation on International Traceability in Analytical Chemistry) initiated a task to assess the comparability of the standards used in different countries. The Laboratory of the Government Chemist (LGC) acted as the project leader. Three certified solutions with nominal ethanol concentrations in water of 50, 80 and 200 mg per 100 ml were distributed to sixteen standards laboratories in Europe, Asia, North America and Australia. The laboratories analysed the solutions by methods of their choice and reported the results, together with measurement uncertainty data and other information such as accreditation status and level of experience. Six out of sixteen laboratories fully met the study target criteria of the mean results for all three solutions being within ±1% of the certified values. Participants were encouraged to assess the adequacy of their performance for their particular application. The paper reports the study protocol, the results of the interlaboratory comparison and discusses the results in the context of the metrological performance required to support routine measurement used for prosecution purposes. Introduction Ethanol in blood and urine analyses are routinely undertaken in both clinical 1 and forensic laboratories, in connection with drinking and driving legislation, for example, under the UK Road Traffic Act The permitted level of ethanol varies from country to country in the range mg of ethanol per 100 ml of blood. Prosecution only takes place when the measured level exceeds the legal limit by a margin which aims to take account of the measurement uncertainty. Current best practice relies on method validation to establish that the method employed is free from bias and uses precision data to establish 99% confidence intervals. A variety of methods are used for the analysis, including gas chromatography (GC), with and without headspace (HS) techniques, infrared spectroscopy (IR), electrochemical detectors, and enzymatic test kits. GC has the advantage of being able to separate ethanol from other alcohols and other interfering substances. Calibration is carried out using ethanol in water standards, often purchased as standards (certified reference materials) supplied by national standards laboratories. NIST sell 95.6%, 2.00%, 0.096% and 0.023% aqueous ethanol standards, whilst LGC sell 80, 107 and 200 mg per100 ml ethanol standards, all with certified ethanol concentrations and stated uncertainties. Alternatively, laboratories may prepare their own in-house standards. The standards may be certified on the basis of gravimetric formulation, where accurately known quantities of wellcharacterized pure ethanol, or a concentrated ethanol in water solution, and pure water are gravimetrically mixed. Some laboratories use the potassium dichromate titration to certify analytical standards and/or to cross-check the gravimetric certification data. 2,3 Recent clinical studies 1 have established reproducibility data for the analysis of spiked ( mg per100 ml) blood serum for a number of methods. The reported mean % CV varied from 3.8% for GC packed (headspace) to 9.4% for an electrochemical analyser. Relative bias varied from +1.5% to 28.9% mean difference from the target value depending on the method. GCbased methods gave bias data ranging from 0.3 to 3.0% mean difference from the target value. The traceability of ethanol in water standards is to references such as pure ethanol and/or pure potassium dichromate and the associated titration method. Given the legal sensitivity of forensic measurement the robustness of traceability claims would be enhanced if the standards and the methods used for certification were shown to be comparable and equivalent on an international basis. CITAC initiated a task to compare measurements of ethanol in water standards undertaken by standards laboratories in various parts of the world. Consultation within CITAC, 4 CCQM 5 and EURACHEM 6 helped refine the study protocol and identify participating laboratories. Of the 18 laboratories who expressed an interest in participating, 16 laboratories submitted results. This paper reports the study protocol and the results of the interlaboratory comparison and discusses the results in the context of the metrological performance required to support routine measurement used for prosecution purposes. Measurement uncertainty estimates were made using the method described in the Eurachem Guide, 7 which also provides a general introduction to the subject. Experimental Preparation of solutions The forensic ethanol solutions were prepared using absolute ethanol obtained from Fisher Scientific UK (product code E/0650/15) and ultra-high-purity water obtained from an Elgastat water purification unit. The purity of the ethanol was determined by measurements made at LGC of the following parameters: water content, congener content (e.g., higher alcohols) and density. The water content was determined by Karl Fischer titration and a mean value (n = 5) of 0.043% was obtained, with an estimated standard uncertainty of %. Analyst, 1999, 124,

2 The ethanol was examined for the presence of the following congeners, using the technique of gas liquid chromatography with flame ionization detection: acetaldehyde, methanol, ethyl acetate, propanol, butan-2-ol, 2-methylpropanol, butanol, acetic acid, 2-methylbutanol and 3-methylbutanol. None of the congeners were detected and accordingly a value of zero was used for the contribution made by the congeners to the impurity content of the ethanol. However, as the detection limit for an individual congener was estimated to be % mass/mass (m/m), the uncertainty associated with the value of zero for an individual congener was calculated as /A3 = , assuming a rectangular distribution. Allowing for 10 possible congeners, the estimated standard uncertainty of the total congener content was 0 ± A10 3 ( ) 2 = %. Based on the above ( % water), the purity value adopted for the ethanol was % (m/m), with an uncertainty of %. As confirmation, a further estimate of the purity of the ethanol was obtained from the density measurements made at 20 ± 0.05 C. These measurements gave a mean (n = 4) density in air value of g ml 21, from which a purity value of % (m/m) was derived, by reference to alcohol tables. The solutions were prepared by transferring about 12 kg of water to a clean, dry, pre-weighed plastic container, fitted with a screw-cap at the top and a stopcock at the base. The container was re-weighed and the mass of water was recorded to 0.1 g. The volume of the plastic container was such that there was minimum headspace above the water surface. Based on the accurately known mass of water, the mass of ethanol required for the target ethanol concentration was weighed into a clean, dry, glass wide necked container fitted with a plastic lid. After transfer of the ethanol to the glass container, the plastic lid was quickly re-fitted and the glass container was re-weighed. The mass of the ethanol was recorded to g. In order to transfer the ethanol to the water, the plastic lid was removed from the glass container and both the lid and glass container were dropped carefully into the water. The required mass of HgCl 2 preservative (about 1.3 g, recorded to g) was also added to the water. The screw-cap was fitted to the large plastic container, which was then thoroughly mixed by shaking for about 5 min. The container was allowed to stand for 1 2 h and was then again mixed by further shaking. This procedure was repeated once more and, after standing overnight, the container was given a final mixing. Three solutions in all were prepared, with nominal target ethanol levels in the prepared solutions of 53, 83 and 197 mg per 100 ml. These solutions were labelled CITAC Solution 50, CITAC Solution 80 and CITAC Solution 200 for the purposes of the interlaboratory study. The solutions were then dispensed into 60 ml amber glass bottles fitted with screw-caps. The screw-caps were fitted with an inner polythene insert to prevent leakage. The bottles were filled to leave only a minimum headspace and bottling was completed in as short a time as practical in order to minimize the possibility of any compositional changes occurring during the bottling procedure. About 100 units (of 60 ml each) of each of the three solutions were bottled (i.e., about half the total volume of each solution was used). Gravimetric calculation of the accurate ethanol concentration values and their uncertainties Based on the accurately known masses of ethanol, water and HgCl 2 used in the preparation of the solutions and the measured purity of the ethanol, the concentration of ethanol in each of the three solutions was calculated on a mass basis. In order to also express the concentrations on a volume basis, the densities (in air) of the three solutions were measured at each of three different temperatures. For each solution, two bottles were selected and the density of the solution in each bottle was measured in duplicate. An electronic density meter was used for the measurements and the mean result (n = 4) obtained for each determination is shown in Table 1. The density meter was calibrated using air and distilled, demineralized water; the temperature measurements were made with a thermocouple that had been calibrated using a platinum resistance thermometer traceable to the ITS-90. In order to determine the uncertainties associated with the calculated concentration values, the following sources of uncertainty associated with the preparation of the solutions were evaluated. All weighings were carried out using balances that had been checked with certified weights that were traceable to the national kilogram, held at the National Physical Laboratory. The uncertainties in the weighings were estimated by comparing the observed values for certified weights to the actual certified values and by determining the standard deviation for replicate weighings of the same weight. The standard uncertainty associated with weighings of ethanol in the range 6 25 g was g. For weighings in the range kg, the standard uncertainty was 0.11 g. The uncertainty in the purity of the ethanol was estimated as described above as %. The uncertainty associated with possible loss of the ethanol by evaporation during transfer to the water was assessed by measuring the weight loss observed when a glass container containing an appropriate quantity of ethanol was left open to the atmosphere for about 10 s. This period of time was chosen to simulate the maximum likely time the glass container would be open and exposed to the air during the preparation of the solutions. From a series of 13 measurements, the maximum weight loss observed was g. Based on this observation, the standard uncertainty arising from ethanol evaporation was estimated as g. The uncertainty of the density determinations was assessed by means of measurements carried out on a dimethyl phthalate reference material with a certified density value (in air) of ± g ml 21 at C. Out of the three results obtained at 20 ± 0.05 C, the measurement showing the largest deviation from the certified value was g ml 21. Based on this result, and making some allowance for the different chemical compositions of the reference material and the forensic ethanol solutions, a standard uncertainty of g ml 21 was attributed to the density measurements when the temperature of measurements is controlled to ±0.05 C. The temperature tolerance of ±0.05 C reflects the temperature control feasible in a density determination by density meter. To reflect the temperature control that is more likely to apply in a typical laboratory environment when a portion (volume) of a forensic ethanol solution is taken for analysis, the effect of temperature on the density of the solutions was investigated. The results are shown in Table 1 and a graph of density versus temperature is, for all practical purposes, linear. The slope of the graph was estimated as g ml 21 C. From this it was calculated that for a temperature uncertainty of ±2 C (e.g., 20 ± 2 C), the corresponding standard uncertainty in a density value is /A3 = g ml 21. Table 2 summarizes the gravimetrically calculated ethanol concentration values in the prepared solutions and their associated expanded uncertainties (k = 3). Note: expanded Table 1 Density values for the forensic ethanol solutions Density Density Density ± 0.05 C/ ± 0.05 C/ ± 0.05 C/ Solution g ml 21 (in air) g ml 21 (in air) g ml 21 (in air) Analyst, 1999, 124,

3 uncertainty values with a coverage factor of three [U(k3)] are used throughout the paper to approximate to the 99% confidence intervals commonly used within the forensic field. Homogeneity and stability The homogeneity of each of the three solutions was assessed by selecting 8 bottles from the batch and analysing each bottle in duplicate for the ethanol content. The potassium dichromate method was used, in which an accurately known excess quantity of potassium dichromate (of certified purity) was used to oxidize the ethanol in a measured mass of sample. The unreacted dichromate was determined by potentiometric titration using ammonium ferrous sulfate, from which the ethanol content of the sample was calculated. The difference in the duplicate results within the bottles was used to estimate the analytical variation, which was then compared to the betweenbottle variation in an analysis of variance. Using the F-test, there was no evidence of inhomogeneity at the 95% confidence level. The long-term stability of aqueous forensic ethanol standards has been monitored at LGC for many years and Table 3 shows the results of the most recent study on a standard, with a nominal concentration of 80 mg per 100 ml, that had been stored at 4 C. Based on these results, it was concluded that the solutions distributed to laboratories would not undergo any significant change during the period of the study. Organization of the interlaboratory study Samples were distributed to laboratories in the week beginning April 28, 1997, and each laboratory was provided with ml bottles of each of the three solutions. Included with the samples was a protocol outlining the procedures to be followed by participant laboratories and the format for reporting results and other information. Laboratories were asked to analyse each bottle in duplicate, using an analytical method of their own choice and to submit their results to LGC by June 30, Laboratories were also asked to provide supplementary information on such matters as the analytical procedures used (methods, calibration standards, etc.), the uncertainties of their reported results and their experience of forensic ethanol analysis. Because results expressed on a volume basis are temperature-dependent, laboratories reporting such results were also asked to record the temperature at which their results applied. In the protocol for the study a target performance level was specified, which stated that results produced by participating laboratories should lie within ±1% of the certified values. The ±1% figure was chosen after consideration of measurement Table 2 Ethanol concentrations in the forensic ethanol solutions mg per 100 ml at mg per 100 ml at Solution mg per 100 g 20 ± 0.05 C 20 ± 2 C ± ± ± ± ± ± ± ± ± 0.33 Table 3 Stability results Results/mg per 100 ml Mean n s 28/3/ /4/ uncertainty aspects and current practice in evaluating results in order to decide whether or not to prosecute. Results The information provided by laboratories indicated that the following types of analytical method were used to analyse the forensic ethanol solutions: potassium dichromate oxidation of the ethanol (5 laboratories), gas chromatography (non-headspace) (6 laboratories), headspace gas chromatography (5 laboratories), infrared (1 laboratory) and enzyme (1 laboratory). All but two of the laboratories used methods that expressed results as mg per 100 ml; the other two laboratories used mg per 100 g. From the density values of the solutions (Table 1) it was concluded that the difference between two essentially identical results, one expressed as mg per 100 ml at ± 0.05 C, and the other expressed as mg per 100 g, would not exceed 0.3%. For those laboratories reporting results as mg per 100 ml, the ambient temperature varied from 17.6 to 26 C. For such a temperature range, a density difference of g ml 21 can be expected. Thus, the difference between two essentially identical results, one expressed as mg per 100 ml at 17.6 C and the other expressed as mg per 100 ml at 26 C, would not exceed 0.4%. As these effects were significantly less than the observed interlaboratory variation of the results, it was decided not to convert mg per 100 g results to an mg per 100 ml basis or to convert results reported as mg per 100 ml to a common temperature basis. Nine of the laboratories reported that they were accredited for this type of analysis, mainly to ISO Guide 25, and four laboratories took part in a proficiency testing scheme for forensic ethanol analysis. Several laboratories had many years experience of this type of analysis, while two laboratories had been involved in this type of work for 2 years or less. The number of forensic ethanol analyses carried out by the laboratories varied considerably, from analyses per year to less than 10 analyses per year. These issues are discussed later. Several laboratories provided information on the legal limits for alcohol in blood that are prescribed in their country s national legislation. Permitted blood alcohol levels vary from 20 mg per 100 ml to 90 mg per 100 ml. A number of laboratories reported that prosecutions are only initiated when the measured blood alcohol level exceeds the legal limit by a certain amount. For one country the legal limit is 20 mg per 100 ml and prosecutions are initiated at 26 mg per 100 ml. For other countries the figures are: 80/87, 90/91, 50/80. Information was requested on the calibration standards used by laboratories, but only five laboratories used standards where the purity of the standard had been certified. Evaluation of data Summary statistics were calculated for each laboratory, which consisted of the mean and standard deviation for each set of results on each of the three solutions. Additionally, the mean of the laboratory means, the standard deviation of the distribution of laboratory means and the 99% confidence interval of the mean of laboratory means were also calculated and are shown in Table 4. Pooling of data was not used, on a point of principle, as laboratories were not using the same method, even when the general approach (e.g., dichromate titration) was similar. An example plot of the results for the solution 50 is given as Fig. 1. Copies of the full set of results are available from the authors. Table 5 compares the mean of the laboratory means (allmethods basis) with the ethanol concentration values calculated Analyst, 1999, 124,

4 Table 4 Summary of results 99% confidence Mean of Standard interval of laboratory deviation of mean of Analytical means/mg laboratory laboratory method n per 100 ml means means Solution 50 : prepared concentration = ± 0.09 mg per 100 ml a All methods Potassium dichromate GC (non-headspace) GC (headspace) Infrared Enzyme Solution 80 : prepared concentration = ± 0.14 mg per 100 ml a All methods Potassium dichromate GC (non-headspace) GC (headspace) Infrared Enzyme Solution 200 : prepared concentration = ± 0.33 mg per 100 ml a All methods Potassium dichromate GC (non-headspace) GC (headspace) Infrared Enzyme a The uncertainty values quoted for the prepared concentrations are expanded uncertainties with a coverage factor of 3. The prepared concentration values apply at ambient temperatures of 20 ± 2 C. Fig. 1 CITAC forensic ethanol solution 50. Results for all methods. Table 5 Prepared and analysed concentrations of the CITAC forensic ethanol solutions Prepared concentration/ Mean of mg per 100 ml at laboratory means Solution 20 ± 2 C a (all methods) ± ± ± ± ± ± 3.43 a For the prepared concentration values, the uncertainties are expanded uncertainties calculated using a coverage factor of 3. For the mean of laboratory means, the uncertainties are 99% confidence limits. from the accurately measured quantities of ethanol and water used in the preparation of the solutions. The agreement between the prepared value and the all-methods mean value is seen to be within 0.1% for the 50 and 80 solutions, while the agreement is within 0.6% for the 200 solution. Discussion Preparation of the solutions There was close agreement between the gravimetrically prepared concentration values and the analysed concentration values, based on the combined data for all the methods used in the study (see Table 5). These observations, together with the detailed and careful approach that was taken in the gravimetric preparation of the solutions, were considered to provide justification for adopting the gravimetrically prepared concentrations and their uncertainties as the certified values for each of the three solutions. Precision of the analytical methods used From the data presented in Table 4, the following estimates of between-laboratory coefficients of variation (CVs) have been calculated. It is recognized that the figures quoted are not true CVs for the method concerned, since each laboratory used its own version of the particular method type. Also, the figures inevitably contain some contribution due to within-laboratory variations. However, the latter are considered to be relatively minor and the data in Table 6 are presented to give a simple, general, comparative picture of the performance of the different methods. Thus, of the three techniques, methods based on potassium dichromate oxidation and titration appear to offer the best reproducibility (i.e., typical between-laboratory CV = 1.1%), while the GC methods, particularly the GC (headspace) methods, have somewhat higher CVs. It is of interest to compare the observed CV values with those predicted by application of the Horwitz function, 8 derived from data obtained from the analysis of a variety of food, drugs and agricultural materials, under routine measurement conditions. As would be expected, the results reported in the present study, obtained from analysis of a simpler system (ethanol in water) under reference measurement conditions, are significantly better than the Horwitz data. The improvement factor varies from about 35 for the dichromate method to 32 for the GC (headspace) method. The details reported for the potassium dichromate technique show that the oxidation conditions used by different laboratories ranged from heating for 2 h at 100 C to allowing to stand for 16 h at 20 C. The small between-laboratory CVs observed for the potassium dichromate procedure suggest that the oxidation step is reasonably robust, that is, the precise oxidation conditions used in this study did not appear to have any significant effect on the results for ethanol concentration ultimately obtained. Likewise, the subsequent titration step was carried out in a variety of ways, but again the actual titration procedure chosen Table 6 Between-laboratory coefficients of variation Solution Solution Solution Method Mean Dichromate 0.66% 1.46% 1.22% 1.11% GC (non-headspace) 1.30% 1.30% 2.57% 1.72% GC (headspace) 2.90% 3.00% 3.56% 3.15% Horwitz value 6% 6% 5% 6% 1126 Analyst, 1999, 124,

5 did not appear to have any significant influence on the results obtained. In the case of the GC (non-headspace) methods a variety of columns were used but the choice of column had no significant effect on the results. The GC (headspace) methods used by laboratories required a larger number of experimental variables to be specified (e.g., incubation time and temperature, volume of headspace, etc.). Variations between laboratories in the choice of these parameters may account for the higher between-laboratory CVs observed for the GC (headspace) technique. Typical within-laboratory CVs for each method studied are shown in Table 7. Thus, the methods appear to be broadly comparable in terms of their within-laboratory CVs. Exceptions are the infrared method, which was only evaluated in one laboratory but shows the best repeatability, and the GC (headspace) methods, which have a tendency to provide results with somewhat higher within-laboratory CV values. Bias of the analytical methods used The methods used by more than one laboratory (i.e., dichromate, GC (non-headspace) and GC (headspace)) were examined for bias by comparing the mean of the laboratory means calculated for each of the three solutions ( 50, 80 and 200 ) with the corresponding certified values (known from the gravimetric preparation of the solutions). The application of the t-test showed that no significant differences were observed in any instance. By applying the t-test in a reverse manner, estimates were made of the level of bias that could have been detected in the present study. The minimum level of bias detectable was calculated and the data are shown in Table 8. In the case of the GC methods, the higher between-laboratory CVs are the main reason why the minimum detectable bias in these methods is higher than for the dichromate method. Performance of laboratories Sixteen laboratories took part in the study and in most cases each laboratory analysed each of the three solutions distributed in 4-fold replicate, using their chosen analytical method. In addition, two of the 16 laboratories analysed the solutions using two methods. One laboratory, when using the potassium dichromate method, only analysed two of the three solutions (the 80 and the 200 ). Consequently, 53 sets of results were received and 53 laboratory means were calculated. Of the 53 mean results, 16 (or 30%) were within ±0.5% of the relevant certified value and 28 (or 53%) were within ±1% of the certified value. Table 9 gives further information on the deviation of the mean results from the certified value. Table 10 gives information on the magnitudes of the withinlaboratory CV values for the 53 sets of results. Thus, it is seen that 72% of the 53 sets of results had a within-laboratory CV of less than 1%. Required analytical performance for the determination of ethanol in blood samples From the data on national drink-driving legislation provided by laboratories, it was apparent that prosecution only takes place when the measured blood alcohol level exceeds the legal limit by a particular margin which takes account of measurement uncertainty. For example, the situation in the UK is that while the legal limit for a driver s blood ethanol level is 80 mg per 100 ml, prosecutions are only initiated when the measured level exceeds 87 mg per 100 ml. Thus, to enforce a legal limit of 80 mg per 100 ml by prosecuting at 87 mg per 100 ml, the standard uncertainty of the result should be 52.3 mg per 100 ml (or 2.9% of the legal limit) for a confidence level of 99%. The measured value of 87 mg per 100 ml can then be held, with 99% confidence, to exceed 80 mg per 100 ml, as the measured value exceeds the legal limit by three standard deviations. When a laboratory analyses a blood sample for ethanol, the stated uncertainty associated with the result must be evaluated in order to ensure that it meets the above requirement (i.e., 52.9% of the legal limit). In the situation where the betweenlaboratory variation of a particular method is significantly greater than the within-laboratory variation (as in the present study compare Tables 6 and 7), the between-laboratory standard deviation may be adopted as an estimate of the standard uncertainty of a result. Adopting this approach it can be seen from Table 6 that, of the methods investigated, the dichromate and GC (non-headspace) procedures meet the criterion. (As only single laboratories reported results for the enzyme and IR methods it is not possible to comment on the likely between-laboratory variations of these methods.) However, it will be appreciated that the ethanol in water samples of the present study are much simpler to analyse than blood specimens, which usually require pre-treatment prior to analysis. Thus, larger standard uncertainties than those found in the present study can be expected when real blood specimens are analysed. In this context it is of interest to note that most of the methods reported in the Wilson and Barnett paper 1 do not meet the 2.9% criterion. However, repeat measurements could lead to standard error values that do meet the criterion. Further work is required to examine these issues, but it seems likely that some ethanol in blood measurement data would not be adequate for the 87 mg per 100 ml prosecution criterion at the 99% confidence level. Required analytical performance for the certification of ethanol in water reference materials When a blood specimen is analysed for ethanol content, the uncertainty of the certified value of the ethanol in water reference material used to calibrate the analytical instrumentation will contribute to the overall measurement uncertainty. Ideally, the contribution that a well-characterized certified reference material makes to the uncertainty of a measurement should be negligible compared with the uncertainty arising from the analytical method itself. This condition applies when the Table 7 Within-laboratory coefficients of variation Typical within-laboratory CV Method Range Median Notes Dichromate 0.02% 3.58% 0.27% Based on data from 5 laboratories GC (non-headspace) 0% 5.88% 0.27% Based on data from 6 laboratories GC (headspace) 0.16% 5.54% 1.08% Based on data from 5 laboratories Infrared 0.03% 0.1% 0.05% Based on data from 1 laboratory Enzyme 0.21% 0.87% 0.84% Based on data from 1 laboratory Analyst, 1999, 124,

6 standard uncertainty of the certified concentration for a reference material is less than 1/3 of the target CV for the routine analysis. 9 Thus, the ethanol in water certified reference material should have a certified value with a standard uncertainty that is < 1% (2.9/3) of the certified value. A finding of this study is that ethanol in water reference materials can be prepared gravimetrically with a standard uncertainty (i.e., k = 1) that is typically 0.06% of the certified (gravimetrically prepared) value (see Table 2), the certified value being expressed as mg per 100 ml at 20 ± 2 C. This is more than adequate for their intended purpose. The certified value of such a solution could be considered traceable to the SI, providing the balances used in the preparation of the solution are appropriately calibrated using masses that are traceable to the kilogram and the ethanol and water are traceable to the mole. Whilst mass traceability is readily demonstrated, traceability to the mole is still under debate within the Consultative Committee for Amount of Substance (CCQM). Nonetheless, gravimetric mixtures prepared from well-characterized pure chemicals are accepted by CCQM as traceable chemical standards. 5 If, alternatively, the certification of such reference materials is to be based on actual analysis of the material, the results of this study suggest that the potassium dichromate method would be the most precise method to use (see Tables 6 and 7). In addition, the method has no detectable bias and is a primary method that does not depend on another ethanol reference material to quantify the ethanol in the aqueous solution. Potassium dichromate can be considered as a primary standard and material with a high and certified purity is available from the National Institute of Standards and Technology, USA (catalogue number SRM 136e). Certified values based on the use of this material could be regarded as being traceable to the SI. The smallest reported 3 full expanded uncertainty estimate (k = 3) for the dichromate method is equivalent to 0.42% of the certified value. By comparison, the smallest expanded uncertainty (k = 3) for a gravimetrically prepared solution is equivalent to 0.04% of the certified value, the certified value being expressed as mg per 100 ml at 20 ± 0.05 C (see Table 2). However, when using the potassium dichromate method, a between-laboratory CV of about % could be expected (see Table 6) which is significantly higher than the best reported 3 standard uncertainty referred to above (0.42%, k = 3) for the dichromate method. The upper end of the CV range (1.5%) is also in excess of the maximum acceptable uncertainty for a reference material of this type (1.4%). However, by basing the certification on (say) 4 independent analyses (e.g., by 4 laboratories), the uncertainty could be reduced to lie in the range 0.35 to 0.75%. From the data reported in this study, the GC methods appear to be less adequate for the certification of ethanol in water reference materials as their between-laboratory CVs are significantly higher, at %. Table 8 Maximum possible bias Table 9 Table 10 Method Bias does not exceed Potassium dichromate 1.1% 1.8% GC (non-headspace) 1.4% 2.8% GC (headspace) 3.6% 4.4% Deviation of mean results from the certified value Number of laboratories with % of the 53 all three mean results within results within Deviation the stated the stated from the deviation deviation from certified value from the the certified (%) value value ± ± ± ± ± ± ± ± Percentage of results with the stated within-laboratory CV Withinlaboratory CV % of the 53 sets of results with the stated CV < 0.1% 17 2 < 0.5% 57 9 < 1.0% < 1.5% < 2.0% < 3.0% < 6.0% Number of laboratories with a mean CV for all three solutions within the stated CV Certified reference materials available for use in ethanol in blood analyses Ethanol in water solutions. Ethanol in water certified reference materials are currently available from the National Institute of Standards and Technology (NIST), USA, and the Laboratory of the Government Chemist (LGC), UK. The NIST reference material (SRM 1828) comprises a series of aqueous solutions of ethanol, for which the certified values are based on the gravimetric preparation of the materials and gas chromatographic analyses of the ethanol content. By way of example, one solution has a certified ethanol concentration of g per 100 g, with an expanded uncertainty (k = 2) of g per 100 g. Thus, the standard uncertainty of this material is equivalent to 0.73% of the certified value, which lies within the acceptable range as discussed above. The LGC materials also comprise a series of aqueous solutions of ethanol, for which the certified values are ultimately based on analysis using the potassium dichromate method. By way of example, one of these materials (LGC 5401, batch 028) has a certified ethanol concentration of mg per 100 ml, with an expanded uncertainty (k = 2) of 0.55 mg per 100 ml. Thus, the standard uncertainty of this material is equivalent to 0.34% of the certified value, which complies with the requirements discussed above. Very recently a thorough examination of the uncertainty of the potassium dichromate method, as carried out at LGC, 3 has been completed, resulting in a revised estimate of the expanded uncertainty (k = 2) of 0.22 mg per 100 ml being produced for a material with a certified value of 79.7 mg per 100 ml. Thus, the revised standard uncertainty is equivalent to 0.14% of the certified value. It is of interest to note that this standard uncertainty is significantly lower than the between-laboratory CV observed for the potassium dichromate method in the current study (typically CV = 1.1%, see Table 6). This would indicate that if certification is based on interlaboratory analysis using the potassium dichromate method, a higher uncertainty could apply to the certified value ultimately obtained. However, in the present study, although 5 laboratories used methods based on dichromate oxidation, each laboratory employed its own 1128 Analyst, 1999, 124,

7 particular version of the method and detailed uncertainty budgets are not available for four of the laboratories. Potassium dichromate. For those methods employing dichromate oxidation, a potassium dichromate certified reference material is available from NIST (SRM 136e). The certified purity of this material is %, with an expanded uncertainty (k = 2) of 0.01%. Thus, the standard uncertainty of this material is equivalent to 0.005% of the certified value and as such the uncertainty of this certified reference material will make a negligible contribution to the overall uncertainty of any ethanol measurements based upon it. Use of certified reference materials in this study Of the 5 laboratories that used the potassium dichromate method, two used the potassium dichromate certified reference material available from NIST (SRM 136e). One laboratory used pure iron (BNM 001) as the ultimate standard, using it to standardize a commercially supplied potassium dichromate solution. Two other laboratories used commercially available potassium dichromate with a purity > 99.5%, but with no formally certified purity value or uncertainty. Laboratories using other methods used either pure ethanol or the certified aqueous reference materials available from NIST and LGC for instrument calibration purposes. The purity of the pure ethanol materials was not usually certified, except in one instance where ethanol supplied by Wako Chemicals was reported to be certified. Laboratories prepared calibration solutions of ethanol in water from the pure ethanol and several laboratories reported the use of verification procedures to check that the solutions had been accurately prepared. Most laboratories concluded that the uncertainty of the ethanol concentrations in their in-house prepared solutions was typically ±1%. Assuming this represents an expanded uncertainty (k = 2), it is equivalent to a standard uncertainty of 0.5%, which is within the acceptable range for calibration standards discussed above. A general observation is that only a minority of laboratories appear to be using certified reference materials (whether potassium dichromate or ethanol solutions) for these analyses. This situation is perhaps not ideal for the analysis of ethanol in blood in connection with drink-driving legislation, the results of which should preferably be traceable to well-characterized reference materials, in view of the serious legal implications of the analytical data ultimately reported. certified value. In fact, some laboratories appear to have significantly overestimated their measurement uncertainties for this determination. Laboratories 9 and 13, in particular, have reported high measurement uncertainties compared with other laboratories, although their observed deviation values are similar to those of other laboratories. Correlations of performance with laboratory features The absolute overall percentage difference from the certified value for each laboratory was computed as the mean of the percentage differences for each of the three solutions analysed, but ignoring the sign. This figure was used as a broad indicator of individual laboratory performance, with better performance being indicated by a smaller value for the absolute overall percentage difference parameter. The absolute overall percentage difference values were then tabulated in a ranked order, with the lowest value ( best performance ) at the top and the highest value ( worst performance ) at the bottom (Table 12). Under each heading in the table, the figure for a particular laboratory is flagged up in bold characters if the laboratory concerned has the particular feature described by the heading. Thus, the absolute overall percentage difference of 0.15 in the Ac- Table 11 Comparison of reported uncertainties with observed deviations (units: mg per 100 ml) Observed Reported Laboratory Method deviation uncertainty 1 GC (headspace) ±2.5 1 Enzyme ±2.5 2 Dichromate +0.2 ± Infrared ±0.4 4 GC (non-headspace) ±2.2 5 Dichromate +0.4 ±0.8 6 GC (headspace) 21.4 ±1.5 7 GC (headspace Not reported 8 GC (headspace) +4 Not reported 9 GC (non-headspace) +1.3 ±6 10 GC (non-headspace) 20.2 ± GC (non-headspace) +1.7 ± GC (headspace) ± GC (non-headspace) ± Dichromate 22.3 ± Dichromate ± Dichromate ± GC (non-headspace) 21.0 ±1.21 Uncertainty of the results reported by laboratories Laboratories were asked to provide information on the sources of uncertainty in their chosen analytical method and to provide estimates of the expanded uncertainties (k = 3) that apply to their reported results. All but two laboratories were able to provide uncertainty estimates and, by way of example, these varied from 0.4 mg per 100 ml to 10.7 mg per 100 ml for the 80 solution, which had a certified ethanol concentration of mg per 100 ml. It is of interest to compare the reported uncertainty values to the actual observed differences between the reported results and the certified values. The observed difference was calculated as the deviation of the mean of each laboratory s set of replicate results from the certified value. Table 11 shows this comparison for the 80 solution, which has a certified value of ± 0.14 mg per 100 ml. It is seen that no laboratory has significantly underestimated its analytical uncertainty, in that the deviation of the mean result from the certified value is the same as or less than the quoted uncertainty for all laboratories. Put another way, the parameter (reported result ± reported uncertainty) always encompasses the Table 12 Correlations of performance with laboratory features > 200 Use of samples > 10 years Use of in-house Accredited PT per year experience CRMs RMs Analyst, 1999, 124,

8 credited column is flagged up in bold to indicate that the laboratory reporting that figure was accredited for this analysis. Examination of the three columns Accredited, PT and > 10 years experience appears to show a fairly uniform distribution of flagged results down the full length of the column, suggesting that these features (i.e., whether a laboratory is accredited, etc.) have no significant effect on laboratory performance. If the effect on laboratory performance was significant, it would be expected that the flagged results would occur predominantly in the top area of the column. The column > 200 samples per year suggests, if anything, that laboratories handling relatively few samples per annum perform better than those dealing with a larger sample throughput, as the flagged results are predominantly towards the bottom of the column. The last two columns indicate that those laboratories that used either purchased CRMs or in-house RMs performed no better than laboratories that did not use RMs, and there is some indication that they performed less well. However, it should be appreciated that random statistical fluctuations can affect ranked lists of this sort, to the extent that the precise position of a data point within the list (i.e., top, middle or bottom) can be due purely to statistical chance. Consequently, the interpretation of findings of this sort should be approached with an appropriate degree of caution. Conclusion 1. The most accurate method of certifying forensic ethanol in water standards is by gravimetric mixing of well-characterized pure ethanol and water [U(k3) 0.04% relative to the certified value, the certified value being expressed as mg per 100 ml at a temperature of 20 C ± 0.05 C]. This may be compared with the dichromate titration, where a full uncertainty budget has shown 3 that the uncertainty is an order of magnitude greater [U(k3) 0.42% relative]. 2. It is possible to calculate the acceptable uncertainty of ethanol in blood measurements required to enable reliable decisions to be made concerning compliance with legal limits for drink-driving. Using the UK as an example, a method CV of 2.9% would be acceptable where a single measurement was used, which, in turn, would require ethanol in water standards with a standard uncertainty that is < 1% of the certified value. 3. Ethanol in water standards (certified reference materials) are available from NIST and LGC and the uncertainty of the certified values complies with the above requirement. 4. Only a minority of laboratories used certified reference materials (whether potassium dichromate or ethanol solutions) as standards for the calibration or validation of their measurements. Improvement in measurement traceability requires a change to this practice. 5. Several laboratories prepared their own ethanol in water standards and some, but not all, verified the concentration values by analysis. Such verification is highly desirable, and whilst information on within-laboratory CVs (e.g., Table 7) showed that the analytical data typically obtained are of adequate precision for the purpose, some of the CVs reported suggest that the data may not always be capable of establishing the ethanol concentration with the required standard uncertainty of < 1.4% relative. 6. Although outside the scope of this study, it could be concluded from literature data 1 that the uncertainty reported for routine measurements of ethanol in blood carried out in clinical laboratories could be significantly greater than that required to yield consistently reliable prosecution decisions in marginal cases. This issue may not apply to measurements made in forensic laboratories but warrants further investigation. 7. Only 6 out of 16 laboratories fully met the study target of the results for all three solutions being within ±1% of the certified value and only 8 laboratories were within ±2% of the Table 13 Measurement/material certified values. All laboratories were within ±7% of the certified values. 8. More than half of the reported sets of replicate measurements had a within-laboratory CV of 50.5% and 94% were 2%. 9. No laboratory underestimated their measurement uncertainty, but in some cases the estimate was considerably greater than the actual uncertainty indicated by this study. In all cases the uncertainty estimates were considerably greater than that estimated by LGC for the dichromate method. 10. The precision (between-laboratory CV) of the measurements varied with the method, within the range 1.1% for the dichromate method, to 3.2% for the GC (headspace) method. 11. No significant bias was detected in any of the methods studied. For example, from the data obtained it was concluded that any bias in the dichromate method did not exceed 1.1%. 12. There are considerable differences between the full standard uncertainty estimate obtained by LGC for the dichromate method 3 and the observed interlaboratory coefficient of variation data for the dichromate method, as illustrated in Table 13 for the 80 solution. For comparison purposes, this table also shows the estimated uncertainty of the ethanol concentration of the gravimetrically prepared 80 solution. 13. From the limited amount of data available it is not possible to discern any correlation between good analytical performance and specific features of a particular laboratory, such as accreditation, participation in PT schemes, number of samples analysed per year or number of years experience of undertaking the measurements. Acknowledgement Elements of the work described in this report were supported under contract with the Department of Trade and Industry as part of the National Measurement System Valid Analytical Measurement Programme. The authors would also like to thank colleagues at LGC and participating laboratories who undertook the experimental work. References Standard uncertainty data Gravimetrically prepared standard. Certified value mg per 100 ml at 20 ± 0.05 C Standard uncertainty/mg per 100 ml Notes 0.01 U(k1) LGC estimate (from this study) Dichromate titration 0.11 U(k1) LGC estimate (from ref. 3) Dichromate titration 1.21 Between-laboratory CV = 1.46% (from this study) Dichromate titration 0.22 Within-laboratory CV = 0.27% (from this study) 1 J. F. Wilson and K. Barnett, Ann. Clin. Biochem., 1995, 32, B. King, Metrologia, 1997, 34, V. J. Barwick, S. L. R. Ellison and S. Burke, Case Studies of Error and Uncertainty. The Certification of Forensic Ethanol Standards, LGC VAM Report, LGC, Teddington, UK, R. Walker, Accred. Qual. Assur., 1997, 2, R. Kaarls and T. J. Quinn, Metrologia, 1997, 34, 1. 6 B. King, Accred. Qual. Assur., 1997, 2, Quantifying Uncertainty in Analytical Measurement, Eurachem Guide, LGC, Teddington, UK, W. Horwitz and R. Albert, Anal. Proc., 1987, 24, M. Thompson and R. Wood, J. AOAC Int., 1993, 76, 926. Paper 9/00487D 1130 Analyst, 1999, 124,