Achieving confidence in trace element analysis for nuclear forensic purposes: ICP-MS measurements using external calibration
- 129 Downloads
Abstract
In this work, problems arising from performing trace element analysis using inductively coupled plasma—mass spectrometry with low measurement uncertainties are addressed. It is shown that some reference materials certified for massic concentration of lanthanides may have either deviating concentrations or underestimated measurement uncertainties. It is also shown that the choice of methods for sample preparation and linear regression to perform external calibration is affecting the outcome of the measurement results and their uncertainties. The results show that, from the selection of methods investigated in this work, the lowest measurement uncertainties can be achieved by using weighted linear regression to evaluate the calibration function and gravimetric dilutions of samples.
Keywords
External calibration Quality control Nuclear forensics Trace element analysisIntroduction
Nuclear forensic science is a discipline that deals with providing information regarding chemical and physical characteristics of material connected to criminal investigations concerning e.g. illicit use of nuclear- and other radioactive material. In nuclear forensics, it is very important that the results from measurements, used for the interpretation of the materials in an investigation, are precise and accurate to stand up in court. Incorrectly estimated measurement uncertainties may lead a comparison between different materials or a comparison between a material and a nuclear forensic library to false conclusions, which at the end may result in wrong decisions. If the measurement uncertainties are underestimated, the comparison may give the result that an investigated material is different from another material or an entry in a nuclear forensic library even though they are similar. On the other hand, a measured material with an overestimated measurement uncertainty may not be possible to distinguish from another material even though they, in fact, are different. Therefore, it is imperative in nuclear forensics that the evaluated measurement uncertainty is as low as possible for a given measurement technique, but still accurate, to sharpen comparisons when materials are similar. For example, isotopic composition of uranium may be performed at an uncertainty level of 1–2% using gamma spectrometry [1] while mass spectrometric measurements using thermal ionization mass spectrometry (TIMS) may reach uncertainty levels as low as 0.1% [2] Even so, the uncertainty of the gamma spectrometric measurement may be adequate if compared materials are different enough.
One important characteristic in nuclear material are trace elements either as contamination from the production process or from the ore [3, 4]. This signature may be used, together with other characteristics such as isotopic and molecular composition [5, 6] to identify the origin of the material or what type of processes the material has undergone. [7, 8]. For example, rare earth elements have previously been used to determine the origin of uranium [3, 9]. Another application is to use trace elements to compare different materials in order to establish a possible common origin or history.
A suitable technique for quantification of trace elements is inductively coupled plasma mass spectrometry (ICP-MS). Depending on the analyte, the quantification can be performed in different ways. In general, isotope dilution is known to provide the lowest measurement uncertainties [10]. However, for many elements, isotopically enriched reference materials are rare and for elements with single isotopes, isotope dilution is not possible. Standard addition is another technique that can be used, but require tedious measurements as each sample requires many measurements in order to obtain a result. Therefore, the most widely used technique for quantification is external calibration.
To achieve confidence in measurement results, quality assurance is essential. Quality control (QC) during measurement assures the instrumental status and the laboratory proficiency. In quantification of elements, a common QC is the use of a QC sample consisting of a certified reference material with a known and certified concentration. Preferably, this material should be of different supplier than the reference material used for the calibration. If the measured concentration of the control sample deviates from the certified concentration this deviation needs to be handled. The most common way of handling deviating results from QC samples is to discard the measurement sequence due to some identified instrumental or sample preparation problem. However, if the deviation remains during multiple measurements and no cause of the deviation can be identified, the anomaly may need to be treated differently. Kessel et al. [11] approached this problem in a similar context by increasing the measurement uncertainty when replicates of the same sample deviated in measured concentration. Another important aspect in performing accurate measurements is the evaluation of the measurement data. It is important that the choice of evaluation method provides accurate results and measurement uncertainties.
In this study, three certified reference materials of different origins have been measured using sector field ICP-MS (ICP-SF-MS) to show that, when attempting to minimize the measurement uncertainties, there are indications that the certified reference materials may not be, within the stated uncertainty, accurate in terms of concentrations. The study showed that the choice of linear regression method and method of sample preparation affects the quality of the measurement results at this level of uncertainty. The study was performed on the lanthanide series but the discussion would be transferable to any element measurable using e.g. ICP-MS.
Experimental
Instrumentation
Instrumental and measurement settings
Forward power (W) | 1250 |
Cool gas flow (L min^{−1}) | 16 |
Auxiliary gas flow (L min^{−1}) | 0.7 |
Nebulizer gas flow (L min^{−1}) | 1.1 |
Resolution | 300 |
Detection mode | Counting |
Runs and passes | 100 × 1 |
Mass window | 5% |
Samples per peak | 100 |
Measured isotopes | ^{103}Rh, ^{139}La, ^{140}Ce, ^{141}Pr, ^{146}Nd, ^{147}Sm, ^{153}Eu, ^{157}Gd, ^{159}Tb, ^{163}Dy, ^{165}Ho, ^{167}Er, ^{169}Tm, ^{174}Yb, ^{175}Lu |
Standard solutions
Three different certified reference materials (CRM) were measured. The standard solutions were Periodic Table Mix 3 for ICP (Sigma Aldrich, Buchs, Switzerland), Spectrascan (Spectrascan, Inorganic Ventures, Christiansburg, USA) and CPAchem (CPAchem Ltd, Stara Zagora, Bulgaria), all certified by mass and traceable to NIST. The measurement uncertainties varied between 0.2 and 0.8%, k = 2, depending on analytes and brands. One of the standard solutions was used as the calibration standard and the other two standard solutions were used as QC samples. The standard solutions were diluted using MQ-water and in-house sub-boiled nitric acid to suitable concentrations. The calibration standard was diluted and measured at concentrations 0 pg g^{−1}, 500 pg g^{−1}, 1000 pg g^{−1}, 1500 pg g^{−1}, 2000 pg g^{−1} and 2500 pg g^{−1}. The standard solutions used as QC samples were diluted to 100 pg g^{−1} and 1000 pg g^{−1}. Rhodium was used as internal standard according to [3], and was added to all measured samples to a concentration of 1 ng g^{−1}. All sample preparations were performed gravimetrically in order to reduce uncertainties compared to volumetric additions. However, uncertainty modelling was also done in order to compare volumetric and gravimetric additions from an uncertainty perspective. The analytical balance used in this work was a Mettler Toledo AX204 (Columbus, Ohio, US) with an uncertainty of 0.3 mg. It should be noted that in the case of measuring real samples, the calibration solutions as well as the QC samples should be matched to have the same matrix as the samples to be quantified. Preferentially, the QC samples should originate from a certified reference material of the same composition as the unknown samples.
Data evaluation
I_{corr,i,j} is the intensity for isotope j in sample i corrected for internal standard, I_{i,j} is the dead-time corrected intensity of isotope j in sample i, I_{IS,i} and I_{IS,blk} are the dead-time corrected intensities of the internal standard in sample i and the blank sample and m_{IS,i} and m_{IS,blk} are the masses of the added internal standard in sample i and the blank sample.
All uncertainties were evaluated according to ISO GUM [18] using the software GUM Workbench Pro (Metrodata GmbH, Weil am Rhein, Germany). All uncertainties are, unless stated otherwise, presented with a coverage factor k = 2, corresponding to an approximate 95% confidence level.
Results and discussion
Considerations for linear regression
Performing different linear regressions on the same calibration data provides the opportunity to evaluate the adequacy of the methods. For OLS to be valid there are a number of conditions that need to be fulfilled. Two important conditions for using OLS in calibration is homoscedasticity in the variance of the dependant variable (in this case I_{corr,i,j}) and that the variance in the independent variable (c_{i,j}) is zero or very small compared to the variance in the dependant variable. The homoscedasticity condition implies that the absolute standard deviation of each point I_{corr,i,j} is constant throughout the calibration interval. Even though this is rarely true in many analytical methods, OLS is commonly used for constructing calibration curves [19]. In ICP-MS, the expected signal variance is similar in relative measures with the exception that the uncertainty is relatively higher close to zero. Since OLS gives each point in the calibration equal importance in the regression, the calibration points closer to zero will be given less weight than they should be given, even though these points are more certain in absolute measures. Whereas OLS might work well on heteroscedastic data in the high end of the calibration, since the equal weight gives the points in the high end of the calibration unreasonably high importance to the calibration model, the lower part of the calibration tends to be badly estimated which in turn will result in severely overestimated limits of detection [20, 21, 22].
It can be discussed what weight to use in the regression. Weighting with the reciprocal variance of the y data has been the classical method for WLS [15, 23]. This is also the weight that has been used throughout this work. However, using the variance for weighting may be difficult in cases where there is only one measurement for each x. In these cases, another weight has to be used. There are a number of suggestions of different weights such as 1/y, 1/x, 1/y^{2} [24, 25]. However, there may be very little statistical difference between using 1/y and 1/x for a linear model [24] and, moreover, using a weight such as 1/x may be impossible if the measurement of the blank, x = 0, is included in the calibration [26]. Therefore, the choice of weight is an important consideration for the evaluation of the calibration function.
Detection limits
The detection limits calculated from the OLS method are orders of magnitude larger than the corresponding results using the WLS methods. This is the result from giving too little weight to calibration points close to the y axis in comparison to calibration points high up in concentration. This results in very high uncertainties in the intercept and also, in many cases, an intercept deviating immensely from the blank measurement. One consequence of this is that if for example a sample containing trace elements at 1 pg g^{−1}, the question whether this concentration is detectable or not is dependent on the type of regression applied. If the sample was evaluated using OLS the concentration would be clearly below detection limit while if the sample was evaluated using WLS there would be quantifiable amounts present even though the same sample is measured, and the same measurement data is used to evaluate the calibration curve.
Quality control samples
The results from Standard solution 2 do not show any significant differences in concentration between measured and certified values. This means that this CRM correlates well with the CRM used for calibration. The conclusion that can be drawn is that the stated concentrations of certain elements in Standard solution 1 are significantly different from those in the calibration solution, or that the uncertainty in one, or both, CRMs are underestimated.
This approach ensures that the result of the measurement of the QC sample corresponds to the certified value within uncertainties at the 95% confidence level. This methodology has previously been applied on replicate samples by Kessel et al. [11] in a similar fashion. It should be noted that if the choice was made to use the same CRM for the calibration as for the QC sample, this anomaly would not have been detected and the risk of reporting analytical results containing bias or underestimated uncertainties would be considerable.
The, in most cases, low measurement uncertainty is the result of diluting all the samples gravimetrically rather than volumetrically and using a CRM certified for mass. If the dilutions were performed volumetrically instead, assuming a combined uncertainty of 0.8%, k = 1, for volumes less than 1 ml and 0.4% for volumes greater than 5 ml, the combined uncertainty would increase to approximately 3%, see Fig. 3. The uncertainty was evaluated according to ISO 8655-6 [28] In this case, most of the uncertainty originates from the uncertainty in the addition of the internal standard and the uncertainty of the slope of the calibration function. In the case of volumetric sample preparation, there was no need for the extra uncertainty, u(δ), for any element at the 1 ng g^{−1} level.
Of course, the easy option to evaluate the measurement results is to use the software of the ICP-MS instrument. This method, however, does not provide the full picture of the uncertainty estimation, i.e. it is not fully transparent to the analyst. The software does not ask for any uncertainty of the dead-time, which, even though it may be small, may affect the overall uncertainty at high count rates. Further, the software does not include any uncertainty from the linear regression into the calculations of the sample concentration and does not give any room for corrections due to addition of the internal standard, which, at least for volumetric additions, is a substantial part of the uncertainty. The uncertainty provided by the software is based on the standard deviation (not the standard deviation of the mean) of the calculated concentrations for each sweep, which basically is the uncertainty of the blank subtracted measured intensity. These uncertainties are, in general, larger than the uncertainties from the volumetrically prepared samples and the uncertainties are, in fact, evaluated on the wrong assumptions, including uncertainties that can be made smaller and leaving out uncertainties that may be significant.
An example of when this methodology has been applied to measurements of trace elements in a uranium matrix can be found in another published paper [30].
Conclusions
Since nuclear forensic evidence, like all evidence presented in a court of law, need to be defensible, it is important that the results and the attached uncertainties are correctly evaluated. This work shows that to perform accurate and precise measurements of elements using ICP-MS, the data evaluation should be made manually with careful considerations regarding sample preparation and choice of regression method prior to performing the measurements, to be able to retrieve correct information from the measurements.
In this paper, it is shown that gravimetric sample preparation is preferred over volumetric sample preparation to achieve the lowest measurement uncertainties and that OLS provides large measurement uncertainties at low concentrations and unrealistically high detection limits. However, depending on the purpose and thus the requirements of the measurement, work effort might be saved if volumetric sample preparations are done but then on the cost of higher measurement uncertainties.
The study also shows that careful quality control is imperative to measurements at this uncertainty level. The risk of biases due to inconsistencies in the certified reference materials needs to be carefully monitored and attended. In this study, the bias was addressed by adding an extra uncertainty to the calculated concentration since it was not possible to know which of the certified reference materials that was deviating either in the value of the certified concentration or in its uncertainty. The discrepancy between the deviating CRMs would not have been observed if volumetric sample preparation had been done.
Notes
Acknowledgements
Open access funding provided by Swedish Defence Research Agency. The Swedish Civil Contingencies Agency, Project No. B40095, is gratefully acknowledged for funding this work. Thanks to Leif Persson, Department of Mathematics and Mathematical Statistics, Umeå University, for valuable comments.
References
- 1.Abousahl S, Michiels A, Bickel M, Gunnink R, Verplancke J (1996) Applicability and limits of the MGAU code for the determination of the enrichment of uranium samples. Nucl Instrum Methods Phys Res A 368:443–448CrossRefGoogle Scholar
- 2.Bürger S, Essex RM, Mathew KJ, Richter S, Thomas RB (2010) Implementation of guide to the expression of uncertainty in measurement (GUM) to multi-collector TIMS uranium isotope ratio metrology. Int J Mass Spectrom 294:65–76CrossRefGoogle Scholar
- 3.Varga Z, Wallenius M, Mayer K (2010) Origin assessment of uranium ore concentrates based on their rare-earth elemental impurity pattern. Radiochim Acta 98:771–778CrossRefGoogle Scholar
- 4.Spano TL, Simonetti A, Corcoran L, Smith PH, Lewis SR, Burns PC (2018) Comparative chemical and structural analyses of two uranium dioxide fuel pellets. J Nucl Mater 518:149–161CrossRefGoogle Scholar
- 5.Mayer K, Wallenius M, Varga Z (2015) Interviewing a silent (Radioactive) witness through nuclear forensic analysis. Anal Chem 87:11605–11610CrossRefGoogle Scholar
- 6.Kristo MJ, Gaffney AM, Marks N, Knight K, Cassata WS, Hutcheon ID (2016) Nuclear forensic science: analysis of nuclear material out of regulatory control. Annu Rev Earth Planet Sci 44:555–579CrossRefGoogle Scholar
- 7.Varga Z, Krajko J, Peńkin M, Novák M, Eke Z, Wallenius M, Mayer K (2017) Identification of uranium signatures relevant for nuclear safeguards and forensics. J Radioanal Nucl Chem 312:639–654CrossRefGoogle Scholar
- 8.Healey G, Button P (2013) Change in impurities observed during the refining and conversion processes. Esarda Bull 49:57–65Google Scholar
- 9.Mercadier J, Cuney M, Lach P, Boiron M-C, Bonhoure J, Richard A, Leisen M, Kister P (2011) Origin of uranium deposits revealed by their rare earth element signature. Terra Nova 23:264–269CrossRefGoogle Scholar
- 10.Vogl J, Pritzkow W (2010) Isotope dilution mass spectrometry—a primary method of measurement and its role for RM certification. J Meter Soc India 25:135–164Google Scholar
- 11.Kessel R, Berglund M, Wellum R (2008) Application of consistency checking to evaluation of uncertainty in multiple replicate measurements. Accred Qual Assur 13:293–298CrossRefGoogle Scholar
- 12.Dulski P (1994) Interferences of oxide, hydroxide and chloride analyte species in the determination of rare earth elements in geological samples by inductively coupled plasma-mass spectrometry. Fresenius J Anal Chem 350:194–203CrossRefGoogle Scholar
- 13.Longerich H, Fryer B, Strong D, Kantipuly C (1987) Effects of operating conditions on the determination of the rare earth elements by inductively coupled plasma-mass spectrometry (ICP-MS). Spectrochim Acta 42B:75–92CrossRefGoogle Scholar
- 14.Appelblad P, Baxter D (2000) A model for calculating dead-time and mass discrimination correction factors from inductively coupled plasma mass spectrometry calibration curves. J Anal At Spectrom 15:557–560CrossRefGoogle Scholar
- 15.Sayago A, Asuero A (2004) Fitting straight lines with replicated observations by linear regression: part II. Test Homog Var Crit Rev Anal Chem 34:133–146CrossRefGoogle Scholar
- 16.Miller JN, Miller JC (2010) Statistics and chemometrics for analytical chemistry, 6th edn. Pearson Education Limited, Harlow. ISBN 978-0-273-73042-2Google Scholar
- 17.ISO 13528:2015 (2015) Statistical methods for use in proficiency testing by interlaboratory comparison. The International Organization for Standardization, GenevaGoogle Scholar
- 18.ISO/IEC Guide 98-3:2008 (2008) Uncertainty of measurement—Part 3: guide to the expression of uncertainty in measurement. The International Organization for Standardization, GenevaGoogle Scholar
- 19.Raposo F (2016) Evaluation of analytical calibration based on least-squares linear regression for instrumental techniques: a tutorial review. Trends Anal Chem 77:167–185CrossRefGoogle Scholar
- 20.Zorn ME, Gibbons RD, Sonzogni WC (1997) Weighted least-squares approach to calculating limits of detection and quantification by modeling variability as a function of concentration. Anal Chem 69:3069–3075CrossRefGoogle Scholar
- 21.Rajaković LV, Marković DD, Rajaković-Ognjanović VN, Antanasijević DZ (2012) Review: the approaches for estimation of limit of detection for ICP-MS trace analysis of arsenic. Talanta 102:79–87CrossRefGoogle Scholar
- 22.del Río Bocio F J, Riu J, Boqué R, Rius FX (2003) Limits of detection in linear regression with errors in the concentration. J Chemom 17:413–421CrossRefGoogle Scholar
- 23.Deming WE (1964) Statistical adjustment of data. Dover, New YorkGoogle Scholar
- 24.Almeida AM, Castel-Branco MM, Falcão AC (2002) Linear regression for calibration lines revisited: weighting schemes for bioanalytical methods. J Chromatogr B 774:215–222CrossRefGoogle Scholar
- 25.Asuero AG, González G (2007) Fitting straight lines with replicated observations by linear regression. III. Weight Data Crit Rev Anal Chem 37:143–172CrossRefGoogle Scholar
- 26.Tellinghuisen J (2008) Weighted least squares in calibration: the problem with using “quality coefficients” to select weighting formulas. J Chromatogr B 872:162–166CrossRefGoogle Scholar
- 27.ISO Guide 33:2015 (2015) Reference materials—good practice in using reference materials. The International Organization for Standardization, GenevaGoogle Scholar
- 28.ISO 8655–6:2002 (2002) Piston-operated volumetric apparatus—Part 6: gravimetric methods for the determination of measurement error, the International Organization for Standardization. Switzerland, GenevaGoogle Scholar
- 29.Ketkar SN, Bzik TJ (2000) Calibration of analytical instruments. Impact of nonconstant variance in calibration data. Anal Chem 72:4762–4765CrossRefGoogle Scholar
- 30.Vesterlund A, Ramebäck H (2019) Avoiding polyatomic interferences in measurements of lanthanides in uranium material for nuclear forensic purposes. J Radioanal Nucl Chem 321:723–731CrossRefGoogle Scholar
Copyright information
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.