You are on page 1of 14

RESIDUES AND TRACE ELEMENTS

Ion Chromatographic Method for Dissolved Hexavalent


Chromium in Drinking Water, Groundwater, and Industrial
Wastewater Effluents: Collaborative Study
EDGELL ET AL.: JOURNAL OF AOAC INTERNATIONAL VOL. 77, NO. 4, 1994
KENNETH W. EDGELL
Bionetics Corp., 16 Triangle Park Dr, Cincinnati, OH 45246
JAMES E. LONGBOTTOM
U.S. Environmental Protection Agency, 26 W. Martin Luther King Dr, Cincinnati, OH 45268
ROBERT J. JOYCE
Dionex Corp., 501 Mercury Dr, Sunnyvale, CA 94086

Collaborators: E. Arar; T. Arnold; J. Capito; R. Dovi; D. Dragotta; P. Grohse; P. Jackson; R. Joyce; B. Mitchell; J. Northington;
W. Pratt; R. Roehl; C. Scutt; J. Sherwell; C. Shoop; T. Surveski; D. Tucker; V. White; F. Williams; W. Wolf; J. Stanley

A collaborative study was conducted by the U.S. 86 statistical procedures and a U.S. EPA computer
Environmental Protection Agency (U.S. EPA) and program. U.S. EPA method 218.6 and the equivalent
the American Society for Testing Materials (ASTM) ASTM method were judged acceptable for the
on an ion chromatographic method for the determi- measurement of hexavalent chromium concentra-
nation of hexavalent chromium (U.S. EPA method tions at 1–1000 µg Cr(VI)/L. The study found that
218.6, and the ASTM equivalent method). This the use of a single linear calibration extending over
study was designed to determine the mean recov- 3 orders of magnitude yielded biased results at the
ery and precision of analyses for hexavalent chro- very lowest concentration levels. Shorter range cali-
mium in reagent water, drinking water, groundwa- bration curves yielded more accurate results.
ter, and industrial wastewaters. The study design Analysis of variance (ANOVA) tests indicated that
was based on Youden’s nonreplicate plan for col- method performance was significantly different be-
laborative studies of analytical methods. The test tween the reagent water matrix and the other ma-
waters were spiked with hexavalent chromium at trixes used. The recovery of Cr(VI) from the other
8 concentration levels, prepared as 4 Youden pairs. matrixes was lower at all concentration levels with
A fifth Youden concentration pair was also includ- slightly less precision when compared with the re-
ed to determine method performance close to the agent water data set. For reagent water, the mean
method detection limit. Twenty-one laboratories recovery and the overall and single-analyst relative
were instructed to filter their test waters through a standard deviations were 105%, 7.8% and 3.9%, re-
0.45 µm filter and to adjust the pH of the filtrate to spectively. For the other matrixes, the same values
9–9.5 with an ammonium sulfate/ammonium hy- were 96.7%, 11.9% and 6.3%, respectively. The
droxide buffer solution before spiking with the method was adopted first action by AOAC INTER-
hexavalent chromium concentrates. A known vol- NATIONAL.
ume, 50–250 µL, was injected into an ion chroma-
tograph which separated the Cr(VI), as CrO42−−, on
an anion exchange column. After separation, the exavalent chromium salts are used extensively in the
Cr(VI) was derivatized with diphenylcarbazide and
the colored complex was detected at 530 nm. The
submitted data were corrected for background con-
H metal plating and leather industries and in the manufac-
ture of paints, dyes, explosives, and ceramics. Chro-
mate compounds are also added to cooling water for corrosion
centrations and analyzed by applying ASTM D-2777- control. As a result, chromium salts are found in the waste-
waters of many industries, potentially contaminating drinking
Submitted for publication July 23, 1992. water supplies. The hexavalent chromium concentration of the
The recommendation was approved by the Committee on Environment- United States drinking waters varies between 3 and 40 µg/L
al Quality, and was adopted by the Official Methods Board of the Associa-
tion. See “Changes in Official Methods of Analysis” (1994) J. AOAC Int.
with a mean of 3.2 µg/L (1). Since hexavalent chromium is
77, Jan/Feb issue, and “Official Methods Board Actions” (1993) The Ref- toxic to humans, animals, and aquatic life (“Health Effects As-
eree, 17, July issue. sessment for Hexavalent Chromium” EPA/540/1-86/019,
1986), and is a suspected carcinogen, the U.S. Environmental Collaborative Study
Protection Agency (U.S. EPA) is very interested in monitoring
the levels of hexavalent chromium in drinking water supplies. The study design was based on Youden’s nonreplicate de-
Sensitive and highly selective colorimetric methods for the sign for collaborative evaluation of overall precision, single-
determination of hexavalent chromium are well established (1, analyst precision, and mean recovery for analytical methods (3,
2); when these methods are used however, matrix interferences 4). Two samples differing in analyte concentration by approxi-
may produce inaccurate results. To this end, U.S. EPA has de- mately 2 standard deviations were analyzed as a Youden pair to
veloped an ion chromatography method 218.6 (Arar, E.J., provide data for estimating single-analyst precision. In this
Long, S.E., & Pfaff, J.D. (1990) Determination of Dissolved study, 4 such concentration pairs were used to encompass the
Hexavalent Chromium in Drinking Water, Groundwater, and range 6–960 µg/L. The study design called for each participant
Industrial Wastewater Effluents by Ion Chromatography, Envi- to spike hexavalent chromium into reagent water and a water-
ronmental Monitoring Systems Laboratory, Office of Research of-choice. In addition to the spiked matrixes, each participant
and Development, U.S. EPA, Washington, DC), which selec- was asked to analyze both a single unspiked sample (blank) of
the reagent water and the selected matrix water. The evaluation
tively removes hexavalent chromium from these matrix inter-
of the results obtained for the spiked reagent water demon-
ferences before derivatization and colorimetric determination.
strated the proficiency of the method in recovering Cr(VI) from
The U.S. EPA Environmental Monitoring Systems Labora-
a sample free of interferences. Analyses of the spiked matrix
tory at Cincinnati, Ohio (EMSL-Cincinnati), develops or se-
waters tested the suitability of the method on representative
lects analytical methods and provides quality assurance (QA) water types and provided a basis for comparison with the re-
support for Agency programs that involve water and wastewa- sults from reagent water.
ter regulations. In EMSL-Cincinnati, the responsibility for pro- Quality control (QC) requirements consistent with the
viding QA support is assigned to the Quality Assurance Re- method were incorporated into this study to monitor the “in
search Division (QARD). One of the Division’s activities is to control” status of the various stages of the analytical process.
conduct collaborative studies, either independently or jointly Each participant was sent a concentrate for preparing a QC
with other government or not-for-profit organizations, to evalu- sample at 40 µg/L. The participants were instructed to analyze
ate analytical methods selected for the Agency’s operating pro- the QC sample after the analysis of the calibration standards
grams. U.S. EPA and the American Society for Testing and Ma- and immediately following the analysis of each set of spiked
terials (ASTM) have agreed to conduct joint collaborative matrix samples. If the sample results fell outside of the accep-
studies. ASTM, through the consensus standard process, pro- tance limits of 36–44 µg/L, the laboratory was directed to re-
vides for review of the methods, study designs, and selection of analyze the sample; if the second analysis also failed, the labo-
appropriate test materials. U.S. EPA conducts the studies and ratory was asked to recalibrate their instrument. All QC results
evaluates the data by using statistical procedures acceptable to were to be reported as part of the study data.
both organizations. A fifth IMDL Youden concentration pair was included in the
The joint study for U.S. EPA method 218.6 was conducted reagent water matrix set to evaluate method performance of
under the direction of QARD through its contract with The hexavalent chromium concentrations close to the reported
Bionetics Corp. U.S. EPA prepared and distributed sample con- MDL for EPA method 218.6. The concentration values of this
centrates, user instructions, and report forms, solicited partici- low-level Youden pair were 1.2 and 1.6 µg/L, which were 3 and
pants through an announcement in the Federal Register (55 FR 4 times higher than the MDL of 0.4 µg/L cited.
14349, April 17, 1990), and evaluated the returned data. Each participating laboratory received the method, 18 sam-
Additional volunteer laboratories were located by ASTM ple spike solutions (10 solutions for spiking into reagent water
and 8 solutions for spiking into a water-of-choice), 1 known
Committee D-19 through ion chromatography task group an-
calibration standard concentrate, 1 QC sample concentrate
nouncements, contacts with ion chromatography instrument
with true values and acceptance limits, report forms, a ques-
manufacturers, and announcements at analytical conferences.
tionnaire, and an instruction manual. Prior to distribution, the
Twenty-three laboratories participated in the study and 21 sub-
test solutions were analyzed and contents were confirmed by
mitted data.
EMSL-Cincinnati against standards freshly prepared from neat
The study had the following objectives: characterization of materials. The sample spiking concentrates, calibration con-
multi-laboratory performance of U.S. EPA method 218.6 in centrates, and QC sample concentrates were buffered to pH 9.2
terms of recovery, overall precision, and single-analyst preci- with ammonium sulfate/ammonium hydroxide and packaged
sion; comparison of the recovery and precision for hexavalent in polypropylene bottles which had been previously acid-
chromium in reagent water and various waters-of-choice; de- cleaned and rinsed with deionized water. Because the buffer
termination of the interlaboratory method detection limit solution was found to leach measurable levels of chromium
(IMDL) for Cr(VI) in reagent water and comparison of IMDL from Pyrex ampules, polypropylene bottles were used to store
with the method detection limit (MDL) in U.S. EPA method samples. The collaborating laboratories were instructed to cali-
218.6; and development of appropriate performance-based ac- brate their analytical instruments by using the calibration stand-
ceptance limits for use in quality control. ard concentrate provided, to analyze the spiked samples in strict
accord with the written method (4), and to complete the analy- summary statistics for the 8 higher concentration samples to
ses within 30 days of receipt of sample. calculate relationships between mean recovery and true con-
centration, and between precision and mean recovery in the
Treatment of Data form of linear regression equations, using the weighted least
Upon receipt, each data set was screened for data points squares technique [6; Edgell, K.W., & Jenkins, E.L. (1989)
which exceeded one-fifth to 5 times the true value. Calculations “U.S. EPA Method Study 39,” EPA/600/4-88/034, Environ-
for such data were checked, and if no error was found, the data mental Monitoring Systems Laboratory, U.S. EPA, Cincinnati,
were retained and submitted with the other study data to statis- OH]. Coefficients of determination of the weighted regression
tical processing. equations (CODw) were also calculated to evaluate the fit of the
regressions to the retained data sets. For these same data, IMVS
Each laboratory submitted a questionnaire which was re-
also tested for matrix effects between the reagent water data and
viewed for adherence to study instructions and QC require-
recovery of the spikes from the waters-of-choice. The IMDL
ments. If a laboratory failed to comply with the study instruc-
was calculated from the within-laboratory precision estimates
tions, its data were removed from the study. Although there
obtained from the IMDL samples and compared with the sin-
were a few instances where QC sample results were outside the
gle-analyst method detection limit reported for the method.
acceptance limits, the study data were not removed.
The participating laboratories were instructed to report
993.23 Dissolved Hexavalent Chromium in Drinking
background concentrations of Cr(VI) separately and not to sub-
Water, Groundwater, and Industrial Wastewater
mit “background-corrected” data. EMSL-Cincinnati per-
Effluents—Ion Chromatographic Method
formed background corrections so the effect of background
concentration on spiked concentration could be evaluated.
First Action 1993
The background-corrected data for the 4 highest Youden
pairs were grouped by water type and arranged into 8 subsets
(Applicable to determination of 1.2–960 µg/L hexavalent
as defined by the 8 different samples. Next, they were evalu-
chromium in drinking water, groundwater, and industrial
ated using the U.S. EPA computer program, Interlaboratory
wastewaters)
Method Validation Study (IMVS) designed for these studies
(Outler, E.C., & McCreery, J.H. (1982) Interlaboratory Method Method Performance:
Validation Study: Program Documentation, Battelle Columbus See Table 993.23 for method performance data.
Laboratories, Columbus, OH). IMVS replaced missing data (Caution: Hexavalent chromium is toxic and possibly car-
points with interpolated values, and “less than” and “nonde- cinogenic. See “Appendix: Laboratory Safety” on special
tect” values were converted to zero. Then outlier tests were ap- chemical hazards of carcinogens)
plied. The first outlier test was Youden’s laboratory ranking A. Principle
procedure (3) which rejected laboratories that had a consis-
tently higher or lower bias in their submitted data for a given After pH of filtered test sample is adjusted to 9–9.5, Cr(VI),
analyte compared with the other laboratories. If a bias was de- as CrO42−, is separated by ion chromatography (IC) using anion
termined, all 8 analyte values for that laboratory were rejected. exchange separator column. Cr(VI) is derivatized with
This procedure was applied to each analyte data set, for each diphenylcarbazide in post-column reactor and detected as col-
water type, at the 5% level of significance. Zeroes, interpolated ored complex at 530 nm.
values, and negative numbers were removed before further
B. Apparatus
analyses. As a final outlier test, Thompson’s test for individual
outliers (5) was applied to the retained data for a specific water (a) IC system.—Equipped with pump capable of 2000 psi
type and sample at the 5% significance level. If an individual at constant flow of 1–5 mL/min and containing no metal parts
data point was rejected on the basis of this test, it was removed in sample, eluant, or reagent flow path. Plastic pressurized
from the subset and the Thompson test was repeated on the eluant container, 1–2 L. Various sample loops, 50–250 µL.
remaining data for that water type and sample. Pressurized post-column derivatization reagent delivery mod-
The results from the IMDL Youden pair reagent water sam- ule with mixing tee and beaded mixing coil capable of constant
ples were evaluated separately from the other concentrations flow, 0.5 mL/min. Detector with low volume flow-through cell
due to limitations of the IMVS software. Data from any labo- containing no metal parts in contact with eluant flow path, ca-
ratory identified by the ranking test as an outlier during evalu- pable of 530 nm. Recorder, integrator, or computer to receive
ation of the 8 higher concentration reagent water samples were analog or digital signals for detector response (peak height or
automatically excluded from the Youden pair before statistical area) as function of time. Chromatographic conditions: eluant
evaluation. Thompson’s test for individual outliers was then flow rate, 1.5 mL/min; post-column flow rate, 0.5 mL/min;
applied to the remaining data points. typical Cr(VI) retention time, 3.8 min.
Summary statistics were calculated for the mean recovery (b) Columns.—(1) Guard column.—(For use with waste-
and overall method precision for each of the 8 higher concen- water samples.) Capable of removing strongly adsorbing or-
tration levels and each sample of the IMDL Youden concentra- ganics and particles that would damage separator column
tion pair. Single-analyst precision was calculated for each of the (Dionex IonPac NG1, Sunnyvale, CA, is suitable). (2) Sepa-
5 concentration pairs. The IMVS computer program used the rator column—Packed with high capacity anion exchange
resin capable of separating CrO42− from other sample constitu- Plot analyte response (peak height or area) versus analyte
ents (Dionex IonPac AS7 is suitable). concentration to construct calibration curve. Coefficient of de-
(c) Glassware.—Class A: volumetric flasks, graduated cyl- termination (COD), R2, for curve should be ≥0.999. Do not
inder, and calibrated pipettes. prepare calibration curve over range >2 orders of magnitude.
(d) Syringes.—10 mL, male Luer-lock, disposable. Verify continuing calibration, using laboratory blank and
(e) Syringe cartridge filters.—0.45 µm, 7.3 cm diam., calibration check standard (mid-level concentration) as surro-
polysulfone filter (Gelman Acro 50A is suitable). gate samples every 10 analyses. If concentration of analyte de-
(f) Storage bottle.—1 L, high density polypropylene. viates from true concentration by greater than ± 5%, recalibrate
(g) Sample bottle.—125 mL, high density polypropylene. and reanalyze previous 10 samples.
(h) pH meter.—Capable of determining pH to accuracy of Verify calibration stock standards, C(h), every 3 months by
± 0.03 unit. analyzing standard prepared from USEPA Certified Reference
Material, or equivalent, using acceptance limits of ± 10% of
C. Reagents established value. If sample results are outside acceptance lim-
(a) Ammonium hydroxide (NH4OH). its, perform second analysis. If second value exceeds accep-
(b) Ammonium sulfate [(NH4)2SO4]. tance limits, stop analyses, find source of problem, and recali-
(c) 1,5-Diphenylcarbazide (DPC). brate instrument.
(d) Methanol (CH3OH).—LC grade or equivalent. F. Procedure
(e) Sulfuric acid (H2SO4).—Concentrated (sp.gr. 1.84).
(f) Reagent water.—ASTM Type I water (ASTM D1193). (1) Equilibrate stored samples to ambient temperature prior
Suitable water may be obtained by passing distilled water to analysis.
through mixed bed of anion and cation exchange resins. (2) Initiate instrument operating configuration and cali-
(g) Sodium chromate tetrahydrate (Na2CrO4⋅4H2O). brate.
(h) Stock solution Cr(VI).—Accurately weigh 4.501 g (3) Draw ca 3 mL sample into new, unused syringe, B(c),
Na2CrO4⋅4H2O, (g), to nearest 0.001 g, dissolve, and dilute to and attach syringe filter, B(d). Discard first 0.5 mL through fil-
volume with reagent water, (f), in 1 L volumetric flask. Transfer ter and load 10× sample loop volume to thoroughly flush loop.
to polypropylene storage bottle, B(f). If concentrations are higher than established linear dynamic
(i) Eluant.—250 mM (NH4)2SO4, 100 mM NH4OH. Dis- range, dilute samples using reagent water, C(f), and adjust pH
solve ca 33 g (NH4)2SO4, (b), in 500 mL reagent water, (f), add if necessary.
6.5 mL NH4OH, (a), and dilute to volume with reagent water G. Calculations
in 1 L volumetric flask.
(j) Post-column reagent.—2 mM DPC, 10% v/v CH3OH, Determine peak height or peak area counts of Cr(VI) in each
1N H2SO4. Dissolve ca 0.5 g 1,5-diphenylcarbazide, (c), in sample. Compare with calibration curve to determine sample
100 mL CH3OH, (d). Add to 500 mL reagent water, (f), con- concentration in µg/L.
taining 28 mL conc. sulfuric acid, with stirring. Dilute to vol-
H. Quality Control
ume with reagent water in 1 L volumetric flask. Stable 4–
5 days. Prepare as needed. Minimum QC requirements:
(k) Buffer solution.—Dissolve ca 33 g (NH4)2SO4, (b), in (1) Determine method detection limit (MDL).—Spike 7
75 mL reagent water, (f), and add 6.5 mL NH4OH, (a). Dilute replicate portions of reagent water, C(f), at concentration 2–5X
to 100 mL with reagent water. estimated detection limit of Cr(VI), buffer to pH 9.0–9.5. Proc-
ess replicate portions through analytical method (F), then ana-
D. Sample Collection, Preservation, and Storage lyze. Calculate MDL as follows:
Filter sample through 0.45 µm filter, B(d). Wet syringe fil-
MDL = t × s
tration unit using portion of sample, then filter and collect re-
quired volume of filtrate. Add buffer, C(k), dropwise, to adjust where t = Student’s t value, 99% confidence level, n-1 degrees
sample pH to 9.0–9.5, periodically checking pH with pH meter. of freedom (t = 3.143 for 7 replicates) and s = standard devia-
Ca 10 mL sample are sufficient for 3 analyses. tion of replicate analyses.
Ship and store samples at 4°. Warm to ambient temperature MDL is concentration that can be distinguished from back-
prior to analysis. Analyze samples within 24 h of collection. ground by analytical instrumentation used.
(2) Analyze method blank daily for contamination.
E. Calibration
(3) Analyze 1 laboratory spiked water blank at 40 µg/L, for
Establish IC operating conditions, B(a). Prepare calibration each 10 samples. Sample recovery from test sample must be
standards at 3 concentration levels, bracketing anticipated con- within 36–44 µg/L, or corrective action should be taken and
centration range of samples. Dilute aliquots of stock standard, documented before analyses continue.
C(h), with reagent water, C(f), in volumetric flasks. Adjust (4) Analyze 1 spiked matrix sample for each 10 samples.
calibration solutions to pH 9.0–9.5 with buffer solution, C(k), Spike concentration should be 40 µg/L or 2× background con-
prior to final dilution. Prepare low-level calibration standards centration, whichever is higher. Calculate percent recovery of
daily. analyte spike. If results are not within 90–100%, reanalyze
spiked and unspiked samples to document sample matrix ef- the laboratory ranking test despite mean Cr(VI) recoveries
fect. across the 8 concentration levels of 89% and 92%, respectively.
Ref.: J. AOAC Int. 77, 994 (1994) In the matrix water data set, the laboratory ranking test re-
jected 3 laboratories for having systematic low bias and 2 labo-
Results and Discussion ratories for having systematic high bias. All of these 5 labora-
tories had mean Cr(VI) biases greater than ±10% of the true
Four of 8 data points from Laboratory 11, in the reagent value.
water data sets, exceeded the review limits of one-fifth to Observations from previous U.S. EPA interlaboratory stud-
5 times the true value. No calculation errors were found. The ies indicate that the laboratory ranking test in ASTM D-2777-
reagent water data set for this laboratory was later identified by 86 is very powerful in identifying laboratories with systematic
the laboratory ranking outlier test as having a systematic low biases even if the submitted data are close to the expected val-
bias and was removed from the study. ues, e.g., Laboratories 16 and 22. This is particularly true when
there is a core set of retained data which are very precise as seen
Quality Control Sample Results in this study. However, the large percentage of rejected labora-
Quality control (QC) samples, prepared at 40 µg/L, were tories should not be equated with poor method performance.
analyzed immediately after the calibration curve and after each On the contrary, this method exhibited good accuracy and pre-
of the 2 spiked matrix sets. The QC results were evaluated cision.
against acceptance limits of 36–44 µg/L (± 10%) to confirm the The “detection-level” Youden pair reagent water samples
accuracy of the calibration curve and to assess the quality of were evaluated separately from the IMVS computer program.
each spiked matrix set. All 21 QC results following the calibra- Before the statistics were calculated, the laboratories identified
tion standard analyses were grouped within the acceptance lim- as outliers by the laboratory ranking procedure were removed
its. However, 2 of 21 QC results following the reagent water from the main study reagent water data set. The majority of the
matrix and 3 of 21 QC results following the water-of-choice data labeled for removal by the laboratory ranking test were
matrix fell outside the acceptance limits. already marked as “nondetect” data by the collaborators.
Laboratory 4 reported unacceptable QC sample results from Thompson’s test for individual outliers was then performed at
both matrix sets. In each case, their QC result was high the 95% confidence interval, which identified only 1 additional
(>110%). As a result, the collaborator analyzed additional cali- outlier.
bration standards (which also reflected enhanced instrument
Method Performance
signal), recalibrated, and recalculated the results of their QC
sample and matrix spike samples with this revised curve. After Dionex instrumentation, AS-7 columns, and 250 mM am-
recalibration, the adjusted QC results were found within the monium sulfate–100 mM ammonium hydroxide eluant were
acceptance limits; however, the laboratory ranking test still used by 18 of the 21 participating laboratories. One laboratory
identified the Laboratory 4 reagent water data set as having a used a Dionex instrument with a CS-5 column with 10 mM
systematic high bias, averaging 126% recovery. pyrridine 2,6-dicarboxylic acid, and 148 mM NH4OH. The re-
Laboratory 6 reported low QC sample results from each ma- maining 2 laboratories used Waters LC instruments, IC-Pak
trix set. The collaborator decided not to reanalyze the QC sam- Anion HC columns, and 25.0 mM ammonium sulfate/10.0 mM
ple or recalibrate the instrument. The laboratory ranking proce- ammonium hydroxide as the eluant. There were no apparent
dure identified the data from both matrix sets as outliers with a differences in the returned data sets among the 3 types of instru-
systematic low bias and the data were removed from the study. mentation and conditions used by the 21 laboratories.
Laboratory 15 reported an unacceptable QC sample result No laboratories reported background Cr(VI) concentrations
after the water-of-choice matrix set. The collaborator recali- in their reagent water; however, Laboratories 6, 10, 14, and 15
brated their instrument with fresh standards, reanalyzed a QC did report background Cr(VI) concentrations in their matrix
sample that met acceptance limits, and reanalyzed the entire set water blanks. Laboratory 6 reported the highest background
of spiked water-of-choice matrix samples. The spiked matrix Cr(VI) concentration of 8.51 µg/L in their drinking water and
water data passed the ranking test and were included in the background correction made no improvement in the data. In
study. fact, the entire Laboratory 6 drinking water data set was re-
Rejection of Outliers jected by the laboratory ranking test. Laboratory 15 reported a
background concentration of 3.76 µg/L which was inconsistent
No laboratory data sets were removed from the study prior with the rest of the matrix water data results. Recovery correc-
to outlier testing. For the entire study, the IMVS computer pro- tion, using the results for the apparently contaminated unspiked
gram rejected 88 (27.0%) of the 326 data points submitted, ex- sample, was judged to be inappropriate. Laboratories 10 and
clusive of “nondetect” responses. Ninety percent of rejected 14 reported background levels around 1.5 µg/L, which was
data in this study came from the laboratory ranking test. In the compatible with their spiked matrix water data. Background
reagent water data set, the laboratory ranking test rejected corrections were made to the study data for these 2 laboratories
4 laboratories for having systematic low bias and 2 laboratories before statistical outlier testing. The background-corrected data
for having systematic high bias. Data received from Laborato- and summary statistics for the wastewater samples are pre-
ries 16 and 22 were removed from the reagent water data set by sented in Table 1.
A slightly high bias was observed in the reagent water re- Weighted least squares linear regression equations over this
coveries across all 8 concentration levels (Table 2). This bias concentration range are presented in Table 5. The coefficients
was negligible for the 4 highest concentration levels. The aver- of determination (CODw) calculated for the weighted linear re-
age percent recoveries of 101–102% were well within the ex- gression equations (Table 5) were greater than 0.999, which
pectations of the method. However, the positive bias continued confirm the suitability of these equations for estimating the
to increase as the concentration level decreased until a percent mean recovery and precision at any concentration level within
recovery of 125% was reached for the lowest concentration the study range.
level of 1.2 µg/L (Table 3). The data reported for the lowest
Effect of Water Type
concentration pair appeared to be of 2 types: data with recover-
ies averaging around 106% and data with recoveries greater A wide variety of matrix waters was used in this study. The
than 130%. treated wastewaters included 1 primary effluent, 6 secondary
An explanation for the 2 types of data was sought in the effluents, and 4 unspecified effluents. The untreated waste-
questionnaires returned by each participating laboratory. It was waters included flyash pond, laboratory, lift station, and animal
discovered that the laboratories could also be divided into storage bin wastewaters. Other waters tested included
2 groups according to the approach used in the preparation of 1 groundwater and 3 finished drinking water sources. The mis-
the calibration curve. Calculation packages supported by cellaneous matrixes included a seawater and a Toxicity Char-
Dionex Instruments provide the option to calibrate the instru- acteristic Leaching Procedure (TCLP) leachate, both of which
ment by fitting a least squares linear regression equation to a were rejected by the laboratory ranking test. The laboratory us-
series of standards. Such a regression can be useful over a nar- ing seawater had recoveries ranging 40–60% across all concen-
row concentration range, but since it tends to be dominated by tration levels. This laboratory reported that the chromate ion
the high concentration members of the data set, it becomes rela- peak had split in the seawater matrix, apparently due to the in-
tively inaccurate the further it is extended, particularly if there creased ionic strength of the sample matrix. The laboratory us-
is any nonlinearity in the calibration curve. One group of labo- ing TCLP leachate could not detect the 6.0 or 8.0 µg Cr(VI)/L
ratories used the regression option to prepare a single calibra- concentration of spike levels; had recoveries of <50% for the
tion curve in the range 1–1000 µg Cr(VI)/L; the other group 16 and 20 µg Cr(VI)/L spike levels; and had recoveries 80–
90% for the spike levels 100–960 µg Cr(VI)/L. Their reagent
used manual calibration curves or prepared shorter range cali-
water data set was very good with no data removed as outliers.
bration curves by using the linear regression option, typically
Although this laboratory had difficulties with the TCLP
1–100 µg Cr(VI)/L and 100–1000 µg Cr(VI)/L. This latter
leachate, there is insufficient data to conclude that the matrix
group reported data much closer to the true value than the group
alone was the cause of the poor data.
using a 3 orders of magnitude calibration curve (Table 4). The
The reagent water and the water-of-choice data sets were
broad-range linear regression produced relatively large errors
subjected to an analysis of variance (ANOVA) test to determine
in the estimates of analyte concentration at the lowest concen-
the effect of water type on recovery and precision. The ANOVA
tration range of the calibration curve. The statistical data (Ta- test detected a statistically significant matrix effect on spike
ble 4) illustrate these errors and suggest, by the positive bias, recovery at the 95% confidence interval for the combined
that calibration curves are not linear over 3 orders of magni- water-of-choice samples compared with the reagent water sam-
tude, and that the positive bias in the reagent water data set is ples. The effect was observed as slightly lower mean recoveries
correctable by proper calibration. from the spiked matrixes at the 4 higher concentration levels,
The overall standard deviation (sR) is the precision associ- compared with the spiked reagent water samples. As the con-
ated with measurements generated by a group of laboratories. centration levels decreased, the mean recoveries from the
Single-analyst standard deviation (sr) is the precision associ- spiked matrixes decreased steadily to a low of 91% while the
ated with performance in an individual laboratory. The mean recoveries from the spiked reagent water increased to
weighted linear regression equations (Table 5) describe method about 110%. The matrix effect on bias observed in this study is
precision as a function of mean recovery. The CODw calculated considered to be of practical significance at the lower concen-
for these weighted equations show them to be representative of trations studied.
the submitted data sets and acceptable for use in estimating the
precision at any concentration value within the range tested. Method Detection Limit
The between-laboratory method precision ranged 2–15% U.S. EPA defines the method detection limit (MDL) as the
for the individual concentration levels in reagent water and 4– minimum concentration of a substance that can be measured
26% for the same concentrations in the waters-of-choice. and reported with 99% confidence that the analyte concentra-
Above 100 µg Cr(VI)/L, the precision was less than 5% in tion is greater than zero (7). Calculation of the MDL requires
spiked reagent waters and less than 8% in spiked matrix waters. the analyses of a minimum of 7 replicate spiked reagent water
Below 20 µg Cr(VI)/L, the precisions ranged from 10 to 15% samples at a concentration of 2–5 times the estimated MDL
in spiked reagent water and 10 to 26% in spiked matrix water. reported in the method. In this study, the concentration levels
The within-laboratory precision ranged 1.5–7% for the spiked of the “detection-level” concentration pair, 1.2 and 1.6 mg/L,
reagent water matrix data and 3–11% for the spiked waters-of- were 3–4 times the single laboratory MDL of 0.4 µg/L cited for
choice data. U.S. EPA method 218.6. Of the 21 laboratories submitting data
for these samples, 5 reported “nondetect” at these spike levels from matrix waters were lower at all concentration levels when
which indicates the closeness of the samples to the MDL value. compared to recoveries of spikes from reagent water samples.
After outlier removal following AOAC guidelines (8), the re- It is recommended that linear regression calibration curves
maining 13 paired analyses (see Table 3) were used to calculate be limited to 1–2 orders of magnitude. Using least squares lin-
the within-laboratory standard deviation (sr) at the average ana- ear regression analysis to fit calibration data over 3 orders of
lyte concentration of 1.4 µg/L. The calculated sr, 0.15 µg/L, magnitude tends to produce very large biases at the lower con-
was used in the following equation to calculate MDL: centration levels. The method was very accurate and precise for
chromium Cr(VI) concentration levels as low as 1.2 µg/L, as
MDL = t(df, α = 0.99) [sr] long as a short range calibration curve (e.g., 1–100 µg/L) was
MDL = (2.65)(0.15)at 13 degrees of freedom used to quantify the sample data.
MDL = 0.40 µg/L The method study data were suitable for the development of
performance-based QC limits. It is recommended that users of
This interlaboratory MDL supports the single-laboratory this method routinely test a QC sample prepared in reagent
MDL reported in U.S. EPA method 218.6 and the number of water and compare the results with the performance-based ac-
acceptable data points reported at the 1.2 µg/L concentration ceptance limits derived from this study. The calculated QC ac-
level supports the applicability of this method to low level ceptance limits using the IMVS estimates fully supported the
Cr(VI) analyses, approaching 0.4 µg/L, as long as a short range ±10% limits imposed on the participants of this study.
calibration curve is used. The outlier procedures employed in this study, which re-
Performance-Based QC Acceptance Limits jected 27% of the submitted data, appear to have rejected a sig-
nificant amount of data that would be judged acceptable by
Since the weighted linear regression equations for precision application of normal QC acceptance criteria, and should be
were found to be representative of the submitted data sets, they further evaluated.
can be used to derive performance-based QC limits for this Two laboratories that produced data of apparently good
method. To establish performance-based QC limits for use with quality were rejected by the laboratory ranking test. Although
a QC sample, estimates of the within-laboratory standard de- the overall study rejection rate exceeds the 2/9 (22%) limit sug-
viation were calculated for Cr(VI) spike level of 40 µg using gested by AOAC guidelines (8), it does not mean that the
the reagent water weighted regression equations reported in Ta- method or the study is flawed. Because of the rejection of these
ble 5. At this concentration, the mean recovery (X) is estimated 2 laboratories, the conclusion should be made that the labora-
to be 40.0 µg/L as the slight (3%) positive bias was attributed tory ranking outlier procedure can be a little too critical when-
to correctable calibration discrepancies. The within-laboratory ever the single-analyst standard deviation is generally so small.
standard deviation (sr) at 40.0 µg/L is estimated to be It is recommended that research be conducted into the
1.22 µg/L. The QC acceptance limits, at the 99% confidence unique characteristics of seawater and TCLP leachate in rela-
level, were calculated as: tion to Cr(VI) analyses using this method. The laboratories us-
ing these 2 matrixes produced low biased data which were re-
QC acceptance limits = X ± 3sr
jected by the laboratory ranking test.
= 36.3–43.7 µg/L The MDL of the method was confirmed by the IMDL to be
These performance-based QC acceptance limits support the 0.40 µg/L. The use of interlaboratory single-analyst precision
continued use of the acceptance limits used in this study, 36– to evaluate MDLs should continue. These studies provide a
44 µg/L (90–110% recovery). Eighteen of the 21 laboratories broad-based assessment of MDLs and give a reliable estimate
had no trouble meeting the study QC acceptance limits of of how well an analytical laboratory can be expected to perform
±10%. One additional laboratory was able to meet the accep- using the method. In addition to low concentration detection-
tance limits upon recalibration and reanalysis of the QC sam- level studies in reagent water, studies should also be considered
ple. The high quality of the returned data in this study supports where practical with other matrixes (i.e. wastewater, ground-
the applicability of these acceptance limits. water, leachate, etc.) at these low levels.

Conclusion Recommendation

U.S. EPA method 218.6 was shown to be accurate and pre- On the basis of the results of this study it is recommended
cise for the analysis of hexavalent chromium in waters and vari- that the ion chromatographic method for determination of dis-
ous matrix waters in a collaborative study involving 21 labora- solved hexavalent chromium in drinking water, groundwater,
tories. The equations presented for method recovery, overall and industrial waste water effluents be adopted first action by
standard deviation, and single-analyst standard deviation can AOAC.
be used to estimate method performance at any concentration
value within the study range. Acknowledgments
The statistically significant matrix effects, identified by the
ANOVA test, were considered to be of practical significance for The authors thank the following:
the water-of-choice matrixes as a group. Recoveries of spikes Elizabeth Arar, Technology Applications, Inc., Cincinnati, OH
Timothy Arnold, American Electric Power, Columbus, OH The authors thank Edward Katz for providing analytical
John Capito, San Diego Gas & Electric, San Diego, CA support for the study and John Ortman for IMVS computer
Robert Dovi, Waste Management, Inc., Geneva, IL processing.
Dominic Dragotta, DuPont, Newark, DE
Peter Grohse, Research Triangle Institute, Research Trian- References
gle Park, NC
(1) Standard Methods for the Examination of Water and Waste-
Peter Jackson, Waters, Milford, MA water (1989), 17th Ed., APHA, AWWA, WPCF, Washington,
Robert Joyce, Dionex Corp., Sunnyvale, CA DC
Barbara Mitchell, Tennessee Valley Authority, Muscle (2) Annual Book of ASTM Standards (1990), “Chromium-Total”,
Shoals, AL D 1687-86, Vol. 11.02, Water, American Society for Testing
Jack Northington, West Coast Analytical Services, Sante Fe and Materials, Philadelphia, PA
Springs, CA (3) Youden, W.Y., & Steiner, E.M. (1975) Statistical Manual of
William Pratt, ENSECO/RMAL, Arvada, CO the AOAC, AOAC, Arlington, VA, p.88
Raimund Roehl, California Department of Health Services, (4) Annual Book of ASTM Standards (1990) “Standard Practice
for Determination of Precision and Bias of Applicable Meth-
Berkeley, CA
ods of Committee D-19 on Water”, ASTM D2777-86, Vol.
Christopher Scutt, DOFASCO, Inc., Hamilton, ON, Canada 11.01, Water (1), American Society for Testing and Materi-
John Sherwell, Radian Corp., Austin, TX als, Philadelphia, PA
Chris Shoop, Tennessee Eastman Co., Kingston, TN (5) Barnett, X., & Lewis, Y., (1984) Outliers in Statistical Data,
Thomas Surveski, Travelers Insurance Co., Hartford, CT John Wiley and Sons, New York, NY
Dave Tucker, City of San Jose WPCP, San Jose, CA (6) Draper, N., & Smith, H. (1968) Applied Regression Analysis,
Vincent White, Department of Environmental Resources, John Wiley and Sons, New York, NY
Harrisburg, PA (7) Code of Federal Regulations, Chapter 40, Part 136, appendix
B, Revision 1.11, Government Printing Office, Washington,
F.W. Williams, Manville, Littleton, CO
DC
William Wolf, Carolina Power & Light Co., New Hill, NC (8) “Guidelines for Collaborative Study Procedures to Validate
John Stanley, Midwest Research Institute, Kansas City, MO Characteristics of a Method of Analysis” (1989) J. Assoc.
Off. Anal. Chem. 72, 694-704
Table 993.23. Method performance for determination
of dissolved hexavalent chromium in drinking water,
groundwater, and industrial wastewater effluents by ion
chromatographic methoda
Mean
recovery,
Spike, µg/L µg/L sr sR RSDr, % RSDR, %

Reagent Water

1.2 1.50
1.6 1.87 0.15 0.57 9.0 33.8
6 6.68
8 8.64 0.53 1.06 6.9 13.9
16 17.4
20 21.4 0.77 2.28 4.0 11.8
100 101
140 143 3.8 4.3 3.1 3.5
800 818
960 966 12.7 21.2 1.4 2.4

Drinking water, ground water, and industrial wastewaters

6 5.63
8 7.31 0.55 1.20 8.2 17.8
16 15.1
20 19.8 1.85 1.85 10.4 10.4
100 98.9
140 138 3.3 4.8 2.8 4.0
800 796
960 944 27.1 66.6 3.1 7.6
a
Samples analyzed as Youden pairs.
Table 1. Background-corrected data and summary statistics for collaborative study to determine hexavalent
chromium in wastewater, µg Cr(VI)/L
True value

Lab. Matrix type 6.00 8.00 16.0 20.0 100 140 800 960

1 2° Effluent 6.42 8.26 15.6 20.6 103 138 773 930


2 Flyash pond 7.35 8.84 17.1 19.5 97.5 136 783 945
a
4 Seawater 2.77 3.45 8.88 8.88 5.05 83.2 459 558
b
5 Ground water 0.00b 2.36 8.53 21.0 101 142 798 952
a c c
6 Drinking water — 0.94 — — — 26.1 245 123
7 Plant effluent 7.48 9.74 18.0 21.8 99.9 148 798 906
8 Plant effluent 4.21 8.46 18.2 19.4 96.4 145 802 1002
a
9 Plant effluent 9.00 10.7 17.0 22.0 105 148 810 954
10 Animal bin 5.87 8.61 16.1 20.4 102 143 832 1019
b
11 Lab. wastewater 111b 4.76 13.9 b
10.3b b
186b 121 706 763
12 1° effluent 6.10 8.57 16.2 19.8 96.8 141 798 949
14 Cold mill waste 4.60 7.60 15.2 20.0 96.4 144 816 931
b
15 Lift station 3.28 4.64 9.14 12.2b b
69.3b 119 898 996
16 Clarifier effluent 4.73 7.18 15.0 18.8 96.7 136 756 910
17 Drinking water 5.86 7.58 15.0 18.8 98.3 135 778 927
18 2° effluent 5.78 77.70 16.7 20.4 106 147 846 1020
19 2° effluent 6.53 8.38 15.8 19.0 103 142 879 1060
a
20 TCLP leachate <2.7 2.7 5.04 9.18 81.3 119 698 863
21 2° effluent 5.46 7.87 15.8 19.0 100 134 830 959
22 2° effluent 5.18 6.43 16.0 18.1 87.7 131 649 840
a
23 Drinking water 7.01 9.34 18.6 21.5 116 162 967 1090
Mean (X) 5.63 7.31 15.1 19.8 98.9 138 796 944
Recovery, % 93.4 91.4 94.6 98.8 98.9 98.63 99.6 98.4
SR 1.17 1.91 2.70 1.01 4.36 8.39 60.6 72.1
RSDR, % 20.8 26.1 17.8 5.1 4.4 6.1 7.6 7.6
sr 0.55 1.85 3.31 27.1
RSDr, % 8.6 10.6 2.8 3.1
a
Data removed by Youden’s laboratory ranking test.
b
Data removed by Thompson’s individual outlier test.
c
Not detected.
Table 2. Raw data and summary statistics for collaborative study to determine hexavalent chromium in reagent
water, µg Cr(VI)/L
True value

Lab. 6.00 8.00 16.0 20.0 100 140 800 960

1 6.78 9.03 16.9 21.6 101 142 802 961


2 7.96 9.13 22.1 25.9 102 139 813 983
a
4 7.72 10.6 22.4 28.8 108 168 1001 1098
5 6.63 9.51 18.9 21.8 102 141 805 972
a b b
6 — 10.0 — — 70.9 98.9 289 328
7 8.96 10.4 18.9 21.5 104 144 782 948
8 5.49 7.38 16.1 18.3 93.9a 143 820 973
9 5.93 8.13 15.7 20.7 99.7 136 797 938
10 6.33 8.78 16.6 21.1 99.1 148 816 983
a
11 4.67 6.38 15.9 17.3 63.7 134 1.94 856
12 7.82 7.82 16.1 19.6 102 143 820 956
14 5.70 7.66 15.8 22.4 101 147 858 998
a
15 8.00 9.97 17.3 21.6 108 166 910 1050
a
16 1.93 7.38 15.5 19.2 99.4 140 775 934
17 6.28 8.23 15.8 19.7 102 141 807 962
18 5.83 7.62 16.2 20.1 97.1 142 816 938
19 6.05 7.87 15.9 20.6 103 149 849 1055c
20 8.00 11.3 15.4 19.1 99.4 131 796 965
21 6.03 8.09 22.3 26.8 134c c
191c 826 956
a
22 5.37 7.74 14.8 19.2 93.1 131 695 874
23 6.41 8.61 17.9 21.1 103 153 871 991

Mean (X) 6.68 8.64 17.4 21.4 101 143 818 966
Recovery, % 111 108 108 107 101 102 102 101
sR 1.03 1.10 2.25 2.31 1.91 5.52 24.3 18.5
RSDR, % 15.4 12.7 12.9 10.8 1.9 3.9 3.0 1.9
sr 0.53 0.77 3.76 12.7
RSDr, % 6.9 4.0 3.1 1.4
a
Data removed by Youden’s laboratory ranking test.
b
Not detected.
c
Data removed by Thompson’s individual outlier test.
Table 3. Raw data and summary statistics for
detection-level concentration Youden pair, µg Cr(VI)/L
Calibration True value, True value, Difference,
Lab curve range 1.20 µg/L 1.60 µg/L µg/L

1 long 1.69 2.14 0.45


2 short 1.33 1.63 0.30
a,b a,b a,b a,b
4 long — — —
5 long 2.21 3.17 0.96
a,b a,b b
6 long — 10.4b —
7 long 2.64 3.10 0.46
8 short 1.02 1.13 0.11
9 short 1.08 1.35 0.27
10 long 1.58 1.71 0.13
b
11 short 1.65b 13.6b —
12 long 1.44 1.68 0.24
c
14 long 3.80c a a
— —
a,b a,b a,b a,b
15 short — — —
b
16 long 0.68b 0.73b —
17 short 1.45 1.89 0.44
18 long 0.90 1.31 0.41
19 short 1.86 2.16 0.30
a a a a
20 long — — —
21 short 1.28 1.54 0.26
a,b a,b a,b a,b
22 long — — —
23 short 1.08 1.56 0.48
d
Mean 1.50 1.87 —
% Recovery 125 117 —
Overall sR 0.50 0.63 —
e
Single-analyst, sr — — 0.15
RSDR, % 33 34 —
f
RSDr, % — — 8.9
a
Not detected.
b
Laboratories rejected on basis of Youden laboratory ranking test
applied to data in 6–960 µg/L range were also rejected here.
c
Rejected data by Thompson’s individual outlier test.
d
Mean values calculated for 13 retained values only.
e
Single-analyst standard deviation, sr, calculated as standard
2 .
deviation of the differences/√
f
Single-analyst relative standard deviation calculated as sr / mean
of pair.
Table 4. Comparison of low-level Youden pair
summary statistics quantified using “short range”
versus “long range” calibration curves
Short range Long range
Statistic calibration curvea calibration curveb

True value,
µg/L 1.20 1.60 1.20 1.60
Number of
pairs 7 7 6 6
Mean
recovery 1.30 1.61 1.74 2.18
Recovery, % 108 101 145 136
Std deviation 0.29 0.34 0.61 0.78
RSD, % 24 21 35 36
a
Short range calibration curves typically 1–100 µg Cr(VI)/L.
b
Long range calibration curves were 1–1000 µg Cr(VI)/L.
Table 5. Weighted linear regression equations
calculated from collaborative study statistics for
chromium in reagent water and matrix watersa
Reagent water, Matrix water,
Statistic 6–960 µg/L 6–960 µg/L

Mean recovery X = 1.020C + 0.592 X = 0.989C − 0.411


Between-laboratory
standard deviation sR = 0.035X + 0.893 sR = 0.059X + 1.055
Within-laboratory
standard deviation sr = 0.021X + 0.375 sr = 0.041X + 0.393
a
X = mean recovery, µg/L; C = true value, µg/L; sR = reproducibility
standard deviation, µg/L; sr = repeatability standard deviation,
µg/L.

You might also like