Download Performance Characteristics of Five Automated CA 19

Clinical Chemistry / FIVE CA 19-9 ASSAYS
Performance Characteristics of Five Automated
CA 19-9 Assays
Sonia L. La’ulu,1 and William L. Roberts, MD, PhD2
Key Words: CA 19-9; Automated immunoassay; Method comparison; Imprecision; Linearity
DOI: 10.1309/H52VET3M6P7GYWG1
Abstract
Serum concentrations of cancer antigen (CA) 19-9
can be useful in monitoring response to therapy in
pancreatic cancer. The objective of this study was to
evaluate 5 automated CA 19-9 assays: ARCHITECT
i2000 (Abbott Diagnostics, Abbott Park, IL), ADVIA
Centaur (Bayer Diagnostics, Tarrytown, NY), UniCel
DxI 800 (Beckman Coulter, Fullerton, CA), IMMULITE
2000 (Diagnostic Products, Los Angeles, CA), and
Elecsys E170 (Roche Diagnostics, Indianapolis, IN). All
methods were evaluated for limit of detection, linearity,
imprecision, method comparison, and reference
intervals. Limit of detection results were all below 2.0
kU/L and met the manufacturers’ claims. Linearity had
deviation from target values that ranged from 4.5% to
26.7%. All methods showed acceptable imprecision
with total coefficients of variation less than 8%. Method
comparison by Passing-Bablok analysis resulted in
slopes ranging from 1.00 to 2.06 and correlation
coefficients of 0.85 to 0.98. Between 97.6% and 99.2%
of results from healthy volunteers were less than 35
kU/L. All methods show acceptable analytic
performance.
436
436
Am J Clin Pathol 2007;127:436-440
DOI: 10.1309/H52VET3M6P7GYWG1
Tumor markers are useful in the management of various
cancers. Among these, cancer antigen (CA) 19-9 is a marker
for pancreatic and colorectal carcinoma.1 However, CA 19-9
has not been shown to be a good screening test for pancreatic
cancer in asymptomatic people.2 The primary use of CA 19-9
is in serial monitoring during palliative chemotherapy in conjunction with imaging tests. Serial measurements are also useful for follow-up after potentially curative surgery.3
Measurement of CA 19-9 may also be useful for monitoring
other cancers, including gastric, hepatobiliary, hepatocellular,
breast, and ovarian cancers.4-6
CA 19-9 is a glycolipid, the sialylated form of the Lewis
blood group antigen.4 In serum, it exists as a mucin, a highmolecular-mass (200-1,000 kd) glycoprotein complex.4 CA
19-9 is synthesized by normal human pancreatic and biliary
ductular cells and by gastric, colon, endometrial, and salivary
epithelia.4 The original monoclonal antibody against CA 199 was developed from a human colon carcinoma cell line,
SW-1116.7
Formerly, immunoradiometric assays were used to test
for CA 19-9 and have mostly been replaced by automated,
nonisotopic immunoassays. Even though imprecision has
improved from the use of nonisotopic immunoassays, the
agreement of CA 19-9 results has worsened in the last few
years when the automated techniques became available.8 The
concentration of CA 19-9 in a given specimen, determined
with assays from different manufacturers, can vary owing to
differences in assay methods, antibodies used, and reagent
specificity. Although many quality assessment efforts have
been initiated, discrepancies are commonly observed.8,9 We
undertook this study to assess the performance characteristics
of 5 automated, commercially available CA 19-9 assays.
© American Society for Clinical Pathology
Clinical Chemistry / ORIGINAL ARTICLE
Materials and Methods
Limit of detection, linearity, imprecision, method comparison, and reference intervals were determined for the
ADVIA Centaur CA 19-9 (Bayer Diagnostics, Tarrytown,
NY), ARCHITECT i2000 CA 19-9XR (Abbott Diagnostics,
Abbott Park, IL), IMMULITE 2000 GI-MA (Diagnostic
Products, Los Angeles, CA), Elecsys E170 CA 19-9 (Roche
Diagnostics, Indianapolis, IN), and UniCel DxI 800 GI
Monitor (Beckman Coulter, Fullerton, CA). All assays were
performed according to manufacturers’ instructions.
The limit of detection was determined by averaging 2
separate runs with each run consisting of 10 replicates of zero
material and 3 replicates of nonzero material. The following
zero and nonzero material was used: on the ADVIA Centaur,
CA 19-9 Diluent (0 kU/L) and Calibrator L (21 kU/L); for the
ARCHITECT i2000, Calibrator A (0 kU/L) and Calibrator B
(30 kU/L); on the IMMULITE 2000, Multi-Diluent 2 (0
kU/L) and Adjustor Low (200 kU/L); for the Elecsys E170,
Diluent Universal (0 kU/L) and Calibrator 1 (24 kU/L); and on
the UniCel DxI 800, Calibrator S0 (0 kU/L) and Calibrator S2
(90 kU/L).
Specimens used for dilution linearity testing were
obtained following completion of clinical testing. Each patient
specimen was serially diluted with respective manufacturer
assay diluent to yield final concentrations of 0.39%, 0.78%,
1.56%, 3.13%, 6.25%, 12.5%, 25%, 50%, 75%, and 100%.
All dilutions were tested in duplicate.
Imprecision studies were performed by using commercially available quality control material. Two concentration
levels (L1 and L2) of lyophilized Lyphocheck Tumor Marker
Controls (Bio-Rad, Hercules, CA) were reconstituted according to the manufacturer’s package insert instructions. Multiple
bottles for each level were pooled, divided into aliquots, and
stored at 4°C until use. Control experiments were run twice a
day for 5 days in replicates of 2, using a fresh aliquot for each
run. A minimum of 2 hours separated each run for a total of
20 replicates for each control level.
Interference studies were performed as previously
described.10 Briefly, a serum pool with a nominal CA 19-9
concentration of 100 kU/L was supplemented with the following: (1) RBC hemolysate to a final hemoglobin concentration
of 1.1 g/dL (11 g/L), (2) bilirubin to a final concentration of
38.2 mg/dL (653 µmol/L), and (3) Intralipid to a final triglycerides concentration of 2,360 mg/dL (26.7 mmol/L). A deviation of more than 15% from the target CA 19-9 concentration
after correcting for dilution was considered a clinically significant degree of interference.
Method comparison was evaluated using 135 specimens
that were obtained following completion of clinical testing.
Specimens chosen contained CA 19-9 concentrations spanning the range of 1 to 1,500 kU/L as measured by the UniCel
DxI 800. Before testing, specimens were thawed, mixed thoroughly, and checked for clots. The ADVIA Centaur was used
as the comparison method because it was the first US Food
and Drug Administration–approved assay. For each method,
specimens with measured concentrations more than the analytic measurement range were diluted on-board according to
manufacturers’ instructions. The institutional review board of
the University of Utah, Salt Lake City, approved all studies
using human samples.
To evaluate reference intervals, 127 samples obtained
from apparently healthy male and female subjects who were
not taking any prescription medications were retrieved from
–70°C storage. Each sample was thawed, mixed thoroughly,
and analyzed by all methods.
EP Evaluator Release 5 software (David G. Rhoads
Associates, Kennett Square, PA) was used to calculate limit of
detection, linearity, imprecision, diagnostic concordance, and
reference intervals. Passing-Bablok and linear regression
analysis was performed using Analyse-It, version 1.71
(Analyse-It Software, Leeds, England).
Results
The limit of detection for each assay was calculated and
compared with the manufacturers’ claimed values. The
ADVIA Centaur had an average limit of detection of 0.43
kU/L with a manufacturer’s claim of 1.2 kU/L; the ARCHITECT i2000 had an average limit of detection of 0.23 kU/L
with a manufacturer’s claim of 2.0 kU/L; the IMMULITE
2000 had an average limit of detection of 0.32 kU/L with a
manufacturer’s claim of 1.0 kU/L; the Elecsys E170 had an
average limit of detection of 0.07 kU/L with a manufacturer’s
claim of 0.6 kU/L; and the UniCel DxI 800 had an average
limit of detection of 0.07 kU/L with a manufacturer’s claim of
0.8 kU/L.
Linearity for all methods was assessed ❚Table 1❚. The target value for each linearity sample was calculated based on the
samples with the lowest and highest concentrations within the
analytic measurement range for each method. The maximum
average deviation from the target recovery ranged from 4.5%
(Elecsys E170) to 26.7% (ADVIA Centaur).
Imprecision ranged from 2.8% to 8.1% for level 1 and
2.5% to 4.7% for level 2. The Elecsys E170 was the most precise for both control levels and the ARCHITECT i2000 was
the least precise for level 1 and the ARCHITECT i2000 and
IMMULITE 2000 were the least precise for level 2 ❚Table 2❚.
Interference studies were conducted for hemolysis,
icterus, and lipemia. The interference detected by the maximum concentration of each substance tested was less than
15% for all methods. Icterus is a relatively common interference in patients with pancreatic cancer owing to metastases to
Am J Clin Pathol 2007;127:436-440
© American Society for Clinical Pathology
437
DOI: 10.1309/H52VET3M6P7GYWG1
437
437
La’ulu and Roberts / FIVE CA 19-9 ASSAYS
❚Table 1❚
Summary of Dilution Linearity Data
Maximum Deviation
for Target Recovery (%)
Method*
ADVIA Centaur
ARCHITECT i2000
IMMULITE 2000
Elecsys E170
UniCel DxI 800
Concentration at Which Maximum Deviation
From Target Occurred (kU/L)
26.7
9.8
15.2
4.5
5.0
Measured Range
(kU/L)
45.25
707.6
100.7
239.2
371.5
1.4-583.1
4.7-1,045.6
3.7-950.0
5.6-996.1
7.5-1,560.0
* ADVIA
Centaur, Bayer Diagnostics, Tarrytown, NY; ARCHITECT i2000, Abbott Diagnostics, Abbott Park, IL; IMMULITE 2000, Diagnostic Products, Los Angeles, CA;
Elecsys E170, Roche Diagnostics, Indianapolis, IN; and UniCel DxI 800, Beckman Coulter, Fullerton, CA.
❚Table 2❚
Summary of Imprecision Data
Coefficient of Variation (%)
Method*/Sample
ADVIA Centaur
L1
L2
ARCHITECT i2000
L1
L2
IMMULITE 2000
L1
L2
Elecsys E170
L1
L2
UniCel DxI 800
L1
L2
*
Mean Concentration (kU/L)
Within Run
Between Run
Between Day
Total
16.23
62.50
4.6
2.8
0.0
2.3
4.9
0.0
6.7
3.6
18.38
122.55
8.1
4.1
0.0
2.1
0.0
0.0
8.1
4.7
8.31
34.72
5.6
3.8
0.0
2.8
2.8
0.0
6.3
4.7
11.57
40.04
1.4
1.3
2.2
2.1
0.8
0.0
2.8
2.5
10.65
34.10
5.6
1.3
3.0
2.3
0.7
0.0
6.3
2.6
See Table 1 for manufacturer information.
the liver and biliary tract obstruction. It is noteworthy that a
total bilirubin concentration of 38.2 mg/dL (653 µmol/L) does
not produce a significant interference for any of the methods
that were evaluated.
Method comparison revealed acceptable agreement when
compared with the ADVIA Centaur ❚Figure 1❚. PassingBablok slopes ranged from 1.00 to 2.06 with correlation coefficients ranging from 0.85 to 0.98. Statistical outliers were
retested and did not change significantly from the original
result. An average of the results from the original and repeated run was used in the final method comparison analysis.
Diagnostic concordance was evaluated using the ADVIA
Centaur as the comparison method ❚Table 3❚.
Results of reference interval studies using samples from
healthy volunteers are shown in ❚Table 4❚. Of the 5 methods,
3 use a cutoff of 35 kU/L (ADVIA Centaur, Elecsys E170, and
UniCel DxI 800) and 2 use 37 kU/L (ARCHITECT i2000 and
IMMULITE 2000). We evaluated reference intervals for 35
and 37 kU/L, and the results were identical. Table 4 shows
results evaluated at a cutoff of 35 kU/L because the majority
of methods use this cutoff. Reference intervals from healthy
volunteers resulted in 97.6% (ADVIA Centaur) to 99.2%
438
438
Am J Clin Pathol 2007;127:436-440
DOI: 10.1309/H52VET3M6P7GYWG1
(IMMULITE 2000, Elecsys E170, and UniCel DxI 800) of
specimens tested at less than 35 kU/L. The 97.5% upper reference limit ranged from 17.0 kU/L (IMMULITE 2000) to
37.1 kU/L (ADVIA Centaur).
Discussion
The limit of detection was lower than the manufacturers’
claimed values for each method and more than adequate for
clinical use on serum samples. When evaluating method linearity, if a maximum deviation from the target values of 10%
was arbitrarily used as the limit, then the ARCHITECT i2000,
Elecsys E170, and UniCel DxI 800 methods would be acceptable. Linearity studies for the ADVIA Centaur and
IMMULITE 2000 were also performed using low serum pools
as the diluent, and deviations from target values were comparable to those seen using the manufacturers’ diluent (data not
shown). Reasons for large deviations from target values, particularly for the ADVIA Centaur method, are unclear.
Deviations do not seem to be attributable to a matrix effect
from the diluent.
© American Society for Clinical Pathology
Clinical Chemistry / ORIGINAL ARTICLE
3,000
2,000
1,000
0
0
500
1,000
1,500
1,500
1,000
500
0
0
500
1,000
1,500
1,000
1,500
0
0
500
1,000
1,500
ADVIA Centaur CA 19-9 (kU/L)
ADVIA Centaur CA 19-9 (kU/L)
ADVIA Centaur CA 19-9 (kU/L)
500
Unicel Dxl 800 CA 19-9 (kU/L)
4,000
D
C
IMMULITE 2000 CA 19-9 (kU/L)
ARCHITECT i 2000 CA 19-9 (kU/L)
5,000
Elecsys E170 CA 19-9 (kU/L)
B
A
1,500
1,000
500
0
0
500
1,000
1,500
ADVIA Centaur CA 19-9 (kU/L)
❚Figure 1❚ Method comparison of CA 19-9 assays. The ADVIA Centaur was used as the comparison method. The dotted line is
the line of identity (x = y). Passing-Bablok analysis is indicated by the solid line. A, Evaluation of the ARCHITECT i2000 method.
Passing-Bablok regression analysis gave a slope of 2.06, an intercept of –18.45 kU/L, and r = 0.98. B, Evaluation of the Elecsys
E170 method. Passing-Bablok regression analysis gave a slope of 1.18, an intercept of 0.76, and r = 0.85. C, Evaluation of the
IMMULITE 2000 method. Passing-Bablok regression analysis gave a slope of 1.00, an intercept of –2.63 kU/L, and r = 0.87.
D, Evaluation of the UniCel DxI 800 method. Passing-Bablok regression analysis gave a slope of 1.29, an intercept of 1.41, and
r = 0.88. See Table 1 for manufacturer information.
Imprecision was acceptable with coefficients of variation
of 8% or less for all methods. Our imprecision data agree with
what has been described previously for similar methods.11,12
Differences in mean concentrations, especially for level 2, are
quite substantial. For the IMMULITE 2000, Elecsys E170, and
UniCel DxI 800, level 2 is close to the cutoff; however, this is
not the case for the ADVIA Centaur and the ARCHITECT
i2000. Even though all methods have similar cutoff values, 35
or 37 kU/L, the level 2 result range is from 34 to 123 kU/L,
suggesting a matrix effect with the quality control material.
Method comparison achieved acceptable performance,
and results are comparable to those of previous studies.11-13
The best correlation was between the ADVIA Centaur and the
ARCHITECT i2000. These two assays are more likely to correlate well based on the use of a common antibody. However,
the ARCHITECT i2000 slope of 2.06 would suggest differences in calibration. Owing to intermethod differences, confirmation of baseline values is required when changing methods.
Diagnostic concordance showed acceptable agreement among
the methods, with positive agreement ranging from 91.0% to
97.4% and negative agreement ranging from 68.4% to 93.0%.
Again, the ADVIA Centaur and ARCHITECT i2000 methods
showed the best overall concordance. This was likely because
these 2 methods use a single monoclonal antibody (1116-NS19-9) for the 2-step sandwich immunoassay, whereas the other
3 use one or more different antibodies.
❚Table 3❚
Summary of Diagnostic Concordance*
ARCHITECT i2000
Negative (≤37 kU/L)
Positive (>37 kU/L)
Total
IMMULITE 2000
Negative (≤37 kU/L)
Positive (>37 kU/L)
Total
Elecsys E170
Negative (≤35 kU/L)
Positive (>35 kU/L)
Total
UniCel DxI 800
Negative (≤35 kU/L)
Positive (>35 kU/L)
Total
ADVIA Centaur Negative
≤35 kU/L)
(≤
ADVIA Centaur Positive
(>35 kU/L)
Total
53
4
57
4
74
78
57
78
135
47
10
57
7
71
78
54
81
135
41
16
57
4
74
78
45
90
135
39
18
57
2
76
78
41
94
135
Overall
Concordance (%)
94.1 (127/135)
87.4 (118/135)
85.2 (115/135)
85.2 (115/135)
* See Table 1 for manufacturer information.
Am J Clin Pathol 2007;127:436-440
© American Society for Clinical Pathology
439
DOI: 10.1309/H52VET3M6P7GYWG1
439
439
La’ulu and Roberts / FIVE CA 19-9 ASSAYS
❚Table 4❚
Summary of Reference Interval Data
Method*
Range of Results (kU/L)
Median (kU/L)
97.5th Percentile (kU/L)
Percentage <35 kU/L
4.3-56.8
2.0-60.3
2.5-38.0
0.6-46.3
1.0-51.0
9.7
4.4
3.3
6.9
7.0
4.7-37.1
2.0-26.4
2.5-17.0
0.6-31.9
1.0-33.2
97.6
98.4
99.2
99.2
99.2
ADVIA Centaur
ARCHITECT i2000
IMMULITE 2000
Elecsys E170
UniCel DxI 800
* See Table 1 for manufacturer information.
The upper 97.5% reference limit for all methods, except
the ADVIA Centaur, was below the manufacturers’ upper limit
of expected values for healthy subjects. The ADVIA Centaur
had an upper limit of 37 kU/L with a manufacturers’ defined
upper limit of 35 kU/L. For the ARCHITECT i2000 and the
IMMULITE 2000, the upper limit was well below the manufacturer’s cutoff. The ARCHITECT i2000 had an upper reference
limit of 26 kU/L with a manufacturer’s cutoff of 37 kU/L, and
the IMMULITE 2000 had an upper reference limit of 17 with a
cutoff of 37 kU/L. The Elecsys E170 and UniCel DxI had upper
limits that were only slightly lower (32 kU/L and 33 kU/L,
respectively) than the manufacturer’s cutoff value of 35 kU/L.
All automated CA 19-9 assays showed acceptable analytic performance. CA 19-9 results are method-dependent, and
continuing efforts to harmonize assays are needed. The availability of an international reference material could facilitate
this effort. As with many tumor markers used in monitoring
cancer, instead of focusing merely on the cutoff values, consideration of the critical differences may be more useful.12 A
single method should be used because values obtained with
different methods cannot be used interchangeably. If the assay
method is changed, measurement by both methods to determine a new baseline is required.
From the 1ARUP Institute for Clinical and Experimental Pathology
and 2Department of Pathology, University of Utah Health Sciences
Center, Salt Lake City.
Supported by Abbott Diagnostics and the ARUP Institute for
Clinical and Experimental Pathology.
Address reprint requests to Dr Roberts: c/o ARUP
Laboratories, 500 Chipeta Way, Salt Lake City, UT 84108.
Acknowledgments: We gratefully acknowledge Abbott
Diagnostics, Diagnostic Products, and Roche Diagnostics for
providing instrumentation to perform testing using their methods.
References
1. Lamerz R. CA 19-9: GICA (gastrointestinal cancer antigen).
In: Sell S, ed. Serological Cancer Markers. Totowa, NJ: Humana
Press; 1992:309-339.
440
440
Am J Clin Pathol 2007;127:436-440
DOI: 10.1309/H52VET3M6P7GYWG1
2. Kim JE, Lee KT, Lee JK, et al. Clinical usefulness of
carbohydrate antigen 19-9 as a screening test for pancreatic
cancer in an asymptomatic population. J Gastroenterol Hepatol.
2004;19:182-186.
3. Goggins M, Koopmann J, Yang D, et al. National Academy of
Clinical Biochemistry (NACB) Guidelines for the Use of Tumor
Markers in Pancreatic Ductal Adenocarcinoma. Available at
http://www.nacb.org/lmpg/tumor/chp3i_pancreatic.pdf.
Accessed May 2006.
4. Chan DW, Booth RA, Diamandis EP. Tumor markers. In:
Burtis CA, Ashwood ER, Bruns DE, eds. Tietz Textbook of
Clinical Chemistry and Molecular Diagnostics. 4th ed. St Louis,
MO: Elsevier Saunders; 2006:773.
5. Engelen MJ, de Bruijn HW, Hollema H, et al. Serum CA 125,
carcinoembryonic antigen, and CA 19-9 as tumor markers in
borderline ovarian tumors. Gynecol Oncol. 2000;78:16-20.
6. Takahashi Y, Takeuchi T, Sakamoto J, et al. The usefulness of
CEA and/or CA19-9 in monitoring for recurrence in gastric
cancer patients: a prospective clinical study. Gastric Cancer.
2003;6:142-145.
7. Koprowski H, Steplewski Z, Mitchell K, et al. Colorectal
carcinoma antigens detected by hybridoma antibodies.
Somatic Cell Genet. 1979;5:957-971.
8. Pilo A, Zucchelli GC, Cohen R, et al. Comparison of
immunoassays for tumor markers CA 19-9, CA 15-3 and CA
125: data from an international quality assessment scheme.
Tumori. 1995;81:117-124.
9. Pilo A, Zucchelli GC, Cohen R, et al. Performance of
immunoassays for CA 19-9, CA 15-3 and CA 125 tumour
markers evaluated from an international quality assessment
survey. Eur J Clin Chem Clin Biochem. 1996;34:145-150.
10. Owen WE, Roberts WL. Performance characteristics of the
IMMULITE 2000 erythropoietin assay. Clin Chim Acta.
2004;340:213-217.
11. Romer M, Haeckel R, Brux B, et al. A multicentre evaluation
of the CA 15-3 assay, CA 19-9 assay and CA 125 II assay on
the Bayer Immuno 1 System. Eur J Clin Chem Clin Biochem.
1997;35:637-644.
12. Stern P, Friedecky B, Bartos V, et al. Comparison of different
immunoassays for CA 19-9. Clin Chem Lab Med.
2001;39:1278-1282.
13. van Dalen A. BR-MA, OM-MA, GI-MA and CEA: clinical
evaluation using the IMMULITE analyzer. Tumour Biol.
1999;20:117-129.
© American Society for Clinical Pathology