Evaluation of Imprecision, Bias and Total

Error of Clinical Chemistry Analysers

S. S. Biswas, M. Bindra, V. Jain &

P. Gokhale

Indian Journal of Clinical

Biochemistry

ISSN 0970-1915

Ind J Clin Biochem

DOI 10.1007/s12291-014-0448-y

1 23
Your article is protected by copyright and all
rights are held exclusively by Association of
Clinical Biochemists of India. This e-offprint
is for personal use only and shall not be self-
archived in electronic repositories. If you wish
to self-archive your article, please use the
accepted manuscript version for posting on
your own website. You may further deposit
the accepted manuscript version in any
repository, provided it is only made publicly
available 12 months after official publication
or later and provided acknowledgement is
given to the original source of publication
and a link is inserted to the published article
on Springer's website. The link must be
accompanied by the following text: "The final
publication is available at link.springer.com”.

1 23
 Author's personal copy
Ind J Clin Biochem
DOI 10.1007/s12291-014-0448-y

SHORT COMMUNICATION

Evaluation of Imprecision, Bias and Total Error of Clinical

Chemistry Analysers
S. S. Biswas • M. Bindra • V. Jain • P. Gokhale

Received: 14 March 2014 / Accepted: 23 May 2014

Ó Association of Clinical Biochemists of India 2014

Abstract Context Two Biosystems analysers are used in Keywords Bias  Imprecision  Total error
our laboratory, a fully automated A25 and a semi-auto-
mated BTS-350. Internal quality control is done for both
but external quality control only for A25. As BTS-350 is Introduction
used for backup, it is important that the results of both
analysers are not just comparable but also within prede- In life, there is continuous fluctuation of the components in
fined limits of systematic, random and total error (TE). Aim biological fluids. The biological variation (BV) of analytes
To evaluate the imprecision, bias and TE of the two Bio- is of three types, namely, variation over the life span,
system analysers. Materials and Methods Biosystems cyclical variation and random variation. The latter causes
level-1 quality control sera lot number 70A was run in subtle variation around the setting point of each individual
duplicate for 32 days on both the analysers. Between day which is responsible for the within-subject or intra-indi-
imprecision (measured by the coefficient of variation), bias vidual BV, while the overall variation is responsible for the
and TE were calculated for ten analytes and were checked between-subject or inter-individual BV. The BV and the
to see whether they are within the acceptable minimum analytical variation both affect the test result, but while the
limits, desirable limits and optimum limits of allowable latter can be minimised, minimisation of the former is not
error based on specifications on Westgard’s website possible. Hence it is important to ensure that the analytical
updated in 2014. Results On both the analysers, all the variation is kept minimised and does not contribute sig-
analytes except alkaline phosphatase were within the nificant additional variation to that contributed by the BV.
acceptable minimum limits of TE and most analytes were The analytical variability is therefore kept appropriately
within the desirable limits of TE. Only TG on A25 was less than the biological variability for the test to be confi-
within the optimum limit of TE. Conclusion The two dently used for clinical diagnosis and monitoring [1].
Biosystem analysers performed comparably with errors Measurement of laboratory analytical errors fall into two
within acceptable limits for most analytes. BTS-350 was main categories, systematic error and random error. Sys-
found to be a suitable and ready backup analyser for A25. tematic errors are predictable problems influencing obser-
vations consistently in one direction, while random errors
are more unpredictable [2]. Systematic errors are assessed
S. S. Biswas (&)  M. Bindra  V. Jain  P. Gokhale
by the bias, while random errors by the imprecision mea-
Department of Biochemisty, LN Medical College,
Bhopal 462042, India sured by the coefficient of variation (CV). Imprecision
e-mail: drssbis@gmail.com; shubho20005@yahoo.com affects the reproducibility and repeatability of results [3].
M. Bindra Reproducibility is the closeness of the results of successive
e-mail: ruby_bindra2005@yahoo.com measurements under changed conditions which require
V. Jain multicentric trials. Repeatability is the closeness of the
e-mail: v.jain33@gmail.com results of at least twenty successive measurements under
P. Gokhale similar conditions. Bias is the average deviation from a true
e-mail: prernadg2319@gmail.com value with minimal contribution of imprecision while

123
 Author's personal copy
Ind J Clin Biochem

Table 1 Imprecision (CV %) of Biosystem A25 & BTS-350 analysers

Analyte A25 A25 BTS-350 BTS-350 A25 BTS-350 Desirable Minimum Optimum
mean SD mean SD CV (%) CV CV CV CV
(%) (%) (%) (%)

Glucose 91.8 2.29 92.4 1.87 2.5 2.0 2.8 4.2 1.4
Urea 28.8 1.54 28.3 1.52 5.4 5.4 6.1 9.2 3.1
Creatinine 1.56 0.05 1.56 0.06 3.2a 3.8a 3.0 4.5 1.5
Cholesterol 148.5 7.10 151.3 6.94 4.8a 4.6a 3.0 4.5 1.4
Triglyceride 57.9 2.96 58.0 5.21 5.1 8.9 10.0 15.0 5.0
T. Bilirubin 2.36 0.20 2.36 0.17 8.5 7.3 10.9 16.4 5.5
D. Bilirubin 0.59 0.10 0.57 0.09 16.9 15.8 18.4 27.6 9.2
ALP 168.9 13.15 168.4 11.08 7.8a 6.6a 3.2 4.8 1.6
ALT 35.2 2.94 33.9 2.48 8.3 7.3 9.7 14.6 4.9
AST 41.4 2.47 41.3 2.42 5.9 5.9 6.2 9.3 3.1
a
Beyond limit of desirable imprecision

inaccuracy is the deviation of a single measurement from periodically updated on Westgard’s website [9]. The min-
the true value with significant contribution by imprecision imum and optimum limits of these specifications are also
[4]. Multiple measurements, at least twenty and preferably provided on Westgard’s website [10, 11]. The most clini-
forty, are therefore required for calculating imprecision as cally and technically appropriate goal is taken as the
well as bias [5]. minimum for imprecision and bias.
Uncertainty of measurement provides a quantitative In our clinical laboratory, we are using two Biosystems
estimate of the quality of a test result. Uncertainty is defined analysers, a fully automated A25 analyser for routine
as ‘‘a parameter associated with the result of a measure- clinical chemistry and a semi-automated BTS-350 for
ment, that characterises the dispersion of the values that back-up purpose. Internal quality control with Biosystem
could reasonably be attributed to the measurand’’. Sources quality control sera is carried out on both analysers to
that contribute to uncertainty may include sampling, sample ensure that the results are within control limits by plotting
preparation, sample portion selection, calibrators, reference the controls on the Levey Jennings chart and checking
materials, input quantities, equipment used, environmental acceptability according to Westgard rules [12]. However,
conditions, condition of the sample and changes of operator only the results on A25 are evaluated by the external
Comparison of result is made possible by an estimate of the quality assessment scheme. It is important therefore to
uncertainty of measurement, calculated as the 95 % confi- ensure that the results of the two analysers are not just
dence interval (±1.96 CV %) [6]. similar, but also within the predefined limits of error.
The uncertainty associated with a test result due to the Hence this study was carried out to evaluate the two Bio-
random errors is termed imprecision, which is determined system’s analysers on the basis of their bias, imprecision
from the data of internal quality control. The total analyt- and TE.
ical error also has to include an estimate of analytical bias.
Anaytical goal setting is required to determine whether a
method is producing ‘fit for purpose’ results. The upper Methodology
acceptable limit for imprecision is taken as a proportion of
the intra-individual BV of the analyte and the upper limit We studied two analysers of Biosystem, Phillipines. The
for analytical bias as a proportion of the overall BV, intra- A25 is a fully automated random access analyser with fol-
individual and inter-individual. Together, this determines lowing major specifications-throughput of 240 tests/h,
goal-setting for the total analytical error (bias ? impreci- minimum reading volume of 200 ll, filter configuration 340,
sion) [6]. 405, 505, 535, 560, 600, 635, 670 nm, measuring range
In the early 1990, recommendations were made by a 0.05–2.5 A, automatic conditioning of fluid system, Levy-
group of European scientists for evaluation of clinical Jennings QC Chart, flexibility in positioning with sample
chemistry and other analysers in terms of its imprecision, and reagents racks and unlimited STAT capabilities, tem-
bias and total error (TE). It came to be known as the perature 10–35 °C and relative humidity\75 %. The BTS-
‘‘European Biologic Goals and Calculated Biologic 350 is a semi-automated analyser with following specifica-
Allowable Total Errors’’ [7]. Later, Ricos et al. [8] built the tions- LED Configuration 340, 405, 505, 535, 560, 600, 635,
biodatabase of these desirable specifications, which are 670, durable and humidity proof hard coated filters, battery

123
 Author's personal copy
Ind J Clin Biochem

Table 2 Bias (%) of Biosystem Parameter QC A25 BTS-350 A25 BTS-350 Desirable Minimum Optimum
A25 and BTS-350 analysers target mean mean bias (%) bias (%) bias (%) bias (%) bias (%)

Glucose 92 91.8 92.4 1.6 1.8 2.3 3.5 1.2

Urea 29.7 28.8 28.3 5.5 5.4 5.6 8.4 2.8
Creatinine 1.55 1.56 1.56 2.7 3.2 4.0 6.0 2.0
Cholesterol 152 148.5 151.3 4.8a 4.1 4.1 6.2 2.1
Triglycerides 59 57.9 58.0 4.2 7.3 9.6 14.4 4.8
T. Bilirubin 2.4 2.36 2.36 6.2 4.7 9.0 13.5 4.5
D. Bilirubin 0.6 0.59 0.57 14.6a 13.5 14.2 21.3 7.1
ALP 179 168.9 168.4 9.3a 7.4a 6.7 10.1 3.4
ALT 35.5 35.2 33.9 7.6 7.6 11.5 17.3 5.8
a AST 41.6 41.4 41.3 4.9 4.5 6.5 9.8 3.3
Beyond limit of desirable bias

power pack, photometric range (0.0–3.5 A) for all wave- The between day imprecision, bias and TE for the two
lengths, flow cuvette 18 ll and Levy-Jennings QC Chart. instruments were checked for each analyte to see if they
Both analysers were calibrated at installation using Biosys- were within the limits of minimum, desirable and optimum
tems calibrators and Biosystems reagent kits. Calibration specifications updated in 2014 respectively [9–11]. These
was done thereafter whenever internal quality control results analytical goals are derived from BV [6] as follows-
failed westgard’s rules. Biosystems level-1 (human) quality
(a) There are three levels of analytical goal for impre-
control sera lot number 70A suitable as an accuracy control
cision derived from intra-individual BV:
was run in duplicate for 32 days on both the analysers and
the average of the two values was noted. The control sera Optimum: CVA ¼ \0:25  CV1
complies with the directions set by the ISO 15189. It’s values Desirable: CVA ¼ \0:50  CV1
are traceable to international certified reference materials: C-
RSE/IFCC, SRM927 c, SRM 909 b, ERM-DA470, ERM- Minimum: CVA ¼ \0:75  CV1
AD455, BRM 97/662, RM W1066, BCR 470. As the A25 where: CVA = Coefficient of variation (analytical)
analyser does not have onboard cooling, we analyse only ten and CVI = Coefficient of variation (intra-individ-
analytes on it that are most frequently ordered. The mean ual), derived from the intra-individual BV
values and standard deviations were calculated for these ten (b) There are three levels of analytical goal for bias
analytes i.e. glucose, urea, creatinine, total cholesterol, tri- derived from intra-individual and inter-individual
glycerides (TG), total bilirubin, direct bilirubin, alkaline BV:
phosphatase (ALP), alanine transaminase (ALT) and
1=2
aspartate transaminase (AST) for each machine. The con- Optimum: BA ¼ \0:125 CV2I þ CV2G
trols were plotted on the Levey Jennings chart to check
1=2
Desirable: BA ¼ \0:250 CV2I þ CV2G
acceptability according to Westgard rules. Student t test was
applied to ensure that there was no significant difference in
1=2
Minimum: BA ¼ \0:375 CV2I þ CV2G
the means of any analyte measured by the two analysers.
Between day imprecision, bias and (TE) were determined for where: BA = analytical bias, CVI = CV of within-
each analyte on each analyser as follows- subject (intra-individual) BV and CVG = CV of
CV% ¼ ðSD=Mean)  100 between—subject (inter-individual) BV.
(c) The two parameters are conveniently combined as
CV (%) is the coefficient of variation for measuring total error allowable (TEa), for which three levels of
between day imprecision, SD is the standard deviation analytical goal are set:
Bias % ¼ ðAverage absolute deviation from the target value= Optimum: TEa ¼ \1:65ð0:25CVI Þ

1=2
Target)  100 þ 0:125 CV2I þ CV2G

The TE (%) was calculated as 1.65 9 CV (%) ? Bias Desirable: TEa ¼ \1:65ð0:50CVI Þ

1=2
(%) [13]. The factor 1.65 implies that 95 % of the results þ 0:250 CV2I þ CV2G
will fall within the TE limit, given a Gaussian distribution.

123
 Author's personal copy
Ind J Clin Biochem

Table 3 Total Error of Parameter A25 BTS-350 Desirable Minimum Optimum

Biosystem A25 & BTS-350 total error total error total error total error (%) total error
analysers (%) (%) (%) (%)

Glucose 5.7 5.1 7.0 10.5 3.5

Urea 14.3 14.2 15.6 23.4 7.8
Creatinine 7.9 9.6a 8.9 13.4 4.5
Cholesterol 12.7a 11.7a 9.0 13.5 4.3
Triglycerides 12.6 22.1 26.0 39.0 13.0
T. Bilirubin 20.2 16.8 26.9 40.4 13.5
D. Bilirubin 42.6 39.5 44.5 66.8 22.3
ALP 22.1a 18.2a 12.0 18.0 6.0
ALT 21.4 19.7 27.5 41.3 13.8
a
Beyond limits of desirable AST 14.8 14.2 16.7 25.1 8.4
total error

Minimum: TEa ¼ \1:65ð0:75CVI Þ authors to select the most appropriate method. In our study,

1=2
þ 0:375 CV2I þ CV2G we checked the level of error on our two analysers in
comparison to the limits specified on Westgard’s website.
On both the analysers, all analytes except ALP were
within minimum limits and most within desirable limits of
Results imprecision, bias and TE. Cholesterol was outside desir-
able limits of both imprecision and bias while outliers
All the analytes on both the analysers except ALP were affected creatinine, affecting only its precision. Only TG
within the minimum limits of between day imprecision. on A25 was within the stringent optimum limit of TE.
Three analytes on both analysers had CV (%) outside the Several authors have reported analysis of results on their
limits of desirable imprecision—total cholesterol, creati- analyser following one or the other guideline with variable
nine and ALP. No analyte had CV(%) within the limits of results. Coudene et al. evaluated 32 common analytes on
optimum imprecision (Table 1). the ABX Pentra 400 according to the National Committee
All the analytes on both analysers were within the for Clinical Laboratory Standards and Valtec protocols and
minimum limits of bias (%). However on A25, three ana- reported imprecision within acceptable limits with mod-
lytes had bias (%) outside the desirable limits—total cho- erate influence of interfering substances [15]. Miler et al.
lesterol, direct bilirubin and ALP, while on BTS-350 there compared Olympus AU2700 with Olympus AU640 with
was one—ALP. Only TG on A25 had bias (%) within guidelines from the Croatian Society of Medical Bio-
optimum limits (Table 2). chemists [16] The results were comparable but control
All the analytes on both analysers except ALP were samples with low concentrations did exceed their allowable
within the minimum limits of TE (%). However, analytes biases with conjugated bilirubin having the maximum bias
with TE (%) outside the desirable limits on A25 were (16.48 %).
two—total cholesterol and ALP, while on BTS-350 there In our study, the two analysers provided comparable
were three—creatinine, total cholesterol and ALP. Only results. Lack of automation did not decrease the perfor-
analyte within optimum limit of TE (%) was TG on A25 mance of BTS-350 although automation is necessary for
(Table 3). handling large sample loads. BTS-350 proved to be an
effective back-up analyser for correct diagnosis and lon-
gitudinal follow up. Although laboratory errors are fre-
Discussion quently pre-analytical or postanalytical [17], ensuring
minimisation of analytical error on both analysers ensures
It is a medical need to have some preset quality specifi- two ready analysers standardized for comparable reporting.
cation of laboratory test results. This ensures quality and There were two major limitations of our study. It used
uniformity not just across laboratories but also within a only level one quality control sera. The imprecision is
laboratory with multiple analysers. Evaluation of perfor- better recorded at more than one level of quality control
mance with preset quality specifications based on BV has depending on the range of reportable values and clinical
been going on since many years, [14] but lack of consensus use of the test. We could have also used patient data to
on the published recommendations makes it difficult for calculate the BV by collecting and storing a number of

123
 Author's personal copy
Ind J Clin Biochem

samples from a number of individuals to analyse them testing: a laboratory implementation guide. Clin Biochem Rev.
simultaneously. The within-subject and between-subject 2004;25(4):S1–24.
7. Fraser CG, Petersen PH, Ricos C, Haeckel R. Proposed quality
variance could be determined by analysis of variance specifications for the imprecision and inaccuracy of analytical
(ANOVA) of obtained results [1]. Controlling pre-analytical systems for clinical chemistry. Eur J Clin Chem Clin Biochem.
factors, the results of BV estimates in adults are similar. The 1992;30(5):311–7.
estimates of BV however, is not only made available but 8. Ricos C, Alvarez V, Cava F, Garcia-Lario JV, Hernandez A,
Jimenez CV, et al. Current databases on biological variation:
updated every 2 years on westgard’s website [9]. pros, cons and progress. Scand J Clin Lab Invest. 1999;59(7):
491–500.
9. Ricos C, Alvarez V, Cava F, Garcia-Lario JV, Hernandez A,
Conclusion Jimenez CV, Minchinela J, Perich C, Simon M (2014) Desirable
specifications for total error, imprecision, and bias, derived from
intra- and inter-individual biologic variation. http://www.west
The two Biosystem analysers performed comparably with gard.com/biodatabase1.htm. Accessed May 2014.
all analytes except ALP within the minimum specified 10. Ricos C, Alvarez V, Cava F, Garcia-Lario JV, Hernandez A,
acceptable limits and most analytes within the desirable Jimenez CV, Minchinela J, Perich C, Simon M (2014) Minimum
specifications for total error, imprecision, and bias, derived from
limits. However, only TG on A25 met the most stringent intra- and inter-individual biologic variation. https://www.west
optimum limit for all the errors. BTS-350 was thus found to gard.com/minimum-biodatabase1.htm. Accessed May 2014.
be a suitable and ready backup analyser for A25. 11. Ricos C, Alvarez V, Cava F, Garcia-Lario JV, Hernandez A,
Jimenez CV, Minchinela J, Perich C, Simon M (2014) Optimum
specifications for total error, imprecision, and bias, derived from
intra- and inter-individual biologic variation. http://www.west
gard.com/optimal-biodatabase1htm.htm. Accessed May 2014.
References 12. Barry PL (2014) QC: the Levey–Jennings control chart. http://
www.westgard.com/lesson12.htm. Accessed May 2014.
1. Ricos C, Perich C, Minchinela J, Álvarez V, Simón M, Biosca C, 13. Oosterhuis WP. Gross overestimation of total allowable error
et al. Application of biological variation: a review. Biochem based on biological variation. Clin Chem. 2011;57(9):1334–6.
Med. 2009;19(3):250–9. doi:10.11613/BM.023. 14. Hackney JR, Cembrowski GS. Need for improved instrument and
2. Bishop ML, Fody EP, Schoeff LE. Clinical chemistry. Principles, kit evaluations. Am J Clin Pathol. 1986;86(3):391–3.
techniques, correlations. 7th ed. Philadelphia: Wolters Kluwer/ 15. Coudène P, Marson B, Badiou S, Flavier S, Anelli S, Cristol JP, et al.
Lippincott Williams & Wilkins; 2013. Evaluation of the ABX Pentra 400: a newly available clinical
3. Evaluating CD, Precision A. Clin Biochem Rev. 2008;29(Suppl chemistry analyser. Clin Chem Lab Med. 2005;43(7):782–92.
1):S23–6. 16. Miler M, Šimundić AM, Štefanović M, Ferenec-Ružić D,
4. Johnson R. Assessment of bias with emphasis on method com- Kvaternik M, Topic E, Vrkic N. A model for results comparison
parison. Clin Biochem Rev. 2008;29(Suppl 1):S37–42. on two different biochemistry analyzers in laboratory accredited
5. Clinical and Laboratory Standards Institute (2005) User verifi- according to the ISO 15189. Biochem Med. 2009;19(3):287–93.
cation of performance for precision and trueness; approved doi:10.11613/BM.2009.027.
guideline. 2. Wayne, PA, USA: CLSI. CLSI document EP15-A2. 17. Plebani M. Laboratory errors: how to improve pre- and post-
6. White GH, Fraser I, AACB Uncertainty of Measurement Work- analytical phases? Biochem Med. 2007;17(1):5–9. doi:10.11613/
ing Group. Uncertainty of measurement in quantitative medical BM.2007.001.

123