Clinical Microbiology: Sigma Metrics For Assessing Accuracy of Molecular Testing
Clinical Microbiology: Sigma Metrics For Assessing Accuracy of Molecular Testing
Clinical Microbiology: Sigma Metrics For Assessing Accuracy of Molecular Testing
Microbiology
Stay Current...
Stay Informed.
N E W S L E T T E R
Why Do Quality Metrics Matter? We contend that proper patient care can be built
The major assumption physicians make prior only on a foundation of quality, which is sup-
to their clinical decisions is that the laboratory’s ported by metrics and analytics. Regardless of
test results are valid - that there is no medically the disease state, assay quality must begin with
important error that obscures the true result, assessment of an assay’s analytical performance
generating either a false-positive or a false-neg- and extend to the use of its results at the relevant
ative result. Laboratory medicine embraces the clinical (medical) decision points. In this review,
paramount importance of quality. With a long we present analytic tools that one can use to criti-
history of quality and accuracy for diagnostic cally assess an assay’s performance by analyzing
testing, it is easy to assume that a test result has results from method verification data or quality
the appropriate accuracy, but how can we ensure control data.
that the necessary accuracy is being achieved? How Do We Know That When We Achieve
How can we ensure that medical decisions are the Proper Level of Quality?
not made based on faulty assumptions? We know
The core requirement of any method used for
inherently that patient diagnosis, and in some
quality assessing is the ability to numerically
cases, treatment may be endangered by poor test
define what we consider good performance and
Corresponding author:
performance, regardless of the skill of the clini-
acceptable quality; therefore, we must also define
Danijela Lucic, Ph.D., Abbott cian or the laboratorian, but do we know how
unacceptable quality and poor performance, i.e.,
Molecular, Inc., 1350 East Touhy to analyze our quality control data to prove the
we must define an unacceptable error, i.e., a defect
Ave., Des Plaines, IL 60018. assay is functioning properly? We certainly col-
Tel.: 224-361-7124. Fax: in the method’s performance. For many diagnostic
lect quality control data, and review it, but do we
224-234-7707. E-mail: tests, the definition of what constitutes acceptable
critically assess the results?
danijela.lucic@abbott.com quality is implicit, but not often clearly articu-
While the reduction in defects is a reward in itself, there are other For analytical methods, like viral load assays, an alternative
obvious benefits to achieving Four-, Five-, and Six-Sigma assay approach must be deployed in order to calculate the Sigma met-
performance. For manufacturing and industry, achieving Six Sigma ric. Rather than count defects, we instead calculate the expected
means operating at the highest efficiency, process reliability, and Sigma metric through an equation that combines the impacts of
customer satisfaction, and ultimately, all of these effects generate imprecision and inaccuracy. The Sigma metric equation (16,17)
the maximum product quality and profit. In contrast, processes is as follows: Sigma metric = (TEa < bias)/CV. In this equation,
operating below Three Sigma are typically the most error-prone, the TEa, bias, and coefficient of variation (%CV) are all expressed
resource-consuming, difficult-to-maintain, and expensive pro- as percentages and the absolute value of the bias is used. For any
cesses in an organization. Outside of health care, processes oper- molecular method, as long as all the variables are expressed in
ating below Three Sigma are usually considered too unstable either logarithmic or non-logarithmic numerical scales, it is pos-
for routine operation because they generate too much re-work. sible to calculate a Sigma metric.
These are processes that must be radically improved, redesigned,
or, ultimately, replaced.
For viral loads, the TEa value is established based on clinical
Similarly, within health care laboratories, test methods that oper- guidelines, so the percent TEa reflects a medically important
ate below Three Sigma typically generate significant re-work change (i.e., 0.5 log copies/ml in and HIV viral load). The %CV
in the form of repeated controls, repeated calibrations, trouble- and bias values are obtained from repeated measurement of labora-
shooting, and even a need for repeated testing of the patient in tory control material, used to assess precision during the method
order to determine a diagnosis. Indeed, in some cases, diagnostic verification or by analyzing longitudinal quality control data. The
methods performing significantly below Three-Sigma have been %CV and bias values can also be obtained from comparisons with
recalled from the market. But the concepts of Six Sigma are rela- peer group surveys or proficiency testing surveys.
tively new to molecular diagnostic laboratories and are not yet
commonly calculated. Awareness of the Sigma metrics should The Sigma metric can be calculated with the variables expressed in
help molecular laboratories better define assay performance and unit-based measurements, as long as all the terms of the equation
improve overall quality. are kept consistent. The graph in Fig. 1 is a visual explanation of
the Sigma metric equation, showing how both bias and impreci-
'H¿QLQJ4XDOLW\5HTXLUHPHQWVIRU6L[6LJPD3HUIRUPDQFH sion affect the distribution of test results. Whenever the curve of
the distribution exceeds the TEa (the lines drawn on either side of
Most molecular laboratories are not fully accustomed to discuss-
ing the “tolerance limits” of our test methods. Laboratorians are
far more comfortable discussing a similar concept, known as the
allowable total error (TEa). The TEa, the net analytical error is
defined as a combination of method imprecision (random error)
and method inaccuracy (systematic error, or bias) (16,17). When
a test method exceeds the TEa, the method or process begins to
generate defects (result errors), which could be classified as either
false positive or false negative results. When we want to prevent
our analytical method from exceeding the Tea, we can rely on
Sigma metrics. If we can achieve Six-Sigma performance, in other
words, maintain our imprecision at approximately less than 1/6 of
the TEa, we keep our defect rate below 3.4 DPM.
A B
Figure 3. (a) Neath et al. analytical analysis at 40, 80, and 90 cps/ml. (b) Method decision chart for viral load change from 40 to 200 cps/ml.
A B
Figure 4. (a) Ruelle et al. analytical analysis at 12, 25, 50, 100 cps/ml. (b) Method decision chart for viral load change from 40 to 400 cps/ml.
A B
Figure 5. (a) Clinical samples at 25 IU/ml were run 10 times. The mean CVs were 26% and 65% for ART and CAP/CTM, respectively. (b) Method
decision chart for viral load change from LOQ to 50 IU/ml.
Discussion Disclosure
Molecular assays have been utilized to manage HIV-1 and HCV Danijela Lucic is an employee of Abbott Molecular Inc.; Sten
therapy response for decades. Differences in accuracy and pre- Westgard is a consultant for Abbott Inc.
cision at clinical decision points where treatment decisions are
5HIHUHQFHV
being made and patient management is impacted have become a
point of laboratory discussion. Historically, laboratories have chal- 1. Ambrosioni, J. et al. HIV treatment outcomes in Europe and North
lenged assay performance only across the full linear range (overall America: what can we learn from the differences? Expert Rev. Anti-
Infect. Ther. 12:523-526.
bias); however, there is a shift towards assessing assay performance
2. Coppola, N. et al. 2015. Treatment of chronic hepatitis C in patients
near or at the clinical decision points. This strategy proved useful with HIV/HCV coinfection. World J. Virol. 4:1-12.
when the first generation of direct-acting antivirals was approved 3. Easterbrook, P.J. et al. 2014. The role of mathematical modelling in
for HCV treatment. Assay sensitivity and precision were critical the development of recommendations in the 2013 WHO consolidated
for appropriate HCV therapy management and duration. The antiretroviral therapy guidelines. AIDS 28(Suppl. 1):S85-S92.
concept is also useful for HIV-1 therapy management, where 4. Holmes, C. et al. 2014. Health systems implications of the 2013
inaccurate viral load results could lead to inappropriate therapy WHO consolidated antiretroviral guidelines and strategies for suc-
cessful implementation. AIDS 28(Suppl. 2):S231-S239.
or failing therapy.
5. Lee, F.J., J. Amin, and A. Carr. 2014. Efficacy of initial antiretroviral
These Six sigma statistical concepts have much broader implica- therapy for HIV-1 infection in adults: a systematic review and meta-
tions than those of a single patient management decision. From a analysis of 114 studies with up to 144 weeks’ follow-up. PLoS One
9:e97482.
therapy management perspective, virus with resistance-associated
mutations is more challenging to treat, with narrower windows for 6. Shaheen, M.A. and M. Idrees. 2015. Evidence-based consensus on the
diagnosis, prevention and management of hepatitis C virus disease.
therapy options. In addition, false-positive rates (blip frequency) World J. Hepatol. 7:616-627.
can add unnecessary costs due to extensive repeat testing or more 7. Webster, D.P., P. Klenerman, and G.M. Dusheiko. 2015. Hepatitis
expensive resistance testing, and increase anxiety for the patient. C. Lancet 385:1124-1135.
Tracking an assay and instrument performance with Sigma met- 8. Blick, K.E. 2013. Providing critical laboratory results on time, every
rics allows laboratories to identify defects in assays prior to down- time to help reduce emergency department length of stay: how our
stream patient management decisions and impact. laboratory achieved a Six Sigma level of performance. Am. J. Clin.
Pathol. 140:193-202.
Conclusions 9. Burnett, D. et al. 2010. Collective opinion paper on findings of the
2009 convocation of experts on quality control. Clin. Chem. Lab.
Six-Sigma analyses provide a useful tool to assay precision and Med. 48:41-52.
overall quality for both HIV and HCV load testing. The Sigma
10. Carlson, R.O., F. Amirahmadi, and J.S. Hernandez. 2012. A primer
metric translates abstract analytical performance characteristics on the cost of quality for improvement of laboratory and pathology
into tangible measures that can impact laboratory operations and specimen processes. Am. J. Clin. Pathol. 138:347-354.
patient outcomes. As diagnostic methods and antiviral treatments 11. Daley, A.T. 2006. Pro: lean six sigma revolutionizing health care of
have evolved, the demands for analytical performance of viral load tomorrow. Clin. Leadersh. Manag. Rev. 20:E2.
testing have increased. In previous decades, extreme precision may 12. Gras, J.M. and M. Philippe. 2007. Application of the Six Sigma
not have been as necessary, but today’s medical treatments demand concept in clinical laboratories: a review. Clin. Chem. Lab. Med.
45:789-796.
higher precision. Without setting a higher bar for analytical per-
13. Landek, D. 2006. Con: six sigma not always the right answer in the
formance, more blips and errors will occur in viral loads, and fewer clinical laboratory. Clin. Leadersh. Manag. Rev 20:E3.
desirable outcomes will be achieved. 14. Nanda, S.K. and L. Ray. 2013. Quantitative application of sigma
Sigma metrics can be implemented only with clinically defined metrics in medical biochemistry. J. Clin. Diagn. Res. 7:2689-2691.
TEa limits, which must relate directly to the treatment guidelines 15. Nevalainen, D. et al. 2000. Evaluating laboratory performance on
quality indicators with the six sigma scale. Arch. Pathol. Lab. Med.
for the disease. When these TEa limits are defined, a Sigma metric
124:516-519.
can be calculated and an objective judgment can be made about
16. Westgard, J.O. and S.A. Westgard. Assessing quality on the Sigma
method precision and acceptability. Once the Sigma metric is cal- scale from proficiency testing and external quality assessment surveys.
culated, that assessment can be leveraged into additional actionable Clin. Chem. Lab. Med. (In press.)
steps for the laboratory, for example, design of the appropri- 17. Westgard, S. 2013. Prioritizing risk analysis quality control plans
ate quality control procedures (i.e., the number of controls, the based on Sigma-metrics. Clin. Lab. Med. 33:41-53.
control limits, and, to some extent, the frequency of controls), as 18. Davis, C. et al. 2008. An international collaborative study to establish
well as estimations of the number of tests that will be required to a replacement World Health Organization International Standard for
human immunodeficiency virus 1 RNA nucleic acid assays. Vox Sang.
detect an error within the viral density range where medical deci- 95:218-225.
sions are to be made. In summary, the Sigma metric is not just a 19. Glaubitz, J. et al. 2011. Accuracy to 2nd International HIV-1 RNA
benchmark; it is the first step in a process of optimizing an assay’s WHO Standard: assessment of three generations of quantitative