close

Вход

Забыли?

вход по аккаунту

?

9783110346268-002

код для вставкиСкачать
2 The elements of a conventional quality
control protocol
2.1 Selection of quality requirements
The first step of any internal quality control protocol is to define appropriate quality
requirements, which have many synonyms: analytical goals, total error allowable
(TEa), and analytical performance specifications (APS). This concept refers to the
maximal amount of variation, generally expressed as a percentage, that can be tolerated in a patient test result, without having a significant impact on patient care. These
quality requirements have to be set a priori [1], before further proceedings in the setup
of a quality control strategy. The setting of an analytical goal is not an easy decision, as
it will have impact on quality control rules that are going to be selected. For a specific
laboratory test, the more stringent the quality requirement is, the more complex will
the QC rules be, and the more QC results will have to be rejected. This will guarantee
that no wrong patient results will be released, but it will take time, cost money, and
generate a lot of downtime in the laboratory. In contrary, the use of a wider allowable
total error will lead to simpler QC rules and fewer QC results infractions, but then,
patient’s results that are clinically wrong may be generated. The selection of quality
specifications is function of a balance between the risk of generating erroneous laboratory results and the need to design a practical QC protocol that will be achievable in
daily routine. Quality requirements have been the subject of much debate in the literature in the last 20 years. Unfortunately, there is no universally accepted analytical
goal for any laboratory test. The selection of an analytical goal remains the choice of
the clinical pathologist. The recent Milan Consensus Conference on APS, organized
by the European Federation of Laboratory Medicine (EFLM) restated a new hierarchy
[2] of APSs based on three different models (table 2.1). This is a simplification of the
previous hierarchy of analytical goals that was defined in the Stockholm Consensus
Conference in 1999 [1].
Tab. 2.1: Hierarchy of APSs as defined by the EFLM Milan 2014 conference.
Hierarchy
Model
1a
1b
2
3
APS based on direct outcome studies
APS based on indirect outcome studies
APS based on biological variation of the analyte
APS based on state-of-the-art
The Milan hierarchy of APSs is composed of three different levels: APS based on the
effect of quality requirements on clinical outcomes, either determined by direct or
DOI 10.1515/9783110346268-002
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
6 2 The elements of a conventional quality control protocol
indirect studies, APS based on biological variation (BV) data of the analyte, and APS
based on state-of-the-art [2].
As laboratory test results have ultimately an impact on patient health, it make
sense that the higher APS on the Milan consensus hierarchy is based on the effect of
test quality requirements on clinical performance, such as diagnostic accuracy or clinical effectiveness. APS based on the effect of quality specifications on clinical outcomes
may be defined by direct or indirect studies [3]. Direct studies would require randomized control trials comparing the effect on health outcomes (such as patient mortality and morbidity) of the same laboratory test being performed with different quality
requirements. Such diagnostic trials are complex to design and generally require a large
number of patients. Moreover, direct studies may only be performed for laboratory tests
for which the effect on patient health is very straightforward. For these reasons, studies
on APS-based direct outcome data are rare in the literature. Indirect studies may consist
in studying the effect of selected APS on diagnostic accuracy (sensitivity and specificity) of a test. It is recognized that while APS based on the effect of performance specification on clinical outcomes is the highest ranking model on the Milan hierarchy, very
few papers using this approach have been published in the last 15 years [3]. As the list
APS based on clinical remains very limited, this is a significant setback in the practical
use of clinical outcomes to set allowable total errors in a clinical laboratory.
Analytical goals based on BV have gained popularity in clinical laboratories since
many years. BV is the random variation of laboratory results around a homeostatic
setting point. There are two components of BV: intra-individual and inter-individual BV.
BV data can be used to define objectives for bias and imprecision, using formulas (table
2.2) that were designed by Fraser in 2001 [4]. The combination of objectives for bias and
for imprecision will give TEa based on BV data. Authors, led by Carmen Ricos, have compiled data regarding BV studies to create tables that are regularly updated. According to
the latest 2014 update [5], data regarding inter-individual CV (CVg), intra-individual CV
(CVi or CVw), goals for desirable accuracy, imprecision, and allowable total error were
available for more than 300 clinical laboratory tests in sectors as diverse as clinical chemistry, immunoassays, hematology cell count, hemostasis, or flow cytometry. There are
three levels of performance based on BV data: desirable, minimum, and optimum.
Tab. 2.2: Formulas used to define objectives for bias, imprecision and total error based on BV.
Level of
precision
Objectives for
bias
Objectives for imprecision
Objectives for TEa
Minimal
0.75 × CVi
0.375 × (CVi2 + CVg2)1/2
Desirable
0.5 × CVi
0.25 × (CVi2 + CVg2)1/2
Optimal
0.25 × CVi
0.125 × (CVi2 + CVg2)1/2
TEa= 1.65 × (0.75 × CVi) +
(0.375 × (CVi2 + CVg2)1/2)
TEa= 1.65 × (0.5 × CVi) +
(0.25 × (CVi2 + CVg2)1/2)
TEa= 1.65 × (0.25 × CVi) +
(0.125 × (CVi2 + CVg2)1/2)
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.1 Selection of quality requirements 7
The Ricos database provides bias, imprecision and TEa objectives based on BV desirable performance. Such APS should be used when the analytical CV (CVa) is less than
the half of the intra-individual CV (CVi). The minimum level of BV-based APS may be
used when CVa is less than three quarters of CVi. Finally, optimum precision may be
used for tests showing excellent precision, for which CVa is less than one quarter of CVi
[4]. Table 2.3 shows theoretical examples of the CVa/CVi ratio for thyroid-stimulating
hormone (TSH), free T4, creatinine, and respective TEa selection.
Tab. 2.3: Examples of CVa/CVi ratio and related analytical goals.
Test
TSH
fT4
Creatinine
CVa/CVi ratio
0.16
0.41
0.58
Analytical goal
BV optimum
BV desirable
BV minimum
TEa (%)
11.9%
8%
13.4%
It should however be mentioned that the selection of BV-based APS according to CVa/
CVi ratio has been arbitrarily defined and could lead to practical problems if used
without verification. Designing QC with for free T4 with 8% as an analytical will lead
to a complex multiple QC rule requiring multiple QC measurements. This is not necessarily cost-effective. The selection of analytical goals should always be verified by a
clinical pathologist, and a good question would be: “is this necessary to provide free
T4 results with a maximum of 8% variation, and is this practically achievable in the
laboratory with a cost-effective QC protocol?” In our laboratory, APS for free T4 is
based on BV minimum performance with a TEa of 12.1%. The clinical pathologist may
decide to use desirable BV-based analytical goals for every test, or to use minimal
precision for tests that too difficult to control using desirable precision (electrolytes
like K+, for example). Also, if the desirable allowable total error seems too large, it is
possible to use the optimum level, which will give a smaller analytical goal, which
may be better suited for clinically sensitive tests, like cardiac and tumor markers. Interestingly, BV-based AG have been criticized for being too stringent, or, in contrary, too
large [6].
BV not only has implications on setting quality specifications, but also on the
analysis of serial patient results [7]. The notion of critical differences, or reference
change values (RCV), is a very interesting concept that must be discussed in this
chapter because RCVs are calculated using analytical imprecision and BV data. Laboratory tests are not only used for diagnostic and screening purposes; the monitoring
of test results in acute (e.g. troponin, creatinine) or chronic (e.g. cholesterol, glycated hemoglobin) is also an important application of clinical laboratory medicine [7].
When monitoring serial patient results, the question is to know whether the observed
variation of test result is clinically significant or not. To answer this question, equations were developed by Fraser [4]. The theory behind the RCV concept is that, in
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
8 2 The elements of a conventional quality control protocol
order to be considered significant, the difference between serial test results should
be higher than the combined biological and analytical imprecision. Formulas to
compute significant (statistical p<0.05) and highly significant (p<0.01) RCVs are given
in table 2.4.
Tab. 2.4: Formulas to calculate RCV (or critical differences).
Critical difference (RCV)
p-Value
Significant difference
Highly significant difference
<0.05
<0.01
Formula
21/2 × 1.65 × (CVa2 + CVi²)1/2
21/2 × 2.33 × (CVa2 + CVi²)1/2
There are several articles that describe RCVs for tests in clinical chemistry [8], protein
testing [9], or other analytes. Critical differences for troponin have been the subject
of much research recently, because it is the example of a critical difference for which
a significant change in serial results has implications both in acute and chronic care
[10, 11]. Examples of RCVs coming from literature or from personal data for highly
sensitive troponin T, NT pro-BNP, total cholesterol, and potassium are provided in
table 2.5. It is not difficult for a laboratory to define its own RCVs because CVi data
are easily found in BV tables or in the literature, while analytical CV may be obtained
from QC records.
Tab. 2.5: Example of RCVs.
Test
hs-Troponin T
NT pro-BNP
Total cholesterol
Potassium
RCV-significant ­
difference (%)
(literature)
85
76
14.19
11.75
RCV-significant
difference (%)
(St. Luc Bouge data)
RCV-highly significant
difference (%)
(St. Luc Bouge data)
39.6
48.9
13.4
11.7
62
76.5
21
18.3
Table 2.5 shows that the RCV data obtained in St. Luc Bouge laboratory are quite close
to critical differences found in the scientific literature. Despite its introduction two
decades ago and its great potential in providing additional information to laboratory
results, the use of RCVs remain limited in clinical laboratories. This is probably attributable to the lack of interest in the concept of critical differences by clinicians and
limitations in laboratory information systems (LIS) to include RCV information on
laboratory reports.
The last strategy included in the Milan hierarchy is APS based on state-of-the-art.
This represents the best possible analytical quality that is achievable with current
laboratory instrumentation. It may be defined as the analytical performance achieved
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.1 Selection of quality requirements 9
by best laboratories (e.g. showing lower imprecision). State-of-the-art information
may be obtained by inter-laboratory comparison programs.
While they are excluded from the Milan consensus statement, other analytical
goals include limits of acceptable performance provided by external quality assessment (EQA)/proficiency testing (PT) providers. These limits, primarily designed
to classify participating laboratory performance as acceptable or unsatisfactory,
may be used to define quality requirement for clinical use of a test or for designing
a QC protocol. Examples of analytical goals include those provided by the Clinical
­Laboratory Improvement Act (CLIA) from USA, by the Scientific Institute of Public
Health (IPH) from Belgium, by Rilibäk Insitute from Germany or by the Royal College
of ­Pathologists of Australasia (RCPA). They still may be used when BV based-APS are
inappropriate or when state-of-the-art performance are lacking.
The different quality requirements show advantages and drawbacks that will
vary according to the laboratory tests on which they are going to be applied. Examples
of different quality specifications for frequent laboratory tests are shown in table 2.6.
Tab. 2.6: Examples or different analytical goals for frequently ordered laboratory tests.
BV desirable TEa
(Ricos table) (%)
Uric acid
Glucose
CA 125
TSH
12
6.9
35.4
23.7
State-of-the- d Value IPH
art TEa (%)
(Belgium) (%)
3.13
2.56
6.63
3.95
8
6.3
14
9.2 (>1.86 mU/L)
0.17 mU/L (<1.86
mU/L)
Rilibäk
(Germany) (%) CLIA (USA) (%)
7
11
N/A
13.5
17
10
N/A
3 SDs
All examples show that quality specifications based on state-of-the-art tend to be very
narrow. In these cases, state-of-the-art TEa was defined using our inter-­laboratory
comparison program (Unity Real Time; Bio-Rad, Hercules, CA, USA). Sometimes,
large differences are notable between analytical goals provided by different sources:
for CA 125, the quality requirement is 35.4% according to Ricos and 14% according to
the Belgian IPH, while Rilibäk and CLIA do not propose APS for this test.
Formulas used to compute BV-based APS, based on the linear sum of bias and
CV, have been criticized in the literature [6, 12]. One of the main critics of these classical formulas is the lack of theoretical basis for the summation of allowable bias
and allowable imprecision, resulting in a constant value of TEa that is often quite
large [6]. To overcome these limitations, a modification of the classical establishment
of BV-based TEa, adapted from Gowans’ model, was proposed [12]. This new model
defines quality specifications according to the medical use of the test (diagnosis or
monitoring) [12]. The impact of the application this new and potentially interesting
model has yet to be studied in the literature.
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
10 2 The elements of a conventional quality control protocol
2.1.1 References
[1] Fraser CG. The 1999 Stockholm Consensus Conference on quality specifications in laboratory
medicine. Clin Chem Lab Med. 2015;53(6):837–40.
[2] Sandberg S, Fraser CG, Horvath AR, Jansen R, Jones G, Oosterhuis W, et al. Defining analytical
performance specifications: consensus statement from the 1st Strategic Conference of the European
Federation of Clinical Chemistry and Laboratory Medicine. Clin Chem Lab Med. 2015;53(6):833–5.
[3] Horvath AR, Bossuyt PM, Sandberg S, John AS, Monaghan PJ, Verhagen-Kamerbeek WD, et al.
Setting analytical performance specifications based on outcome studies – is it possible? Clin
Chem Lab Med. 2015;53(6):841–8.
[4] Fraser C. Biological variation. From principles to practice. Washington, DC: AACC Press. 2001.
[5] http://www.westgard.com/biodatabase1.htm. Accessed 17 August 2016.
[6] Oosterhuis WP. Gross overestimation of total allowable error based on biological variation. Clin
Chem. 2011;57(9):1334–6.
[7] Fraser CG. Improved monitoring of differences in serial laboratory results. Clin Chem.
2011;57(12):1635–7.
[8]Bugdayci G, Oguzman H, Arattan HY, Sasmaz G. The use of reference change values in clinical
laboratories. Clin Lab. 2015;61(3–4):251–7.
[9] Katzmann JA, Snyder MR, Rajkumar SV, Kyle RA, Therneau TM, Benson JT, et al. Long-term
biological variation of serum protein electrophoresis M-spike, urine M-spike, and monoclonal
serum free light chain quantification: implications for monitoring monoclonal gammopathies.
Clin Chem. 2011;57(12):1687–92.
[10] Nordenskjold AM, Ahlstrom H, Eggers KM, Frobert O, Jaffe AS, Venge P, et al. Short- and longterm individual variation in cardiac troponin in patients with stable coronary artery disease.
Clin Chem. 2013;59(2):401–9.
[11] Simpson AJ, Potter JM, Koerbin G, Oakman C, Cullen L, Wilkes GJ, et al. Use of observed w
­ ithin-person
variation of cardiac troponin in emergency department patients for determination of biological
variation and percentage and absolute reference change values. Clin Chem. 2014;60(6):848–54.
[12] Oosterhuis WP, Sandberg S. Proposal for the modification of the conventional model for
establishing performance specifications. Clin Chem Lab Med. 2015;53(6):925–37.
2.2 Quality control rules
Quality control (QC) rules are statistical rules that are used to evaluate QC results.
These rules were developed by Westgard in the 1970s. Many QC rules are available,
some are simple, and some are more complex to use. They all have different capabilities to detect significant analytical errors; this capability is called Ped, for probability
of error detection. They have also different probability that their application may lead
to the blockade of results release, due to a statistical, but not analytical problem; this
phenomenon is called Pfr, for probability of false rejection. Ped and Pfr are ­determined
using power function graphs, which is easily done using different QC software. Although QC rules were developed 40 years ago, many problems are still ­associated with
them. First, difficulties with their selection are frequently ­encountered. Some laboratory managers decide to use multiple QC rules for every test: although this approach
offers some standardization and guarantees that a minimal number of unreliable
patient results will be generated, the use of multiple QC rules for each test may lead to
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.2 Quality control rules 11
an inadequately high probability of false rejection. This will have the effect of slowing
down the whole laboratory activity because technologists will look to solve analytical
problems that, in fact, do not exist. In contrary, it was also recently shown by different
authors [1, 2] that many laboratories still use a unique QC rule, generally 12s. This will
also generate an excessively high false rejection rate. A second problem with QC rules
is that they have been defined 40 years ago, when most laboratory analyzers were
operating in batches. It is not the case for high-throughput analyzers, like the ones
encountered in chemistry, hematology, and hemostasis.
Although the specimen throughput capacity and reliability of analyzers have
greatly improved over the years, the statistical QC rules have remained the same: the
capabilities of modern analyzers and the design of modern laboratories have created
challenges for QC rules application. This chapter will describe the different available
QC rules and discuss their implementation in the laboratory with practical examples
coming from the daily routine. A section will discuss the graphs and metrics that are
needed to characterize and select optimal QC rules and will also show the relationship between QC rules and Sigma metrics.
QC rules have a different sensitivity to random and systematic error. Random
errors are caused by unexpected and unpredictable changes in the measurement
procedures. Causes of random errors in laboratory medicine include air bubbles in
samples or reagent packs, obstructed pipetting systems, and fluctuations in the electrical power supply [3–5]. Systematic errors are caused by a systematic change in the
measurement system that will result in a shifted mean. Examples of systematic errors
in clinical laboratories include change in reagent, control, and calibrator lot numbers,
deterioration of reagents, and variation of temperature [3, 5, 6]. Table 2.7 describes
the QC rules frequently used in clinical laboratories, their respective sensitivities to
random or systematic error, and the probable laboratory causes for these errors.
One of the most frequently used QC rule in laboratory QC protocols is 13s. It defines
QC failure as a QC value of + or –3 standard deviations (SDs)
as compared to the mean.
The 13s rule offers a good Ped and a low Pfr for tests showing a good performance (high
Sigma metric). It is generally a good QC rule to start with. If the analytical performance of a test is very good as compared to the analytical quality requirement (e.g.
high Sigma metric), then a less stringent rule may be used. 13s is sensitive to random
errors and large systematic errors [7].
1ks (13.5s, 14s, 15s) rules are similar to the 13s rule. They may be used for high Sigma
metric tests. Applied in this context, they offer a high Ped (generally >98%) and a lower
Pfr as compared to 13s. However, analyzers and QC software rarely permit to use rules
wider than 15s (16s, 18s, etc.), while theoretically, it make sense to use a 110s for a test
with a Sigma value of 20. However, in the case of overenthusiastic Sigma metrics, the
choice of the quality requirement may certainly be questioned. 1ks rules are sensitive to
large random errors [7], but they have to be used only for tests with high Sigma metrics.
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
12 2 The elements of a conventional quality control protocol
Tab. 2.7: Statistical rules used in clinical laboratory QC.
QC rules
Interpretation
Sensitivity to
13s
QC failure occurs when Random error
one QC result is ±3 SD
limits
1ks (13.5s, 14s, 15s) QC failure occurs when
one QC result is ±k SD
limits
R4s
QC failure occurs
when there is a 4 SD
difference between
two consecutive
control measurements
22s
QC failure when 2
Systematic
successive QC results error
exceed the same +2 or
–2 SD limits
2 of 32s
QC failure when two out
of three consecutive QC
results are located + or
–2 SD from the mean
31s, 41s
QC failure when 3 (31s)
of 4 (41s) consecutive
QC results exceed the
same +1 or –1 SD
7t
QC failure when
7 consecutive QC
results move upwards
or downwards
8x, 10x, 12x
QC failure when 8 (8×),
10 (10×), or 12 (12×)
consecutive QC results
are located on the
same side of the mean
12s
QC failure occurs
when one QC result is
+/–2 SD limits
12s, single
repeat
QC failure occurs
when the first QC
result is +/–2 SD
limits and repeat
testing of the same QC
is also +/–2 SD
Possible error cause
––
––
––
––
––
––
Clogged pipettors
Obstructed rinsing needles
Air bubbles in reagent
Insufficient mixing of reagent
Incorrect fitting of disposable tips
Variation of power supply
––
––
––
––
––
––
––
Deterioration of QC material
Reagent lot change
Calibration bias
Evaporation of reagent of calibrator
Reagent cross-contamination
Alteration of light source
Change in environmental
conditions (increase or decrease in
temperature and humidity)
–– Deterioration of water quality
Random and
systematic
errors
All random or systematic causes
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.2 Quality control rules 13
The 12s rule, while similar in conception to the ones discussed previously, must be treated
differently because its use will lead to many practical problems in the laboratory, mainly
related to its high Pfr. 12s has been used in the past as a rejection rule: if one QC measurement is two standard deviations above or below the mean, then, all patient results
from the precedent QC result are rejected. Used in this context, the 12s rule will lead to
a very high number of false rejections. The Pfr of 12s increases with the number of QC
measurements, as is the case with other QC rules. If rule 12s is used with three control
measurements, the Pfr is going be close to 15%. It means that using 12s with three control
measurements in a shift for 60 tests in a corelab will generate nine QC failures that will
have to be investigated. This is way too many. For this reason, the use of 12s should be
discouraged. Because the Pfr of 12s is so high, many people have used it not as a rejection
rule but as a warning rule. In this case, what will happen is that technologists will not
even see the 12s rule violation because it will happen continuously. The use of the 12s rule
as a rejection or warning rule should be discouraged. If a QC selection process using QC
software leads to the selection of the 12s rule for a test, it is probable that the analytical
performance of the test is not good enough or that the analytical goal is too stringent.
A wiser application of that procedure would be consist in using the 12s rule, and,
in the case of 12s violation, to repeat immediately the same QC sample. This QC rule is
called 12s, single repeat and has been proposed by Parvin [1]. The author postulated that
this practice was common in American laboratories and has studied Ped and Pfr of this
rule in comparison with other QC rules. Interestingly, this work showed that this rule
had a low PFr and a Ped similar to the 13s rule. One drawback of the 12s single repeats is
that it uses additional QC material, and thus, it increases the costs associated with QC.
22s aims at rejecting patient results when two consecutive QC results are above or
below 2 SDs each. It will detect systematic error [7]. 22s is frequently associated with
13s and R4s in multiple QC rules. Another rule sensitive to systematic error is the 2 of
32s QC rule. In this case, a QC failure happens when two QC results out of three consecutive measurements are located +2 or -2 SDs compared to the mean. Although it is a
rule that is frequently proposed in analyzer and QC software, its use seems to remain
limited, probably because it is a rule that is less intuitive than other rules.
The R4s rule aims at rejecting results when two consecutive QC results have a 4 SD
difference. This may occur even if all QC results fall between the ±3 SD control limits.
It is a powerful rule; however, it is infrequently encountered in clinical laboratories. 31s
and 41s are similar rules that are sensitive to systematic errors. Rules like 8x, 10x, and 12x
detect a QC failure when 8, 10, or 12 results are located on the same side of the mean.
These rules generally show a high Pfr. They are going to be frequently in infraction for
tests performed with reagents showing significant lot-to-lot variation; therefore, some
authors discourage their use [4]. Another rule that is sensitive to systematic errors is
the 7t rule. It detects QC failure when seven consecutive QC results are moving upwards
or downwards. The impact of this rule has been insufficiently studied in the literature.
Also, we can imagine that similar rules like 5t, 10t, or 12t could be interesting, but these
are not available in analyzers or QC software. Although this 7t rule seems interesting,
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
14 2 The elements of a conventional quality control protocol
particularly when it is used in conjunction with wider rules like 15s for high Sigma metric
tests, it may also lead to some frustrating situations: imagine the case of a 7t rule that is
triggered while all QC results are very close to the mean. In this case, it is difficult for QC
analysts to understand why a QC failure is triggered.
QC rules may be used within a QC level and some rules (like 41s and 22s) may be
used between levels. In this case, the 41s rule is triggered when two QC measurements
of two different levels are + or – 1 SD compared to the mean; 22S may be triggered when
one QC measurement of two different QC concentrations are at +2 or −2 SDs. This
inter-level application of QC rules may offer quicker detection of QC failure, but it is
not applicable to every QC rule and it is not proposed by all softwares.
Two important metrics are used to describe the performance of QC rules: Ped and
Pfr [8, 9]. Ped is the probability of detecting an error that is significant on an analytical
point of view. Pfr represents the risk to detect a QC failure that is due to a statistical
incident and not to the occurrence of a serious error that is medically significant. These
two metrics are determined using a power chart. This chart plots Ped as a function of
the critical systematic error, noted ∆SE crit, which represents the critical error that has
to be detected by the QC procedure. Plotting of power charts is routinely possible using
different available software programs (EZ Rules 3 from Westgard QC Inc, Madison, WI,
USA, or Westgard Advisor, a module in Unity Real Time™, from Bio-Rad Laboratories,
Hercules, CA, USA). Figures 2.1 and 2.2 provide examples of power function graphs. To
understand how a power chart is plotted, the reader is advised to read the publication
of Hyloft Petersen, which provides several examples of power chart plotting [10].
Fig. 2.1: Power chart for CK-MB, the 15s rule, Ped 100%, and Pfr 0%.
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.2 Quality control rules 15
As there is a link between the Sigma metric equation and the ∆SE crit, Sigma metrics
are closely related to QC rules.
∆SE crit = Sigma metric – 1.65
or
Sigma metric = ∆SE crit + 1.65
Figure 2.1 shows the example of a power chart for CK-MB, a test which shows an excellent performance: in this case, Sigma metric is 8.60 and ∆SE crit is 6.95. The QC rule
automatically selected by the software in use at our laboratory (Westgard Advisor,
Unity Real Time™) is the 15s rule with a Ped of 100% and a Pfr of 0%. Unfortunately,
not all tests will show similar performance and will have to be managed with QC rules
that are more complex and that generates an higher value for Pfr. Figure 2.2 shows the
power function chart for HDL-cholesterol; for the test, using BV, the desired precision
as quality specification of the Sigma metric is 3.5 and ∆SE crit is 1.85, the selected
QC rule is a multiple rule using 13s, 2of32s, R4s, 31s and 12x, resulting in a Ped of 96.8%
but a considerable Pfr of 8.9%.
Fig. 2.2: Power chart for HDL, Multiple QC rule, Ped 96.8%, and Pfr 8.9%.
An alternative strategy to this multirule would be to use 12s, single repeat, or to define
an alternative quality goal, for example, BV minimum precision. This will lead to a
simpler QC strategy, and the level of quality guaranteed for the test would be lower.
Because of its link with ∆SE crit, Sigma metrics offer a simple single approach to
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
16 2 The elements of a conventional quality control protocol
select appropriate QC rule for any specific test. Authors have proposed a guideline for
the selection of QC rule based on Sigma metrics [11]. It also provides a QC frequency
recommendation. While this simple guideline offers Sigma metric-based QC that is
easily implementable, there is little doubt that the definition of specific QC rules for
each laboratory test using power charts will provide a better QC strategy; however,
this process takes time and need to be repeated at a regular interval or when significant changes occur, such as implementation of a new QC material or a new QC lot.
Previously, some QC rules were used as a rejection rule, and some, especially
12s, were used as a warning. In our experience, warnings are not really helpful and
the use of the 12s rule should be discouraged. Therefore, the recommendation is to
design QC rules using power charts or Sigma metrics, to apply the rules that are
provided, and to consider every rule violation as a QC failure. This does not mean
that all patient samples from the previous QC have to be retested, but that a study of
patient impact has to be performed in case of QC failure. This process is discussed
in section 2.5.
The classical view of QC results is the plotting of a Levey-Jennings (LJ) chart.
Fig. 2.3: Control chart for CA 125.
The conventional LJ chart shows the mean and the ±1, 2, and 3 SDs. In some software
programs, it is possible to “zoom out” in order to view the chart with, for example,
±4, 5, or 6 SDs.
Figure 2.3 provides a conventional LJ chart for CA 125. Some software programs
offer the possibility to view the LJ chart without the conventional limits, but to add
the quality specifications (TEa) as control limits.
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.2 Quality control rules 17
Fig. 2.4: Conventional LJ graph for creatinine.
While figure 2.4 represents a conventional LJ chart for creatinine, the corresponding
LJ chart with TEa as boundaries is shown in figure 2.5.
Fig.2.5: TEa LJ graph for creatinine.
This view may be helpful in the determination of the importance of the QC failure;
however, the confirmation of the occurrence medically significant error has to be performed with patient specimens and not with QC samples.
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
18 2 The elements of a conventional quality control protocol
Ceriotti et al. [12] have recently proposed an alternative model to follow QC results
and to define QC failure. This model incorporates the concept of measurement uncertainty (MU). In the proposal of Ceriotti et al., QC results are followed on a graph with
an acceptance zone based on MU and absolute limits are based on selected TEa. The
incorporation of MU and TEa on QC graphs is an interesting development to be followed in the future.
2.2.1 References
[1] Parvin CA, Kuchipudi L, Yundt-Pacheco JC. Should I repeat my 1:2s QC rejection? Clin Chem.
2012;58(5):925–9.
[2] McFarlane A, Aslan B, Raby A, Moffat KA, Selby R, Padmore R. Internal quality control practices
in coagulation laboratories: recommendations based on a patterns-of-practice survey. Int J Lab
Hematol. 2015;37(6):729–38.
[3] Kinns H, Pitkin S, Housley D, Freedman DB. Internal quality control: best practice. J Clin Pathol.
2013;66(12):1027–32.
[4]Miller WG. Quality control. In: McPherson RA, Pincus MR, editors. Henry’s clinical diagnosis
and management by laboratory methods. 22nd ed. Philadelphia, PA: Elsevier/Saunders;
2011.
[5] Cembrowski GS, Smith B, Tung D. Rationale for using insensitive quality control rules for
today’s hematology analyzers. Int J Lab Hematol. 2010;32(6 Pt 2):606–15.
[6]Miller WG, Erek A, Cunningham TD, Oladipo O, Scott MG, Johnson RE. Commutability
limitations influence quality control results with different reagent lots. Clin Chem.
2011;57(1):76–83.
[7] Cembrowski GS, Clarke G. Quality control of automated cell counters. Clin Lab Med.
2015;35(1):59–71.
[8]Westgard JO. Statistical quality control procedures. Clin Lab Med. 2013;33(1):111–24.
[9] Westgard JO, Westgard SA. Quality control review: implementing a scientifically based quality
control system. Ann Clin Biochem. 2016;53(Pt 1):32–50.
[10] Petersen PH, Ricos C, Stockl D, Libeer JC, Baadenhuijsen H, Fraser C, et al. Proposed guidelines
for the internal quality control of analytical results in the medical laboratory. Eur J Clin Chem
Clin Biochem. 1996;34(12):983–99.
[11] Cooper G, DeJonge N, Ehrmeyer S, Yundt-Pacheco J, Jansen R, Ricos C, et al. Collective opinion
paper on findings of the 2010 convocation of experts on laboratory quality. Clin Chem Lab Med.
2011;49(5):793–802.
[12] Ceriotti F, Brugnoni D, Mattioli S. How to define a significant deviation from the expected
internal quality control result. Clin Chem Lab Med. 2015;53(6):913–8.
2.3 Sigma metrics
Six Sigma is a management methodology that was developed by Motorola in the
1980s. Since then, this breakthrough methodology has been applied by many companies with tremendous success in terms of cost savings and overall profitability. Six
Sigma principles aim to measure quality on an industrial scale, so that quality can be
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.3 Sigma metrics 19
evaluated in a more quantitative fashion. Each Sigma metric corresponds to a certain
value of defects per million opportunities (DPMOs). There are two different Sigma
metrics: short term and long term (table 2.8). Long-term Sigma metrics include the
possibility of the occurrence of a 1.5 SD shift that will give larger DPMOs as compared
to the short-term Sigma metric.
Tab. 2.8: The Sigma scale.
Sigma metric
Error rate (DPMO)
Error rate with 1.5 SD
shift (DPMO)
Error rate with 1.5 SD
shift (%)
1
2
3
4
5
6
317,400
45,400
2700
63
0.57
0.002
691,462
308,538
66,807
6210
233
3.4
69
31
6.7
0.62
0.023
0.00034
Six Sigma can be applied in laboratories [1]. Authors have used the DMAIC (define,
measure, analyse, improve, control) strategy to reduce data entry errors and to
better define the variation of a pneumatic tube system in the pre-analytical phase
or to reduce post-analytical errors [1]. An important application of Six Sigma in
laboratories is to allow the performance evaluation of laboratory tests on the
Sigma scale [1]. An equation, derived from the process capability concept, was
popularized by Westgard 10 years ago and allows to define Sigma metrics for laboratory tests.
Sigma metric = (Quality specifications – |Bias|)/Test variation
or
Sigma metric = (TEa – |Bias|)/CV
Sigma metrics are computed using data coming from measurement of QC ­materials
and may be used as quality indicators that represent the balance between quality
requirements (TEa) and test variation (bias and CV). Quality requirements are easily
obtained from tables or from the literature, whereas CV data are readily available
in clinical laboratories. The determination of bias is more problematic. The classical definition of bias implies the calculation of the difference between the results of
a reference material being assayed with a routine instrument and with a reference
method. Unfortunately, reference materials are not widely available. There are three
options available regarding inclusion of bias in a Sigma metric equation: [1] including bias calculation obtained with a reference material, when available; [2] including bias value provided with inter-laboratory comparison with a peer group; [3]
considering bias as equal to zero. This last option is a possibility when instrument
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
20 2 The elements of a conventional quality control protocol
manufacturers try to minimize bias by providing calibration traceability to a reference
method, which is often the case [2]. However, withdrawing bias will have a significant
impact on Sigma metrics, and therefore, on QC rules that are going to be chosen. This
Sigma metric equation has many applications in clinical laboratory medicine. The
Sigma metric of a test is a measure of how many standard deviations will fit into its
quality requirement. If the Sigma metric of a test is very high, it means that its analytical performance is good [3], and it will need a very important shift in its measurement
in order to generate a result that will not fit in its quality requirement (a result that will
exceed his TEa requirements). Important shifts are easily detected by simple quality
control rules (13s, 14s, even 15s) with a limited number of control measurements. In contrast, tests with a low Sigma metric have limited analytical performance [3], and small
shifts in the measurement may generate a result that exceeds TEa specification. Small
shifts cannot be detected using a simple QC rule; in the case of low Sigma metric tests,
complex, multiple QC rule implying multiple control measurements have to be used.
During a consensus conference, authors have defined categories of tests using Sigma
metrics, and have proposed QC rules to be applied according to the related Sigma
metric [4]. These recommendations are summarized in table 2.9.
Tab. 2.9: Definition of QC rules according to Sigma metrics.
Sigma metric
Category
Proposed QC
rule
Proposed QC frequency
Additional
action
>6
4–6
13,5s
12.5s
<3
Problematic tests
One QC per day
Two levels of QC
per day
Two levels of QC
two times a day
Three levels of QC
three times a day
None
None
3–4
Excellent tests
Tests that are suited
for purpose
Poor performers
13s 22s R4s 41s
10×
13s 22s R4s 41s
10×
None
Consider
testing patient
specimens in
duplicate
This concept is simple to implement in a laboratory and provides a good approach for
QC standardization; however, while it provides QC guidelines based on test performance, it does not take into account the specific risk level for a test. It also does not
take into account the frequency performance measures that were recently introduced
[5–7]. Sigma metrics may be utilized also to compare test performance between different analyzers or methods and to monitor analytical performance over time [8]. This
quality indicator may be part of an integrated laboratory dashboard. Sigma metrics
have also been applied to EQA/PT [9, 10].
While the Sigma metric equation is simple, it needs to be discussed. It is important to note that Sigma metrics and resulting QC rules are relative to the TEa, bias,
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.3 Sigma metrics 21
and CV chosen in the equation. Sigma metrics will be very different if the TEa selected
is based on BV or on state-of-the-art data coming from an EQA program. The same
is true regarding bias and CV selection. Table 2.10 shows the impact of different TEa
selection for potassium, glucose, CA 19.9. and TSH. Using BV with desirable precision
as quality specification, the four tests exhibit excellent Sigma metrics.
Tab. 2.10: Impact of TEa selection on Sigma metrics.
Test
K+
Bias (%) CV (%) TEa
#1
–0.306 0.517 BV
des
Glucose 0.48 0.752
CA 19.9
0.99 2.73
TSH
–2.99 1.87
TEa #1 Sigma TEa #2 TEa #2 Sigma TEa #3
(%)
#1
(%)
#2
5.61
10.3
6.96
46.2
23.7
8.62
16.6
11.1
BV opt
2.8
3.48
23.1
11.9
4.82 IPH
Belgium
3.99 RCPA
8.1 RiliBäk
4.78 GOST
TEa #3 Sigma
(%)
#3
4.8
8.7
8
14
28.5
10
4.8
13.7
The choice of TEa for CA 19.9, with a value of 46.2%, may be questioned. In the case
of CA 19.9 measurement, it is probably more cautious to select BV optimal precision
with a TEa value of 23.1%. In that case, the resulting Sigma metric is 8.1, which is
still an excellent value. Using BV optimum precision or the other tests will have an
impact, as the performance will switch from excellent to moderate for potassium,
glucose, and TSH. The third TEa chosen shows an example of alternative analytical
goals, such as those provided by the Institute of Public Health (IPH) from Belgium,
the Royal College of Pathologists of Australasia (RCPA), the German institute Rilibäk
and the Russian institute GOST. Such analytical goals may be helpful when no analytical goals based on BV are available or when they are impossible to achieve. The
selection of the quality specification remains the choice of the clinical pathologist,
and the justification for the choice on a TEa has to be recorded in the quality management system (QMS). Possible value for bias include comparison with a reference
method, with a peer group (laboratories working with the same platform, method,
and reagent), with a group using the same method, or all laboratories participating
in an inter-laboratory comparison program for a specific test. Different bias values
may also be generated if the data taken into account are cumulative over time or if
the value of bias that is chosen is the one occurring over a single month. Table 2.11
shows the impact of bias on Sigma metric for the four tests that were previously
described in the tables about TEa selection. In the first situation, no bias is taken
into account for the Sigma metric calculation. In the second and third examples,
bias in comparison to the peer group (laboratories using same method, analytical
platform, and reagents) or in comparison with the method group (laboratories using
same method, irrespective of manufacturer) is taken into consideration for computation of Sigma metrics.
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
22 2 The elements of a conventional quality control protocol
Tab. 2.11: Impact of bias selection on Sigma metrics.
Test
TEa
K+
Glucose
CA 19.9
TSH
BV des
BV des
BV opt
BV des
TEa (%)
5.61
6.96
23.1
23.7
CV
0.517
0.752
2.73
1.87
Bias #1, Sigma Bias #2,
no bias #1
peer
group
0 10.8
0 9.25
0 8.46
0 12.7
−0.306
0.48
0.99
−2.99
Sigma
#2
10.3
8.61
8.1
11.1
Bias #3, Sigma #3
method
group
1.44
0.862
0.931
–2.88
8.06
8.11
8.12
10.8
Interestingly, for potassium, the peer-group bias is slightly negative, whereas the
method-group bias is positive. For the four tests, the impact of bias on Sigma metrics,
and therefore, on evaluation of test performance, remains limited. Regarding CV
selection, Sigma metrics will differ according to the number of data points that are
considered for CV determination. Sigma metrics will also differ according the concentration of the QC material that is used. Generally, a QC material with low analyte concentration will show a lower Sigma metric, because imprecision is frequently higher
at low concentration. Table 2.12 shows the impact of using a 2- or 6-month cumulative
coefficient of variation (CV). In the case of 6-month cumulative CV, the Sigma metric
for CA 19.9 and TSH will drop a few units, but this remains nonsignificant on the
­performance classification of these two tests.
Tab. 2.12: Impact of CV selection on Sigma metrics.
Test
TEa
K+
Glucose
CA 19.9
TSH
BV des
BV des
BV opt
BV des
TEa (%)
5.61
6.96
23.1
23.7
Bias
–0.306
0.48
0.99
–2.99
CV #1, 2-month
cumulative
Sigma #1
CV #2, 6-month
cumulative
Sigma #2
0.432
0.635
1.78
1.29
12.2
10.2
12.4
16
0.517
0.752
2.73
1.87
10.3
8.61
8.1
11.1
A 6-month CV is considered representative of true test variation because many different laboratory technologists will perform QC measurements, and significant events,
such as calibration or manufacturer maintenance will occur in that timeframe. Unfortunately, there are currently no guidelines published regarding the Sigma metric calculation. In our experience, we favor the computation, for each level of QC material,
of Sigma metrics using BV as quality requirement, monthly bias in comparison to
the peer group issued from the inter-laboratory comparison program, and long-term,
6-month cumulative CV obtained on at least 180 data points from routine QC material
measurement. In recent years, several authors have discussed the practical implementation of Sigma metrics in clinical laboratories [1, 3, 11]. An interesting article has
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.3 Sigma metrics 23
been published by Schoenmakers et al. in 2011 [11]. In their work, the focus was on
using Sigma metrics in order to select appropriate QC rules for several clinical chemistry tests that were run on four different analyzers. The allowable total error was
based on BV data, the CV was a long-term CV extracted from the internal QC software
program, and bias was calculated in comparison with a reference method. When no
reference method was available, bias was computed using difference with the peer
group value. The resulting Sigma metrics were excellent (>6) for many tests, including
enzymes such as CK, ALAT, gamma-GT or amylase; the performance was intermediate
(Sigma metrics between 4 and 6) for glucose, cholesterol, and potassium, whereas
the performance was unacceptable (Sigma metrics <3) for tests such as magnesium,
calcium, chloride, and sodium. This first paper on the application of Sigma metrics for
clinical chemistry showed the feasibility of determining QC rules based on the Sigma
metric equation for frequent clinical chemistry tests and has also identified several
difficulties encountered with the use of this concept. First, the CV used in the Sigma
metric equation is concentration dependent. Therefore, Schoenmakers et al. [11] have
concluded that it is better to compute different CVs at different concentrations and
to compute a mean CV to input in the Sigma metric equation. Second, determination
of bias is a significant problem. If there is no reference method available, it is necessary to turn to an inter-laboratory comparison program, and sometimes, a bias can be
positive at one concentration level and negative at another. Therefore, using a mean
bias is not recommended, but this fact renders the use of an average Sigma level more
complex. Moreover, in modern laboratories, multiple analyzers performing the same
test are frequently encountered. The authors have also mentioned that a laboratory
test being performed in a modern corelab is frequently the result of a virtual analyzer
composed of several individual instruments. Therefore, virtual biases, CVs, and resulting Sigma metrics may also be computed. Despite these different drawbacks, the
authors have concluded that a QC roadmap based on Sigma metrics is a fast and easy
way to individualize QC rules in a chemistry laboratory, resulting in potential cost
savings [11]. This statement has been confirmed in a recent report, where the implementation of Sigma metrics for the selection of QC rules with a lower probability of
false rejection (Pfr) has allowed a decline of 75% in run rejection for high-volume clinical chemistry parameters [8]. Another article published by our Belgian colleagues has
investigated the performance of Abbott Architect® systems for 41 commonly ordered
analytes, using Sigma metrics [3]. To input in the Sigma metrics equation, CVs were
determined using the CLSI EP5-A2 guideline. Biases were also included in the equation and were defined in comparison with the target value provided in the insert kit of
the control materials. Sigma metrics were determined for two concentrations of independent, third-party control material. In this article, authors have studied in details
the importance of the TEa selected. Three different analytical goals were used: BV,
Rilibäk and Clinical Laboratories Improvement Act (CLIA). Sigma metrics obtained
using Rilibäk as TEa were mostly excellent, except for tests like lithium, creatinine
and calcium, due to high biases occurring at one concentration. Many tests showed
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
24 2 The elements of a conventional quality control protocol
impressive Sigma metrics using BV-based TEa, but several tests exhibited bad performance for one QC level (IgG, total protein, alkaline phosphatase, etc.) or for both QC
levels (magnesium, potassium, sodium, albumin, bicarbonate, calcium and transferrin). Interestingly, several of these tests showed also Sigma metrics <3 in the paper
from Schoenmaker et al. [11]. Using TEa values based on CLIA, which are known for
being less stringent, most analytes showed excellent performance. Notably, Sigma
metrics encountered on several similar analyzers were comparable. Hens et al. [3]
concluded that BV-based TEas are sometimes too demanding and that mixing sources
of allowable total error in a clinical laboratory could probably be a correct way to set
appropriate APSs. Moreover, Sigma metrics are a valuable tool to monitor the performance of several analyzers [3, 8]. In a very recent article, Coskun et al. [12] suggested
that in the analytical phase, Sigma metrics should not include the extra 1.5 SD shift
that is usually used to compute the corresponding DPMOs [12]. They have also proposed to express Sigma metrics as an abbreviation with the corresponding DPMOs as an
index. An example is as follows:
SM (DPMO) = 3.5 (460)
Time will tell if this new way of expressing Sigma metrics will facilitate its use and
overcome some hurdles that are limiting its understanding and implementation in
clinical laboratories.
2.3.1 References
[1] Gras JM, Philippe M. Application of the Six Sigma concept in clinical laboratories: a review.
Clin Chem Lab Med. 2007;45(6):789–96.
[2] Cembrowski GS, Cervinski MA. Demystifying reference sample quality control. Clin Chem.
2016;62(7):907–9.
[3] Hens K, Berth M, Armbruster D, Westgard S. Sigma metrics used to assess analytical quality of
clinical chemistry assays: importance of the allowable total error (TEa) target. Clin Chem Lab
Med. 2014;52(7):973–80.
[4] Cooper G, DeJonge N, Ehrmeyer S, Yundt-Pacheco J, Jansen R, Ricos C, et al. Collective opinion
paper on findings of the 2010 convocation of experts on laboratory quality. Clin Chem Lab Med.
2011;49(5):793–802.
[5] Parvin CA. Assessing the impact of the frequency of quality control testing on the quality of
reported patient results. Clin Chem. 2008;54(12):2049–54.
[6] Yundt-Pacheco J, Parvin CA. Validating the performance of QC procedures. Clin Lab Med.
2013;33(1):75–88.
[7] Yago M, Alcover S. Selecting statistical procedures for quality control planning based on risk
management. Clin Chem. 2016;62(7):959–65.
[8]Jones JB. A strategic informatics approach to autoverification. Clin Lab Med. 2013;33(1):
161–81.
[9] Westgard JO, Westgard SA. The quality of laboratory testing today: an assessment of Sigma
metrics for analytic quality using performance data from proficiency testing surveys and the
CLIA criteria for acceptable performance. Am J Clin Pathol. 2006;125(3):343–54.
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.4 Quality control frequency 25
[10] Westgard JO, Westgard SA. Assessing quality on the Sigma scale from proficiency testing and
external quality assessment surveys. Clin Chem Lab Med. 2015;53(10):1531–5.
[11] Schoenmakers CH, Naus AJ, Vermeer HJ, van Loon D, Steen G. Practical application of Sigma
Metrics QC procedures in clinical chemistry. Clin Chem Lab Med. 2011;49(11):1837–43.
[12] Coskun A, Oosterhuis WP, Serteser M, Unsal I. Sigma metric or defects per million opportunities
(DPMO): the performance of clinical laboratories should be evaluated by the Sigma metrics at
decimal level with DPMOs. Clin Chem Lab Med. 2016.
2.4 Quality control frequency
Designing a quality control (QC) protocol using power function graphs and OPSpec
charts will generate important QC metrics such as the probability of error detection
(Ped), probability of false rejection (Pfr), and the number of control measurements
(noted N). N represents the number of control measurements that needs to be available when making a decision regarding the evaluation of a specific QC rule. While these
QC metrics are certainly important, they will not propose the optimal QC frequency.
Because modern automated laboratories perform so many tests with different performances and different risk levels, it is impossible to design a universal QC schedule
(e.g. two QC concentrations, run twice a day, on 7 AM and 3 PM) for every test in every
laboratory department. QC materials represent an important cost for laboratories
and QC frequency will have an impact on the laboratory financial budget. The selection of QC frequency will also have an important impact in the daily routine of MLTs.
Therefore, QC frequency has been the subject of much discussion and articles have
proposed guidelines for the frequency of running QC materials [1–3]. Seminal papers
discussing new concepts and metrics related to QC frequency have also appeared in
the peer-reviewed literature [4–7]. The objective of this section is to provide information about different concepts that have to be integrated in order to select the optimal
QC frequency in different situations.
Factors that influence QC frequency are summarized in table 2.13. These include
analytical system events, test performance, risk level, frequency metrics, and regulations. It is important to run QC after every significant change in the test system. The
events include routine maintenance, company maintenance, reagent lot changes, introduction of a new reagent pack, and calibration [2]. While it is intuitive to run a QC
sample after every significant change in the instrument, authors have also noted that
QC has also to be performed before any important action [2, 8]. If a QC value significantly changes after calibration or maintenance and that no QC was run before these
actions, it will be impossible to determine if the QC change is caused by the event or
was already existing before [2]. QC examination following any significant action on
an instrument is also a requirement in ISO 15189:2012 [9]. However, after implementation of a new reagent lot, the verification process should be performed using patient
specimen and not with QC materials, as these may show some non-commutability
with patient samples [10]. As a consequence from non-commutability, QC results may
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
26 2 The elements of a conventional quality control protocol
shift significantly when patient results do not. This phenomenon may be seen irrespective of the type of QC used (lyophilized or liquid, manufacturer or independent QC
material) [10].
Tab. 2.13: Factors that influence QC frequency.
Factors
Comment
System events, such as maintenance,
calibration, reagent lot change,
introduction of a new reagent pack
Test performance
QC has to be run before and after any significant system
change
Number of patient samples being
processed
Frequency metrics
Regulations
–– Test performance may be quickly estimated by the
computation of Sigma metric.
–– All other factors (number of patient processed, risk and
frequency metrics) being constant, tests with higher
Sigma metrics will require less frequent QC than tests
with a low Sigma metric
High volume tests will require more frequent QC than low
volume tests
QC frequency may be selected using expected
increase in the number of incorrect results,
E(Nuf) and E(Nuc)
Some countries (e.g. USA) have regulations about the
frequency of QC
Test performance is an important factor to take into account when selecting QC
frequency. Test with good performance (high Sigma metric tests) will require less
­frequent QC as compared to tests with marginal or low performance (low Sigma
metric). A consensus paper proposed recommendations on QC rules and QC frequency based on Sigma metrics [1]. While this approach is interesting and easy to
implement, it is a little oversimplistic and do not consider the number of patient
samples being run. It also does not include the new frequency metrics that were
recently introduced [4, 6, 7].
Testing volume is intuitively an important factor to consider. If an analyte is performed 500 times a day, it will require a higher QC frequency compared to a specialized test that is run only 10 times a day. In their QC harmonization paper, Jones et al.
[2] have proposed that for low volume testing (less than 200 tests/day), QC should be
performed before every work shift (e.g. every 8 hours). For high volume testing, the
authors suggest samples to be run every 250 patient samples. This recommendation
does not include consideration of test performance, level of risk, or newly introduced
frequency metrics.
The risk level of a test is also an important factor at play. If a test is critical and
immediately acted upon (e.g. troponin, glucose, creatinine), QC frequency has to be
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.4 Quality control frequency 27
higher as compared to tests with lower risk, or that are not immediately acted upon
(e.g. AST, LDH). Defining risk level may be easily performed using a risk matrix, such
as the one presented in CLSI EP23-A document [11]. However, this process remains
a little arbitrary. The assessment of laboratory risk and resulting risk matrixes are
discussed in chapter 6. A useful tool to design optimal QC frequency, easily obtainable in a laboratory, is the ratio between patient samples and QC materials analyzed.
Examples of this ratio are provided in table 2.14 and will be necessary to use advanced frequency metrics. For potassium, 324 patient samples are tested per day in our
laboratory, on four different ion-specific electrode (ISE) modules. Three levels of QC
materials are run three times a day on each ISE module. Therefore, the ratio between
patient and QC samples is 9.
Tab. 2.14: Examples of ratio between QC and samples analyzed.
Test
Patient Patient
QC levels Times when Number of QC
Ratio patient/
samples samples per
QC samples samples per day
QC samples
per day day and per
are run
and per instrument
instrument
Potassium
324
81
Calcium
194
97
TSH
175
87
14
25
14
25
CA 19.9
CT/NG PCR
3 7 am
5 pm
12 am
3 7 am
5 pm
2 7 am
5 pm
2 7 am
2 During
batch
9
9
6
16
4
21
2
2
7
12
For calcium and TSH, the ratio is 16 and 21, respectively. For CA 19.9, the ratio is 7,
which tends to indicate overcontrol of this clinically important test. For Chlamydia
trachomatis (CT) and Neisseria gonorrhoeae (NG) testing, the ratio between patient
samples and QC materials is 12. However, this ratio excludes internal controls that are
extracted with each sample in order to monitor the effectiveness of DNA extraction
and amplification.
To select QC frequency based on patient risk, Parvin [4] introduced a new
concept, called the expected increase in the number of unacceptable patient
results and noted E(Nu). E(Nu) is divided into two distinct components, the expected increase in the number of unacceptable results that are final (E(Nuf)) and the
expected increase in the number of unacceptable results that may be corrected
(E(Nuc)) [6]. E(Nuf) are unreliable results that are produced before an accepted QC
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
28 2 The elements of a conventional quality control protocol
and represent tests that are impacted by an analytical error that was small enough
to be undetected by the following QC specimen. In contrary, E(Nuc) are unreliable
results that are produced after the last accepted QC and before a QC rule violation.
In that case, E(Nuc) represent results that were impacted by an error that was large
enough to be detected by the subsequent QC [6]. Because QC was able to detect
the error, E(Nuc) represents patient results that may be retested and that will be
the subject of corrected laboratory reports. E(Nuf) and E(Nuc) will depend on the
total error allowable (TEa), the size of the systematic error (∆SEcrit), the performance of the method (Sigma metric) and the number of patient specimens being
tested between QC measurement (noted NB). E(Nuf) and E(Nuc) may be plotted
using a commercial software that was launched in 2015 (Mission Control ™, BioRad, Hercules, CA, USA). To define E(Nuf) and E(Nuc), necessary inputs will be the
following: average Sigma metric, QC rule in use, number of tests per day (ND), and
number of tests between QC measurements (NB). Table 2.15 summarizes the necessary QC parameters to input and resulting frequency metrics E(Nuc) and E(Nuf) for
three tests routinely performed in our laboratory.
Tab.2.15: QC metrics related figures 2.6, 2.7, and 2.8.
Test
TEa
Potassium
TSH
CA 19.9
BV des
BV des
BV opt
TEa (%)
Average
Sigma
5.8
23.7
19.5
8.2
10.5
11.9
QC rules
14s
14s
14s
ND
NB
Max E(Nuf)
Max E(Nuc)
81
175
14
27
87
7
5.1
7.5
0
13.5
42.5
3.5
Figures 2.6, 2.7, and 2.8 represents the visual plotting of E(Nuf) and E(Nuc)
for potassium, TSH, and CA 19.9, respectively. While the plotting of E(Nuc) shows
a curve of comparable shape for the three tests, the evolution of E(Nuf) is testspecific.
For potassium (fig 2.6), the maximum E(Nuf) is reached for a small systematic
error, whereas for TSH (fig 2.7), the maximum E(Nuf) is obtained with an SE of + or
−13.9%. This represents an SE that is large enough to cause a significant error, but is
also small enough to go undetected by the QC procedure.
For CA 19.9, we should probably consider performing fewer QC measurements
in our laboratory, but the real problem is the low number of patient samples
that we run daily for this tumor marker. We cannot aim to test patient samples
without running any QC measurements. Another option would be to perform CA
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.4 Quality control frequency 29
Fig. 2.6: Graphical representation of E(Nuf) and E(Nuc) for potassium at St. Luc Bouge clinical
laboratory.
Fig. 2.7: Graphical representation of E(Nuf) and E(Nuc) for TSH at St. Luc Bouge clinical laboratory.
19.9 measurements in batches; however, this is difficult to set up with analyzers
coupled with pre-analytical automation, like Cobas 8000 coupled to a Modular
Pre-Analytics (MPA) (Roche Diagnostics International AG, Rotkreuz, Switzerland).
Moreover, the turnaround time (TAT) for tumor markers are rather short, as clinicians are expecting results to monitor disease progression and possibly modify
patient treatment. The CA 19.9 case is an illustration that, sometimes, QC frequency,
and associated costs, cannot be reduced without a negative impact on patient care
and clinician satisfaction.
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
30 2 The elements of a conventional quality control protocol
Fig. 2.8: Graphical representation of E(Nuf) and E(Nuc) for CA 19.9 at St. Luc Bouge clinical
laboratory. For CA 19.9, E(Nuc) is 3.5 and E(Nuf) is 0 for all possible systematic errors; such a low
E(Nuf) is obtained due to a combination of an excellent average Sigma metric, a QC rule with an
elevated Ped and a very low number of patient specimen between QC samples.
QC frequency may be chosen following the definition of a maximal number of
E(Nuf) and E(Nuc). Understandably, the maximal E(Nuf) should ideally be <1, as
no clinical laboratorian wants to generate unacceptable patient results. However,
imputing a maximum E(Nuf) of 1 will sometimes generate a QC protocol that is not
achievable in routine. Therefore, an alternative tolerable E(Nuf) may have to be
selected. The value of maximal E(Nuc) is less intuitive. When unacceptable correctable results are generated following QC failure, these samples must be retested.
This process is costly in terms of reagents and consumables and is especially time
consuming for laboratories that do not use a post-analytical refrigerator connected
to a track. The greater the number of E(Nuc), the more time and money will be spent
for picking up and retesting the samples. Therefore, we would suggest to target a
maximal E(Nuc) of 50. A value of 20 would be even better, but may be difficult to
achieve in practice.
An interesting development on the concept of the unacceptable number of
patient results has been published very recently in Clinical Chemistry [7]. Yago
and Alcover have studied the relationship between E(Nuf) and Ped for multiple QC
rules with different number of control measurements [7]. They have concluded that
there is a close relationship between Ped and maximum E(Nuf) for common QC procedures; therefore, the maximum E(Nuf) may be predicted from the value of Ped.
As Ped is dependent on test performance, Ped, and therefore, E(Nuf) can be estimated
from the Sigma level. This allowed Yago and Alcover to design nomograms plotting
E(Nuf) versus Sigma metrics for different QC rules and different numbers of control
measurements. In the mentioned article, the authors focused on analyzers operating
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.4 Quality control frequency 31
in continuous mode with 100 patient specimens being assayed between QC measurements, a situation that is frequently encountered in modern automated laboratories. The nomograms from Yago and Alcover are a ready-to-use tool to design QC
frequency based on risk indicators [12]. This recent work may lead toward simplification of QC procedures [12].
The introduction of E(Nuf) and E(Nuc) concepts by Parvin and recent developments by Yago and Alcover are significant steps toward a scientific, risk-based definition of QC frequency. The principal limitation of these concepts is that the automatic
evaluation of QC materials every 25, 50, or 125 patient specimen is not currently available on many analytical platforms. Moreover, every test will have a different optimal
QC frequency. Therefore, an alternative design of QC frequency based on time units
would be welcome.
2.4.1 References
[1] Cooper G, DeJonge N, Ehrmeyer S, Yundt-Pacheco J, Jansen R, Ricos C, et al. Collective opinion
paper on findings of the 2010 convocation of experts on laboratory quality. Clin Chem Lab Med.
2011;49(5):793–802.
[2] Jones G, Calleja J, Chesher D, Parvin C, Yundt-Pacheco J, Mackay M, et al. Collective opinion
paper on a 2013 AACB workshop of experts seeking harmonisation of approaches to setting a
laboratory quality control policy. Clin Biochem Rev. 2015;36(3):87–95.
[3] Kinns H, Pitkin S, Housley D, Freedman DB. Internal quality control: best practice. J Clin Pathol.
2013;66(12):1027–32.
[4] Parvin CA. Assessing the impact of the frequency of quality control testing on the quality
of reported patient results. Clin Chem. 2008;54(12):2049–54.
[5] Hatjimihail AT. Estimation of the optimal statistical quality control sampling time intervals
using a residual risk measure. PLoS One. 2009;4(6):e5770.
[6] Yundt-Pacheco J, Parvin CA. Validating the performance of QC procedures. Clin Lab Med.
2013;33(1):75–88.
[7] Yago M, Alcover S. Selecting statistical procedures for quality control planning based on
risk management. Clin Chem. 2016;62(7):959–65.
[8]Miller WG. Quality control. In: McPherson RA, Pincus MR, editors. Henry’s clinical
diagnosis and management by laboratory methods. 22nd ed. Philadelphia, PA: Elsevier/
Saunders. 2011.
[9] ISO. Medical laboratories – requirements for quality and competence. ISO document
15189. Geneva, Switzerland: International Organization for Standardization. 2012.
[10] Miller WG, Erek A, Cunningham TD, Oladipo O, Scott MG, Johnson RE. Commutability limitations
influence quality control results with different reagent lots. Clin Chem. 2011;57(1):76–83.
[11] CLSI. Laboratory quality control based on risk management; approved guideline. CLSI document
EP23-A. Wayne, PA: Clinical and Laboratory Standards Institute. 2011.
[12] Cembrowski GS, Cervinski MA. Demystifying reference sample quality control. Clin Chem.
2016;62(7):907–9.
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
32 2 The elements of a conventional quality control protocol
2.5 Investigation and corrective actions following
QC or EQA/PT failure
Too often, QC results that are violating QC rules go unnoticed and do not lead to appropriate corrective actions [1]. Lack of knowledge regarding QC troubleshooting is probably due to apparent complexity and limited actions that clinical laboratorians may
take on modern, complex, random access analyzers. There is a limited number of available references about QC troubleshooting. However, review articles, p
­ ublished 20 years
ago [1] more recently [2, 3], or book chapters [4] and updated reference ­documents [5]
do provide very useful information about QC troubleshooting and corrective actions.
Table 2.16 provides the different steps to prevent, investigate, and correct QC failure.
Because they share multiple similarities, investigation and c­ orrective actions following
EQA/PT failures are also going to be discussed in this chapter.
2.5.1 Investigation and corrective actions following QC failure
First, to prevent excessive QC failures and time-consuming QC troubleshooting, it
is of paramount importance to design QC rules that have a high probability of error
detection (Ped), with the lowest possible probability of false rejection (Pfr) [1, 4].
A combination of QC rules that gives the lowest possible Pfr guarantees that a QC rule
violation is a true QC failure and needs further investigation. Special attention has
also to be put on the selection of an appropriate and realistic analytical goal. If the
quality requirement is too tight, QC results will frequently violate the resulting QC
rules. In practice, it is also important to compute a QC target value that is based on a
good estimate of the mean [4]. This needs at least 20 QC measurements, ideally performed in 20 days, in order to include sufficient sources of variation. If a new QC lot
has to be implemented in emergency, this could be reduced to 20 QC measurements
in 4 days (5 measurements per day). If a lower number of QC measurements is used,
the estimate of the mean will not be realistic enough and QC results will frequently
deviate from this mean. When a new QC material is validated, the analytical system
should be controlled using another QC material (for example, a similar QC material
from the previous lot or manufacturer QC being assayed in complement to the new
lot of third-party QC). This process should guarantee that the analytical system is in
control and that the mean of the new QC material will be correct. Determining a good
estimate of the standard deviation (SD) is of equal importance. Because the measurement of 20 QC material will provide an unrealistic estimate of the SD, Miller [4]
proposes to apply the cumulated SD of the previous similar QC material to mean of the
new QC lot, as SD tends to remain similar between two QC materials of different lots.
Other preventive actions include the design of clear standard operating procedures
(SOPs) relating to QC failure management. It is also important to teach QC concepts to
medical laboratory technologists (MLTs).
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.5 Investigation and corrective actions following QC or EQA/PT failure 33
Tab. 2.16: How to deal with QC failure.
Phase
Actions to perform
Preventive
–– Select a QC strategy with the lowest possible Pfr
–– Write clear QC SOPs
–– Provide adequate QC training for MLTs
General
–– Identify which QC rules are violated, if it is a random or systematic error, and
investigation
how many analytes are impacted
–– Identify if QC results changed progressively (drift) or abruptly (shift)
–– If available, compare statistical QC results with patient-derived QC data
–– If available, compare QC means and SDs to its peer group using inter-laboratory
comparison software
–– Evaluate the impact of QC violation on patient results
Specific
Elements to check Possible cause of QC failure
investigation
Operator
–– Inversion of QC levels
–– Use of inappropriate QC materials
–– Non adherence to SOPs
Corrective
actions
QC material
––
––
––
––
––
––
Expired QC material
Expired QC vial
Evaporation
Insufficient volume
Inadequate storage conditions of QC material
Analyte deterioration
Reagent
––
––
––
––
––
––
Insufficient reagent
Expired reagent pack
Inadequate reagent reconstitution
Cross-contamination of reagents
Inadequate storage conditions of reagent pack
New reagent lot or new reagent shipment
Calibrator
––
––
––
––
Expired calibrator
Inadequate calibrator reconstitution
Inadequate storage conditions of calibrator
New calibrator lot or new calibrator shipment
Analyzer
–– Incorrect maintenance procedure
–– Clogged pipettors, obstructed rinsing needles
–– Hydropneumatic leakage
Environment
–– Inadequate temperature or humidity level
–– Variation in power supply
–– Water problems
Actions to perform
–– Correct the root cause
–– If necessary, reassign QC value
–– If unacceptable patient results were generated, retest and issue corrected reports
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
34 2 The elements of a conventional quality control protocol
In case of a QC result that violates a QC rule, there is a need to verify that QC failure is
not due to a human error. Common QC result violation due to human errors include
inversion of QC levels (a technologist placing the cup containing level 3 on the rack
position supposed to contain level 1) or use of inadequate QC material (a technologist
assaying QC lot 46460, while QC lot 46440 is still in use). Therefore, the first step in
QC troubleshooting is generally to test another QC material, ideally from a vial different from the one that generated the QC failure. When human error is ruled out, QC
troubleshooting should start.
Reasons for QC failure may concern the following elements: operator, quality
control material, reagent, calibrator, analyzer, environment, or a combination of
several of these elements. The identification of the violated rule may provide some
information on the type of error that is occurring and on the element responsible for
QC failure. Violation of 1ks rules such as 13s, 15s , and violation of R4s is associated
with random error, such as the ones encountered with pipetting errors and electrical interferences, whereas violation of rules like 22s, 41s, and 10x indicates systematic
error that may occur following calibration bias or reagent lot change [2]. If available,
comparing QC data failure with alternative patient-based strategies such as average
of normals (AoN) may provide additional interesting information. Confronting QC
mean and SD of the test with peer-group data using inter-laboratory comparison
software may also add some insight. Once the preliminary steps have been achieved,
the impact of QC failure on patient results must be assessed. Because of the frequent
non-commutability of QC materials with patient samples [6], evaluation of the impact
on patient results is best performed with the selection of a limited number of previously
assayed patient samples (e.g. 5), covering the measuring range [4] and to retest these
specimens after the occurrence of QC failure. If the difference between measurements
exceeds the predefined total error allowable (TEa), patient testing must be stopped
and QC failure investigation should start. In contrary, if there is no patient impact
(no patient results exceeding TEa specifications after QC failure), investigation should
start, but there is no need to stop patient testing. Such events will generally lead to a
reassignment of the target values of QC materials.
QC failures due to QC material are generally easy to recognize. These include
the use of an expired QC bottle, insufficient volume of QC material in cups or tubes,
inadequate storage conditions, and analyte deterioration. One example of this latest
category occurred in 2014 in our laboratory. HDL-cholesterol was controlled using a
multiparametric, independent control material. We did notice that HDL-cholesterol
results on one QC concentration were progressively shifting downwards. We were
thinking at first about a reagent problem, but the normal results following testing of
additional QC material from the manufacturer and a specific third-party lipid control
showed that QC results were in the acceptable range. We concluded to a deterioration
of HDL-cholesterol values with the multiparametric QC material. This problem was
encountered close to the expiration date of this lot of QC material. Following implementation of a new lot of the same QC material, the HDL- cholesterol problem disappeared. Validation of new QC lots, as well as verification of new reagent and calibrator
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.5 Investigation and corrective actions following QC or EQA/PT failure 35
lots should limit QC failures due to these elements. However, verification of lots and
shipments of crucial elements, as stated in ISO 15189:2012 [7] represents an additional
cost for a clinical laboratory. Due to the frequent occurrence of non-commutability
of QC materials with patient samples [6], verification of new reagent lots should be
performed using patient samples instead of QC samples. Due to internal checks in
modern analytical systems, QC failures due to expired reagents should not be encountered. In case of a QC result failure, the following information regarding reagent pack
should be obtained: reagent pack identification; number of remaining tests in the
reagent container; location and temperature of storage; if the reagent pack was the
subject of a recent manufacturer notice; date of last calibration; reagent lot number.
In our laboratory, different QC failures due to reagents were encountered. One of these
was about thyroglobulin testing. In 2012, a new reagent lot for this test was implemented in our laboratory. Two levels of independent QC material showed a upward shift
immediately after implementation of this new lot. To exclude a possible matrix effect,
we did test several patient samples that were previously assayed using the preceding
reagent lot. All the patient results shifted upwards in a similar manner as the independent controls. Therefore, patient testing was suspended in our laboratory and we
requested for another reagent lot for thyroglobulin. This resulted in a security notice
emitted by the manufacturer regarding this reagent lot, as falsely increased thyroglobulin results may generate unwanted clinical investigations; this is especially true for
results that show an increase in the low range while they were expected to be negative
(e.g. patients with thyroid cancer that were treated by total thyroidectomy). Another
complex QC problem that was encountered recently using a similar clinical chemistry
platform was due to cross-contamination between hs-CRP and HIV reagents [8]. The
problem only occurred if HIV and hs-CRP testing were performed on the same module
of immunoassays. This contamination led to a shift in the calibration signal, resulting
in low levels of manufacturer control material showing 13S rule violation. The manufacturer provided two possible corrective options: the first was to perform hs-CRP and
HIV on two different analyzers of the same class; the second possibility was to discard
affected reagent containers in case of documented contamination. Calibrators may
be responsible for QC violations and the causes of calibrator problems are similar to
those encountered with reagents.
While automated analyzers have become increasingly more complex, visual inspection may show analyzers problems such as clogged pipettors, obstructed rinsing
needles, and hydropneumatic leakages. Causes of QC failures due to equipment are
not rare, especially when an analytical platform is aging. Additional investigation of
QC failure related to the analytical platform includes the review of information contained in the instrument log, such as system alarms and error messages.
Generally, the ultimate corrective action that clinical laboratorians may take is
calibration using a properly reconstituted calibrator on a new reagent pack, performed on a correctly maintained instrument. If the problem persists, then it may be
caused by a complex problem on the analyzer and manufacturer customer service has
to be called. Additional root cause analysis may reveal environmental problems such
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
36 2 The elements of a conventional quality control protocol
as change in electrical power supply or decrease in water quality. Finally, a QC failure
may be due to multiple causes. Practical examples of QC failure and corresponding
corrective actions are given in chapter 7.
In the case of a significant impact on patient results, it is impossible to know the
exact time when the error occurred; it is therefore recommended to retest the last 10
patient samples since the QC failure was discovered. If unacceptable results (exceeding TEa specifications) are encountered with these 10 patient samples, it is necessary
to retest the immediately preceding 10 samples, issue corrected reports, and proceed
backwards until no significant difference in patient results is observed [3].
2.5.2 Investigation and corrective actions following EQA/PT failure
The investigational steps, potential root causes, corrective actions, and monitoring following EQA failure are well described in the reference document on PT [9] and are summarized in an easy to use table in the 2011 review article on EQA by Miller et al. [10]. The
first step is checking the sample that was used in the EQA program if there was no inversion of EQA materials and if the PT specimen was received in good condition. Thereafter, a general investigation verifies the different elements that possibly generated the
problematic EQA result, such as the reagents and controls used and the maintenance
log of the analyzer. SOPs that were used by technologists to test the EQA sample have
to be verified as well. It is also necessary to crosscheck the EQA results with internal
quality control (iQC) data to verify if no iQC point was violating a rule when PT was performed. Finally, laboratory specialists need to define if this EQA failure is a single event
or if difficulties regarding this PT program were already encountered previously. After
the investigational phase, the classification of the EQA failure may begin. CLSI GP27-A2
defines six categories that may explain EQA/PT failure [9]. These include problems with
EQA materials, clerical errors, methodological problems, equipment problems, technical problems related to personnel, and errors related to PT evaluation protocol. CLSI
describes also the possibility that no explanation may be found after extensive PT evaluation: this is encountered in up to 24% of PT failure cases. Table 2.17 describes the
different steps in EQA failure investigation and classification of error causes. Problems
affecting PT materials may be due to incorrect storage condition of PT samples, sample
deterioration due to bacterial contamination, or a ruptured container. Clerical errors are
frequently encountered regarding EQA failure and may occur in the pre- or post-analytical phase of PT testing. As the PT materials have to be tested like patient samples, their
information has to be entered into the LIS, and this step may be subject of mistakes
such as mislabeling of the EQA samples. Consequently, results of a PT sample may be
sent in place of the other and will generate a double EQA failure. As PT results have to
be transcribed or typed on the results page of the EQA provider website, transcription
error are quite frequent. Finally, regarding clerical errors, an error in measurement units
or in the denomination of the method used will lead to an aberrant result (outlier) or
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.5 Investigation and corrective actions following QC or EQA/PT failure 37
Tab. 2.17: How to deal with EQA/PT failure.
Step
Actions to perform
Investigation ––
of PT failure ––
––
––
––
––
––
PT specimen received in good conditions?
No inversion of QC material?
Verify internal controls, reagents and calibrators
Check maintenance log of the analyzer
Correct SOP used?
SOP followed by operator?
Perform crosscheck of iQC results and inter-laboratory comparison with PT
results
–– Single or multiple analyte failure in this EQA program?
–– Recurring PT failure for same analyte?
Problem
Category
classification
Problems with
EQA materials
Laboratory phase Possible cause of EQA/PT failure
Pre-analytical
–– Incorrect storage conditions of PT sample
–– PT sample deterioration
–– Damaged PT sample container
Clerical problems Pre- or postanalytical
–– Mislabeling of PT samples
–– Transcription errors
Methodological
problems
––
––
––
––
––
Analytical
Equipment
problems
Incorrect SOP used
Incorrect use of correct SOP
Calibration problems
Biased methods
Methods lacking sensitivity and
specificity
–– Analyzers used without validation and
verification
–– Manufacturing problems for reagents or
calibrators
––
––
––
––
Inadequate maintenance
Obstructed needle or probe
Damaged tubing
Defective pipettors
Inadequate reconstitution of PT material
Excessive delay in managing PT samples
Pipetting errors
Dilution errors
Misinterpretation of test reaction
Technical
problems
Pre-, analytical
and postanalytical
––
––
––
––
––
Problems with PT
evaluation
Post-analytical
Error in peer group assignment
Non-commutability across reagent lots
No identified
cause
Some PT/EQA failure remained unexplained after investigation
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
38 2 The elements of a conventional quality control protocol
in a misclassification in the peer group, both leading to unsatisfactory performance
evaluation.
Methodological problems may be related to the analyzer in case of a test performed
on an automated instrument or to the SOP in case of a manual method. Concerning
automated analyzers, methodological problems may include calibration problems,
measurement methods that are biased or lacking sensitivity or specificity, analyzers
used without validation or verification, and analyzers using reagents or calibrators
that were impacted by manufacturing problems. For manual methods, methodological causes of EQA failures include the absence or incorrect use of an SOP. Equipment
problems are due to a specific malfunctioning of an analyzer’s component, such as
obstruction of a needle or probe, damaged tubing, defective automatic pipettors, and
inadequate maintenance. Methodological and equipment problems concern the analytical phase of the testing process. Technical problems are due to errors by laboratory personnel, such as inadequate reconstitution of EQA material, excessive delay
in managing PT samples, pipetting and dilution errors, and misinterpretation of test
reaction. Technical problems may occur in the three different phases of the testing
process. One last cause that may explain QC failure is a problem regarding the EQA/
PT evaluation protocol, with one frequent cause being the assignment of a participating laboratory to the wrong peer group. As recently shown, non-commutability of a
PT material across different reagent lots may also lead to errors in the performance
evaluation of a laboratory [11]. Finally, some PT failure may not be explained, even
after extensive investigation [9]. If the cause of a PT failure is identified, corrective
action will generally be easy to perform. All the steps in the investigation of a PT
failure, including cause identification and following corrective actions, have to be
registered and documented in accordance with guidelines [7, 9].
2.5.3 References
[1] Petersen PH, Ricos C, Stockl D, Libeer JC, Baadenhuijsen H, Fraser C, et al. Proposed guidelines
for the internal quality control of analytical results in the medical laboratory. Eur J Clin Chem Clin
Biochem. 1996;34(12):983–99.
[2] Kinns H, Pitkin S, Housley D, Freedman DB. Internal quality control: best practice. J Clin Pathol.
2013;66(12):1027–32.
[3] Jones G, Calleja J, Chesher D, Parvin C, Yundt-Pacheco J, Mackay M, et al. Collective opinion
paper on a 2013 AACB workshop of experts seeking harmonisation of approaches to setting a
laboratory quality control policy. Clin Biochem Rev. 2015;36(3):87–95.
[4] Miller WG. Quality control. In: McPherson RA, Pincus MR, editors. Henry’s clinical diagnosis and
management by laboratory methods. 22nd ed. Philadelphia, PA: Elsevier/Saunders; 2011.
[5] CLSI. Statistical quality control for quantitative measurement procedures: principles and
definitions. 4th ed. CLSI guideline C24. Wayne, PA: Clinical and Laboratory Standards Institute.
2016.
[6] Miller WG, Erek A, Cunningham TD, Oladipo O, Scott MG, Johnson RE. Commutability limitations
influence quality control results with different reagent lots. Clin Chem. 2011;57(1):76–83.
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
2.5 Investigation and corrective actions following QC or EQA/PT failure 39
[7] ISO. Medical laboratories – requirements for quality and competence. ISO document 15189.
Geneva, Switzerland: International Organization for Standardization. 2012.
[8]Wils J, Boudewijns M, Vandermarliere M, Callewaert N. Monitoring the patient response
as an alternative to commercial negative quality control in infectious serology. J Clin Virol.
2015;72:30–5.
[9] CLSI. Using Proficiency Testing to Improve the Clinical Laboratory; Approved Guideline- Second
Edition. CLSI document GP27-A2. Wayne, PA: Clinical and Laboratory Standards Institute. 2007.
[10] Miller WG, Jones GR, Horowitz GL, Weykamp C. Proficiency testing/external quality assessment:
current challenges and future directions. Clin Chem. 2011;57(12):1670–80.
[11] Stavelin A, Riksheim BO, Christensen NG, Sandberg S. The Importance of Reagent Lot
Registration in External Quality Assurance/Proficiency Testing Schemes. Clin Chem.
2016;62(5):708–15.
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
Brought to you by | SUNY Binghamton
Authenticated
Download Date | 10/29/17 2:46 AM
Документ
Категория
Без категории
Просмотров
13
Размер файла
465 Кб
Теги
002, 9783110346268
1/--страниц
Пожаловаться на содержимое документа