Professional Documents
Culture Documents
Allowable Imprecision for Laboratory Tests Based on Clinical and Analytical Test
Outcome Criteria
James 0. Westgard’, Julie J. Seehafer, and Patricia L. Barry
The allowable imprecision for laboratory tests has been be used in judging the acceptability of laboratory per-
estimated from criteria based on clinical and analytical formance for -80 tests that are regulated in the US (1 )2
test outcome. The analytical outcome criteria studied are Given that the origins of the CLLA PT criteria are not
the Clinical Laboratory Improvement Amendments (CLIA) well-documented, we consider it of interest to under-
criteria for proficiencytesting. The clinical outcome criteria stand how they compare with clinical outcome criteria.
ing than the clinical outcome criteria for aspartate ami- ing more tests, changing therapy, or considering an-
notransferase (asymptomatic patients), cholesterol, cre- other diagnosis.”
atinine (asymptomatic patients), glucose, thyroxine, total The usefulness of clinical outcome criteria based on
protein, urea nitrogen, hematocrit, and prothrombin time. physicians’ opinions vs the quality goals for imprecision
The clinical outcome criteria are more demanding for derived from biological variation has been discussed (3),
bilirubin (acute illness), iron, potassium, urea nitrogen as has the use of analytical outcome criteria for PT vs
(acute illness), and leukocyte count. The estimates of quality goals for imprecision (4). Unfortunately, direct
allowable imprecisionfrom analytical and clinicaloutcome comparisons are not possible between clinical outcome
criteria overlap for aspartate aminotransferase (acute ill- criteria, analytical outcome criteria, and goals for im-
ness), bilirubin (asymptomatic patients), calcium, creati- precision and inaccuracy because each set of criteria
nine (acute illness), sodium, triglyceride, and hemoglobin. describes limits for different factors that affect the vari-
ation in a test result. Clinical outcome criteria describe
IndexIngTerms: laborafolymanagement/qualitycontrol/proficiency a limit for both preanalytical and analytical components
testing of variation in the testing process. Analytical outcome
criteria describe a limit for only components of the an-
Test outcome criteria provide consumer-oriented re- alytical process, e.g., imprecision, inaccuracy, and qual-
quirements for quality, in contrast to performance goals ity control (QC). Goals for imprecision and inaccuracy
for imprecision and inaccuracy, which provide labora- are operating specifications for individual components.
tory specifications for operating a testing process. Out- Comparisons can be made when clinical and analyti-
come criteria define a limit for the total variation that is cal outcome criteria are used to derive operating speci-
allowable in a laboratory test result, rather than setting fications for the imprecision and inaccuracy that are
a limit for an individual factor or component that con- allowable and for the QC that is necessary in routine
tributes to the variation of a test result. Such outcome operation of an analytical testing process (5, 6). Previ-
criteria can be formulated to reflect either analytical or ously, we illustrated how proposed European specifica-
clinical requirements for test performance. tions for imprecision and inaccuracy could be compared
Analytical outcome criteria 4escribe the total analyt- with US CLLA proficiency testing criteria (7). Here, we
ical errors that would cause a test result to be judged as determine the maximum imprecision that is allowable
analytically unacceptable. For example, the 1988 Clin- based on the clinical outcome criteria of Skendzel et al.
ical Laboratory Improvement Amendments (CLIA) de- (2) and compare these estimates with the maximum
fine the total error criteria for proficiency testing (PT) to
Department of Pathology and Laboratory Medicine, Medical 2Nonstandard abbreviations: PT, proficiency testing CLIA,
School, University of Wisconsin, 600 Highland Ave., Madison, WI Clinical Laboratory Improvement Amendments (1988); QC, qual-
53792. ity control; and NCEP, National Cholesterol Education Program.
‘Author for correspondence. Fax 608-263-0910. Received April 4, 1994; accepted May 24, 1994.
N
1/2,./R441
0.03 4 1
-bIe 1255
Table 1 summarizes the allowable imprecision for for bilirubin (acute illness), iron, potassium, urea nitro-
those tests for which clinical outcome criteria are de- gen (acute illness), and leukocyte count. The estimates
fined by Skendzel et a!. (2) and analytical outcome cri- of allowable imprecision overlap for aspartate ami-
teria are defined by CLIA PT criteria (1). Column 4 notransferase (acute illness), bilirubin (asymptomatic
shows the clinical outcome criteria in the form of the patients), calcium, creatinine (acute illness), sodium,
medically significant changes from Table 1 by Skendzel triglyceride, and hemoglobin. All the operating specifi-
et al. (2). Column 5 shows the estimates of within- cations for allowable imprecision are more demanding
subject biological variation that were used to derive the than the medically useful CVs recommended by Skend-
clinical maximum allowable imprecision shown in col- zel et al. (2).
umn 6. These estimates of imprecision are to be com-
pared with the values in column 7 for the medically Discussion
useful CVs [taken from Table 3 of Skendzel et a!. (2)] The difficulty in comparing different requirements for
and with the values for the analytical maximum allow- analytical performance is ifiustrated in Table 1 by the
able imprecision (column 8), derived from the CLIA iron test, where the medically significant change or clin-
analytical outcome criteria shown in column 9. ical decision interval of 33% is almost entirely con-
These estimates of the maximum allowable impreci- sumed by the within-subject biological variation of
sion show that the CLIA analytical outcome criteria are 19.8%, requiring analytical imprecision of <1% when
more demanding than the clinical outcome criteria for biological variation is taken into account, rather than
aspartate aminotransferase (asymptomatic patients), the medically allowable CV of 17.2% recommended by
cholesterol, creatinine (asymptomatic patients), glucose, Skendzel et a!. (2). Similar situations are seen for bill-
thyroxine, total protein, urea nitrogen, hemathcrit, and rubin (acute illness) and urea nitrogen (acute illness),
prothrombin time. The clinical outcome criteria are where the within-subject biological variation by itself
more demanding than the analytical outcome criteria will cause changes that exceed the intervals for medi-
cally significant change. These examples illustrate why sion, whereas the earlier estimates assumed all of the
there has been considerable skepticism about the valid- variation could be analytical imprecision. Linnet recog-
ity of these earlier estimates of medically useful CVs nized these difficulties and accounted for both preana-
(14) and may also help explain why there has been little lytical and biological variation when deriving the max-
ongoing work to determine clinical outcome criteria in imum allowable errors to be used in selecting QC
the form of medically significant changes in test results. procedures (15).
Our purpose here is to consider the difficulties in trans- The rigorous demands for analytical imprecision re-
lating medically significant changes into specifications sult from the design objective to achieve 90% detection
for operating laboratory testing processes, rather than of critical-sized systematic errors by using only the low
the difficulties in estimating medically significant numbers of control measurements (N = 2-4) that are
changes. However, we think that there is merit in de- deemed acceptable by professional practice and regula.
termining how customers actually use laboratory tests, tory guidelines. The 90% objective is based on the desire
that guidelines are needed for obtaining reliable esti- to detect medically important analytical errors in the
mates of medically significant changes, and that knowl- first run in which they occur (expected average run
edge of customer requirements should be valuable when length ARL = 1/Pedor 1/0.90 = 1.1). Use oflowNvalues
establishing outcome criteria and operating specifica- limits the power of statistical QC procedures and makes
tions. it difficult to detect critical systematic errors that are
The operating specifications for imprecision that we <-3 CV of the method. Improving precision (reducing
derived in this study are consistently more demanding the CV) has the effect of changing those critical system-
than earlier estimates of medically useful CVs because atic errors to higher multiples of the CV, enabling
the additional factors of within-subject biological varia- greater error detection with commonly used QC proce-
tion and the sensitivity of QC procedures are included in dures.
the quality-planning models that were applied. These These rigorous demands for analytical performance
quality-planning models provide error budgets for the are achievable in routine operation, as demonstrated in
various factors that may affect the variation of a test an earlier study (16), where 90% error detection was
result. Factors such as within-subject biological varia- obtained for 15 out of 18 tests on a multitest chemistry
tion and the sensitivity of the QC procedures reduce the analyzer with use of N = 2 and single-rule QC proce-
amount of variation that is left for analytical impreci- dures. Lower probabffities for error detection may be