46
| 4 | 2011
Klinisk Biokemi i Norden
tain deviation from the regression line or equal line
is often considered. This is reported as a claim that a
certain percentage of all observations, e.g. 95 %, shall
be within for instance
±
10 %
of the regression function
or equal function (fig 4). These limits are called LER,
Limit of Erroneous Results and the interval between
the limits ATE or Allowable Total Error. Acceptable
performance is individual for different quantities and
at least theoretically also for different purposes of the
measurement.
Other characteristics that are useful in evaluating
the performance of a new procedure are estimation of
the diagnostic sensitivity and specificity. This approach
estimates to what extent a test method provides the
same number of correct and false diagnosis as the
previous, comparative procedure, desirable or not! The
outcome will depend on the cut-off e.g. the suggested
reference limits. The approach may also be applicable
to ordinal quantities, e.g. obtained by urinary sticks.
Three different situations may be recognized: the
disease is characterized by an increased concentration,
e.g. S-T4, or decreased, e.g. S-HDL cholesterol or both,
e.g. B-Glucose or S-TSH. The two first are trivial, the
third may be more controversial. Here I suggest that
all results above the high cut-off when estimation the
performance of the low cut-off and
vice versa
.
Thus,
the results within the reference interval (TN) will be
used twice (table in figure 4) whereas the FP, FN and
TP of the other cut-off are disregarded. Two sets of
result will thus be reported. With a suitable software
the cut-offs can be optimized for the intended use of
the measurand.
The allowable difference is related to the Reference
Change Value (RCV) and thus of importance for the
utility of the quantity for clinical purposes. The RCV
is usually determined by the observed uncertainty of
the studied procedures.
ROC-curve
The clinical performance of a procedure is described
by a ROC-curve (Receiver Operating Curve). This is
obtained by plotting the calculated sensitivity on the
Y-axis and (1-specificity) on the X-axis for each tested
cut-off. An efficient measurand will produce a hyper-
bolic curve with a maximal ratio between sensitivity
and (1-specificity) (LR, Likelihood Ratio) close to the
upper left corner of the diagram. LR on the diagonal
(
slope (b) =1, intercept (a) =0) or below have no diag-
nostic value. The LR takes any values between 0 and
1,
the Youden index describes the maximal distance
between the diagonal and the ROC curve, thus a high
value is preferable. The K-index describes the distance
between the upper left corner and the ROC curve and
a small value is thus preferable.
References
1.
NCCLS/CLSI. Evaluation of Precision Perfor-
mance of Quantitative Measurement Methods;
Approved Guideline - Second Edition.EP5-A2
(
ISBN 1-56238-000-0). Wayne, Pennsylvania
USA, 2004.
2.
CLSI. Method Comparison and Bias Estimation
Using Patient Samples; Approved Guideline
–
Second Edition. EP9-A2 IR. ISBN 1-56238-
472-4.
CLSI, Wayne, Pennsylvania USA, 2010
3.
CLSI. User Verification of Performance for
Precision and Trueness; Approved Guideline –
Second Edition. EP15-A2. ISBN 1-56238-574-7.
Wayne, Pennsylvania USA, 2005.
4.
JCLM 100-2008 Evaluation of measurement
data — Guide to the expression of uncertainty
in measurement (
)
5.
EURACHEM / CITAC Guide CG 4 Quanti-
fying Uncertainty in Analytical Measurement.
QUAM:2000.1
6.
CLSI Uncertainty in measurements. C51. To
be published
7.
Bland JM, Altman DG. Statistical methods for
assessing agreement between two methods of
clinical measurement. Lancet 1986; i: 307-10.
8.
Carstensen, Bendix. Comparing clinical measu-
rement methods, A practical Guide John Wiley,
Chicester 2010, ISBN 978-0-470.69423-7,
9.
Monti KL Folded Empirical Distribution Fun-
ction Curves – Mountain plots. Amer Statisti-
cian 1995;49(4):342-345.
(
Fortsat fra side 45)
Suitable software for EXCEL is available from the author. All formulas are available in Anders Kallner. Commonly
used terminology and formulas in laboratory medicine. (ISBN 978-91-633-3272-2)
