![]()
21
| 3 | 2004
Klinisk Biokemi i Norden
Abstract
Measurements on ordinal scale
– often called semi-quantitative
tests – of the type dipsticks, are
used for a number of quantities
in urine. The results are reported
as – or + (++, +++ etc.) but
should be reported as 0 or 1 (2,
3, etc.). These measurements on
ordinal scale are more variable than assumed, and
measurements on a single sample may give 15 times of
0 and 85 times of 1, due to the nature of the process.
Consequently, it is not possible to use a control mate-
rial with such a concentration in daily routine. These
distributions, may however be used in external control
in order to characterise and control different kits.
In order to obtain the maximum information from
such projects, it is necessary to know the concen-
trations of control materials and to perform a large
number of independent measurements – preferable
more than 100 measurements of each sample.
A model for evaluation of results from series of
control samples is based on a rankit-transformation
of the percentage (fractions) of 1-results for each
control, using ln-transformation of the abscissa.
The process is comparable to the transformation of
a Gaussian distribution or histogram by cumula-
tion and rankit-transformation. If the points fit a
straight line in the rankit-plot, it can be described
as a cumulated Gaussian distribution.
The model is applied to data from external con-
trol of Urine-hCG and to Mononucleosis, and for
both components several kits apply to the model,
and it is discussed whether deviations are due to
poor performance or defects of reagents – or if the
model only is useful for some kits. The model is
also tested for a semi-quantitative test (CRP-Nyco
card visual reading) where each reading of > 10,
> 25, > 50 mg/L is treated as a 0-1 test. Also here
the resulting percentage points in a rankit-plot fit
to straight lines.
The usefulness of the model in external quality
assessment is discussed.
Nordic EQA-organizers/external control orga-
nizations have decided a project, where the urine
components hCG, haemoglobin, glucose, albumin,
esterase and nitrites are investigated. The project is
running until spring 2005, at which time it will be
evaluated and validated for possible continuation as
a routine procedure. Before these results are avail-
able, some of the project participants want to pres-
ent the model and some pre-project results.
Keywords
External control; Ordinal scale; Rankit model;
Qualitative measurements; Semi-quantitative meas-
urements
Introduction
Qualitative and semi-quantitative measurement
results have always been considered simple analy-
ses of the type of dipsticks to get a quick and easy
estimate of the possible presence of components like
glucose, bilirubins, blood/haemoglobin etc. in urine.
The results are usually reported as – or + (and some-
times also ++, +++, and ++++), or more correct as
0 and 1 (2, 3, and 4), and the general interpretation
is usually as simple as the result: ‘No presence’ or
‘presence’ of the component.
When this type of measurement is simply a quick
or cheap rough estimate of the concentration of a
quantity, which could be measured on a ratio scale,
this is a serious oversimplification of the problem,
as the outcome of such a 0/1-test is not reproduc-
ible within considerable concentration intervals,
within which some results can be 0 and other 1 for
Basic model and preliminary results on a Nordfond project
on measurements reported on an ordinal scale
Per Hyltoft Petersen
1
,
2
, Sverre Sandberg
2
, Nina Gade Christensen
2
, Esther Jensen
1
,
Ole Blaabjerg
1
-
1.
Department of Clinical Biochemistry, Odense University Hospital,
Odense, Denmark
2.
NOKLUS, Norwegian quality improvement of primary care labora-
tories, Division for General Practice, University of Bergen, Norway.
E-post: per.hyltoft.petersen@ouh.fyns-amt.dk
