* Many people are not yet subscribed to the statslausanne mailing-list;
please forward appropriately to your colleagues *
Our next statistics forum for people interested in quantitative methods
in biology and medicine will take place this Thursday 22 March at 11h00
in the reading room (2020) of the Genopode building (2nd floor, same
level as the entrance from the Amphimax building).
The speaker will be Valentin Rousson from IUMSP/CHUV, who will ask
« What if the conditions of application of Bland and Altman’s limits of
agreement are not satisfied?»
And the abstract follows below.
Hope to see you there !
Frédéric
---
The limits of agreement of Bland and Altman have been introduced as an
alternative to the concept of intraclass correlation to measure the
agreement between two methods of measurement. Whereas intraclass
correlation is a measure of reliability “in a relative sense”, informing
on a percentage of variability in the measurements which is due to the
measurement errors, depending hence crucially on the range of
observation considered, the limits of agreement, which are based on a
simple plot of the differences between the measurements from the two
methods versus their averages, provide information on measurement errors
expressed in the units of the data, which may facilitate the task of
clinicians to judge whether or not they are acceptable in practice.
Conditions of application of this method include in particular that the
size of measurement error is not related to the size of the “true
measurement”. It is however not clear how to interpret such limits of
agreement when these assumptions are not met, in particular when there
is a trend in the systematic difference between the two methods of
measurement. The statistical literature does not help us much in this
regard. Worse, some authors have actually criticized the limits of
agreement of Bland and Altman, pointing out situations where such a
trend could be an artifact of the method. When one of the two methods is
a gold standard, they propose to plot and analyze the differences of the
two measurements versus the gold standard (instead of the averages of
the two measurements), a method that Bland and Altman had themselves
criticized. Using a real example from a statistical consultation, where
the goal was to compare the temperatures provided by two kinds of
thermometers, we shall try to understand this controversy.”
Literature :
Bland JM and Altman DG (1986). Statistical methods for assessing
agreement between two methods of clinical measurement. Lancet i, 307-310.
Bland JM and Altman DG (1995). Comparing methods of measurement: why
plotting differences against standard method is misleading. Lancet 346,
1085-1987.
Hopkins WG (2004). Bias in Bland-Altman but not regression validity
analyses. Sportscience 8, 42--46.
Krouwer JS (2008). Why Bland-Altman plots should use X, not (Y+X)/2 when
X is a reference method. Statistics in Medicine 27, 778-780.
