More information about text formats
In the paper by Schmidt et al, measurements of the left and right
extremities of the same subjects seem to have been pooled to calculate
standard reference values for musculoskeletal ultrasonography .
However, it is common knowledge that right and left measurements are tied
and should not be mixed in order to get a larger “study sample”.
In the same paper, correlations have been...
In the same paper, correlations have been used as measures of
agreement between two observers (inter-observer reliability). Correlations
are inappropriate for that purpose, however. Correlations express the
strength of association between the paired values, not the agreement
between them. Systemic variation between observers is a common phenomenon.
Moreover, the correlation is influenced by the distribution of data. A
large between-subject variation will lead to a high correlation. It is
clearly not reasonable to assess agreement by a statistical method that is
highly sensitive to the choice of the sample of subjects. Altogether, the
correlation may be high, when the agreement is poor. Calculating the
standard deviation of the differences between the pairs of measurements
would be more meaningful [2,3].
In the Schmidt et al paper, correlations were also used as measures
of repeatability (intra-observer reliability). The problems with
correlations are the same in this case. Furthermore, the correlation is
not very useful when the clinician wants to know whether a given change is
due to expected within-subject variation or to real change. Other measures
of repeatability have a more direct interpretation which can be applied to
individual measurements, for example limits of agreement calculated on the
basis of within-subject standard deviations [4,5].
(1). Schmidt WA, Schmidt H, Schicke B, Gromnica-Ihle E. Standard
reference values for musculoskeletal ultrasonography. Ann Rheum Dis 2004;
(2). Altman DG. Practical statistics for medical research. Chapman
& Hall, London, 1991.
(3). Bland JM, Altman DG. Statistics notes: measurement error and
correlation coefficients. BMJ 1996; 313: 41-2.
(4). Bland JM, Altman DG. Statistics notes: measurement error. BMJ
1996; 313: 714.
(5). Magid E, Petersen PH, Christensen M. A note on the theory of
reference changes. Scand J Clin Lab Invest 1992; suppl. 208: 95-101.