I'd like to share the contents of a memo from our principal scientist to the
director of pathology about what can be regarded as a significant change in
glycated Hb results. I do so of course with his permission - I think it's a
clear statement of analytical goals.
Gary,
The basic statistical rule for interpreting serial trends is:
There is a 95% probability that there is a real change if the difference between
the two results is greater than 2.7 times the long term assay standard
deviation. This SD will vary with concentration - so you need to know the
laboratory's performance for its internal QC closest to the decision level.
It is safer to use a factor of 3 to allow for occasional short-term excursions
in assay performance.
Therefore, choosing a method with better precision improves your ability to
detect trends. For HbA1c, experts in the field recommend a goal of 2%
coefficient of variation (SD/concentration). Few methods can achieve this - all
of them automated HPLC assays like our Primus method.
For example, we achieved an overall CV of 2.1 % in the February - June AQAP.
This is similar to our current internal QC performance at 6% HbA1c.
So a statistically significant change for our assay near 7% HbA1c is 3 x 2.1 x
7/100 = 0.44% HbA1c; near 9% is 3 x 2.1 x 9/100 = 0.57% HbA1c.
Other commonly used methods are not as good. The equivalent AQAP CVs for Biorad
Diastat is 4.6% and Roche/Hitachi assays range from 3.1 to 6.8 %. However, it
is the individual laboratory precision over the comparison period which should
be sought.
If the assay has a CV of 5%, then for the examples of repeat assays of 8.7, 9.6
and 6.7, 7.6, the difference of 0.9% is well within the normal imprecision of
the test and a trend cannot be assumed.
The other advantages of our method are : no errors due to specimen storage
delays, haemoglobin variants, or carbamylated haemoglobin in renal failure. I
think it is fair to say that all other HbA1c methods in use are subject to at
least one of these interferences; the competitive high-precision HPLC assays
are subject to all of them.
Rob
|