I think I will make this the last of my inputs to the mail base on this
subject, lest I be mistaken to be pondering along on “a subject abandoned by
most analytical chemists, clinical chemists and accreditation bodies”. The
fact remains that there is a considerable drive to adopt the Total Error
concept in the UK, as outlined in the presentations by WEQAS (a major UK EQA
provider) at their user-group meeting I attended in December 2009. Their
requirements will be mandated to their customers (NHS labs) through the CPA
scheme, in, I am lead to believe, a fashion not dissimilar to the CLIA in
the USA; through the assignment of Total Error goals to which adherence must
be demonstrated monthly/periodically.
One can discuss the relevance of the approach, and disagree with the way in
which statistics are derived; however, one can not decide to opt out. As
such, Total Error not only warrants discussion, but should be central to a
UK laboratory's attention. I would be concerned if the situation were
otherwise. Further, it is taken for granted that since the Total Error
concept is not new, that it is well understood and has been rejected for
better models. I would argue that the whole concept of error determination
on this level is very poorly appreciated, not through a lack of skills or
knowledge (far from it), but more so due to a lack of accountability in the
UK; we have never had to demonstrate quantified error in this way.
Also, I would like to reassure Dr Ekins that I have thoroughly considered
the varying nature of precision across a concentration range, and have
attached two screenshots of how I do so. Typically, the worst-case measured
imprecision is taken as a point for calculation, unless one is willing to
provide proportioned Total Errors for the method, or specific-concentration
Total Errors perhaps for specialist clinics where certain critical decision
limits are defined. I generally tend not to discuss this aspect of analysis
initially, as it can be confounding and distracting when discussing the
basics. I certainly recognise the importance of including such
considerations, and thank him for bringing the point to light.
I would also like to thank Mr Parry for his clarifying letter on the
distinction between TE, TEa, and Uncertainty. I could indeed be accused of
using the terms interchangeably, and I accept the points made. However, I
would like to give a brief account of why I justify doing this.
A calculation of Total Error that does not include an account of the
potential for error assumes that there is NO error above and beyond stable
variance/bias. On a day to day basis, this is of course un-knowable; no
routine laboratory (that I know of) performs sensitive method verification
daily for all their analytes. What actually happens is that a handful of QC
materials are analysed, and, if the results of which fall between set
limits, we assume the analyser to be fault-free. A substantial degree of
error can be assigned to this practice, and must be described in our final
calculations.
One can, perhaps, retrospectively examine QC data and make a fair assumption
that a method was error free; however, ‘retrospective’ is no good for our
patients. We need a method that gives a somewhat reliable account of error
quickly and efficaciously. Calculating Total Error without SDcrit does not
do this, and I can imagine that a laboratory that provides Total Error
estimations in this way would be quickly found to be breeching their own
goals when tested. As such, I do not consider any practical use for the
other approaches, and favour TEa (as described by Westgard here:
http://www.westgard.com/critical-error-graphs.htm ) as a description of
measured uncertainty.
I do not want to deliberate on the matter any further; I feel that I have
made my point here and in previous correspondence. Also, I do not believe
the purpose of Mr Parry’s letter was to evaluate the methods, rather to
clarify the different equations. I will perhaps in future consider my use of
the terms more carefully.
I agree completely with the sentiment in Mr Challand’s letter, and would
give a similar response as to Mr Kallner. Whilst we may have our own ideas
on what is good for the patient and what may actually constitute
improvements, we are still bound to the requirements set to us by our
governing bodies, and conformance is likely to be assessed though this type
of metrology. I find myself at the opposite end of the spectrum to Mr
Challand, and have perhaps 35-40 years of NHS service left. It is
fascinating to think where clinical governance will take the NHS in this
time, and I agree with Mr Kallner that there should be a drive for global
standardisation.
To finish, I would like to assure Mr Kallner that I have pursued (and am
still pursuing) the publications he has kindly referenced in previous
letters. I enjoy reading on such matters (which usually end in my writing a
computer program of some sort to test the statistics), and feel privileged
to have been able to discuss this subject with him and others. I am
currently producing a webpage on which I will host the various statistical
tools I have created for free download (including source code for critique);
I will post here with a link as time permits. Whilst I intend for this to be
the last of my responses on this topic here on the mail base, I would of
course be happy to reply to direct emails on this or any other subject!
My best regards,
Andy
------ACB discussion List Information--------
This is an open discussion list for the academic and clinical community working in clinical biochemistry.
Please note, archived messages are public and can be viewed via the internet. Views expressed are those of the individual and they are responsible for all message content.
ACB Web Site
http://www.acb.org.uk
Green Laboratories Work
http://www.laboratorymedicine.nhs.uk
List Archives
http://www.jiscmail.ac.uk/lists/ACB-CLIN-CHEM-GEN.html
List Instructions (How to leave etc.)
http://www.jiscmail.ac.uk/
|