Dear list members
I've been watching with interest the debate on assessing diagnostic tests in
the absence of gold standards and it seems to me that there are some
fundamental philosophical issues to be answered about the meaning of a
'diagnosis'. We have a great tendency to see the world as 'black and white' -
those with or without a diagnosis, disease, outcome or positive test result.
However, we all recognised that the world is not dichotomous but has infinite
shades and hues.
It may be more helpful to think of diagnoses not as an underlying 'absolute
truth' but as a shorthand for a predictive model that, on the basis of past
experience, describes expected outcomes and responses to treatment. Thus the
'gold standard' becomes the classification which best defines a subset of
patients with the greatest predictive power for the question being asked.
This might be clearer with an example - venography has frequently been
considered the 'gold standard' for defining DVT. By this definition
radio-isotope scanning 'over-predicts' DVT and duplex scanning 'misses'
below-knee thrombosis. But there is no absolute truth about the diagnosis of
DVT - whether we consider small thromboses in the tibial vessels to be DVT is
simply a matter of definition. From the pragmatic point of view the important
issue is how well the diagnostic criteria define a subset of patients for
whom particular outcomes or responses can be predicted. For example we may
wish to predict PE, post-phlebitic syndrome, benefit from anticoagulation,
lysis or caval filter and it may be that for each of these purposes a
different definition (and thus gold standard) is most accurately predictive.
Part of our difficulty may be that most 'scientific' medicine is based on
dichotomous thinking - the entire basis of RCTs is the analysis of a 2 x 2
table. Having struggled to learn (and teach) these relatively simple
statistics it's a daunting thought that perhaps medical diagnosis may be
better modelled using non-linear regression, neural networks or fuzzy logic!
Jonathan
Jonathan Michaels, MChir. FRCS.
Consultant Vascular Surgeon,
Northern General Hospital,
Herries Road
Sheffield S5 7AU
UK
Tel: +44 114 271 4968
Fax: +44 114 271 4747
email: [log in to unmask]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|