Simon - Congratulations on a clear exposition of some difficult
concepts. We should stop teaching our students 'classic'
presentations and stop looking for 'magic bullet' tests. Probability
theory should be an important part of everyday undergraduate
teaching - even though realisation of the amount of chance invoved
in medicine may worry them!
Tim. Coats
> Well, they may not be completely useless. Like any other diagnostic test,
> the clinical utility of the test is based upon the performance of the test
> AND the characteristics of the patient.
>
> What was needed in this case (and in many other emergenct assessments) is an
> estimate of the pretest probability of disease for PE in the patient
> described. All the information needed was not given, but by virtue of the
> fact that the patient had recent surgery they are AT LEAST moderate risk!
> This group of patients cannot have PE excluded on d-dimers alone.
>
> If you want the maths:[an alternative is a nomogram as shown in many of the
> ebm books]
>
> 1) What's the pretest probability?
> If we clinically guestimate (ok this bits a bit woolly for PE, you can
> generate good pretest probabilities for things like DVT and chest pain, for
> PE we can still make a clinical estimate) that this patients pretest
> probability of disease is 30%???
>
> 2) What's the likelyhood ratio?
> If d-dimers in PE perform at 85% sensitivity and 64% specificity [1] (there
> are several papers - I have chosen one for demo- this is for whole blood
> assay NOT the ELISA tests which do not perform as well(allegedly)). The
> likelihood ratio for a negative result is therefore 0.23. i.e you are about
> 0.23 as likely to have a PE with a negative test as you were before.
>
> 3) Combine the likelihood ratio with the pretest probability
> 0.3 x 0.23 = 0.069 = 7% So you've reduced the risk, but not to a level at
> which I would feel happy to send the patient home (because it's such a
> dangerous condition). Further investigation is clearly needed (though none
> of those are perfect either!!!).
>
> [You can also do the calcs for a postive result (the LR=2.7), this gives a
> post test result of 81% - so a positive result does not even prove the
> diagnosis.]
>
> HOWEVER, if your pretest probability was as low as 5%, a negative test would
> give you a post test probability of 1%. That's a much better risk than most
> of us manage at the moment!! They could go home (remember no medicine is
> perfect - we just reduce the risk, 1% would be pretty good) with
> instruction. I don't know what others think, but from what I have seen in UK
> we just don't investigate the low risk ones at all!
>
> You can make this sort of thing even more useful if you calculate different
> LR's depending on how abnormal the test result is, making the tests even
> more powerful.
>
> What is needed for PE is a scoring system (like exists for things like dvt
> which would allow us to get a good estimate of pretest disease). Diagnostic
> tests virtually NEVER fully exclude or prove a diagnosis. They just make it
> more or less likely. What you then do depends on 1) what you have changed
> the risk to and 2) what the consequences of getting it wrong are.
>
> Simon
> NB: I am not proposing that we issue a huge list of LR's and pretest
> probabilities to our SHO's (maybe as a PDA???). But this is exactly the kind
> of questions we should be asking when we set up
> protocols/guidelines/diagnostic strategies etc.
Timothy J Coats MD FRCS FFAEM
Senior Lecturer in Accident and Emergency / Pre-Hospital Care
Royal London Hospital, UK.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|