Does the pretest probablity of a condition really determine which test you should do?

The December 25, 2008 issue of NEJM has a excellent article on "Acute Pulmonary Embolism."

It seems that standard practice has become to use (or at least to say that one should use,I am not sure how often that really happens)) one of the clinical prediction rules or "CPRs" ( in this case the Wells or the Geneva criteria) to make an assessment of how likely is the diagnosis of PE .

Of historical note, the early studies on how best to diagnose PE included a global assessment by the physician and the clinical prediction rules came on the scene later in part, it was said, to aid the less experienced physician.

In one version of this game plan, one determines if there is high,intermediate or low pretest probability of PE. Then the test you order depends on into which risk category the patient is placed. In the first two instances, one then proceeds to obtain a highly sensitive D-dimer test ( e.g.,enzyme-linked immunosorbent assay).If this test is less than some cut-off value (usually 0.5m/liter) the diagnosis is said to be effectively ruled out.However, if there is a high pretest probability of PE then one should not order the D-dimer but proceed directly to a multidetector chest CT .(This is basically the recommendations made by the PIOPED ii investigators in their 2007 summary article.)

This is said to be the case because-according to the NEJM review- clinical data have indicated that in patients with a high pretest likelihood of PE the negative predictive value of the D-dimer is too low to be safely determinative. Interestingly, when one reviews the quoted reference in the Archives of Internal Medicine for that advice it turns out the study involved only 13 instances of high probability PE cases all of whom had a elevated d-dimer. To my perhaps antiquated way of thinking this is a bit strange that we can determine the predictive value of a negative test from data that seems to contain no negative tests.There may well be other studies which I have not yet found that, in fact, provide enough data to conclude the negative predictive value is too low.

Ignoring that anomaly for the time being, what about the advice for those patients in the intermediate risk category. There are some data indicating that about 5% of such patients have false negative D-dimer tests.See the Wikipedia article for that reference.

I think a better reason not to rely on the D-dimer a high probability case is , is concern that a potentially fatal condition may be missed because of one erroneous negative biological lab test considering all the many ways lab tests can get screwed up.Plus you have to feel more confident that clots are in the pulmonary arteries when you can point to the their images plugging up segmental or larger vessels (seeing sub-segmental occlusions may be a bit more iffy).

In any event, I have some lingering trouble in getting my brain around the notion for that for a given condition there is not one most accurate and reliable test but that the test of choice varies with the pretest likelihood. My unease is not made better by the fact that one of the more heavily weighted Wells criteria (a value of 3) is credited to the patient's risk score when PE is thought to be "more likely" that an alternative diagnosis,such determination being made by undefined criteria and can not be standardized. ( O.K., the Geneva criteria seem to fixed that problem.)

Nevertheless, the above discussed game plan seems to work most of the time. The importance of not allowing the D-dimer's negativity to exclude PE is emphasized in this 2008 article from Holland in the journal CHEST.The authors reported data that indicated that almost 10% (9.3%) of patients with a likely diagnosis (they used a version of the Wells criteria that had only two categories, "likely" or "unlikely") and a negative D-dimer had evidence of thrombotic disease in the subsequent 3 months.

I have written about the D-dimer and Pulmonary embolism prediction rules before in a discussion about an earlier larger study from Holland which is generally considered to validate the rule of thumb that says "low likelihood Wells score plus negative high sensitivity D-dimer test equals no PE." As thumb rules go this one does seem pretty good but as an example of brazen self absorption let me quote a paragraph from an earlier posting.

" I wonder if the "dichotomized"version of the Wells decision rule is too simple.All patients in the "unlikely" category are not equally unlikely to have PE. I have a problem with reducing a physician's assessment of how likely the diagnosis may be to a mechanistic rule and apparently excluding any and all other elements that a physician may call upon to decide the likelihood of a diagnosis, for example a decreased 02 saturation. A clinician's "global assessment"( i.e. considering the overall clinical picture not just the check list from Wells) should trump the decision rule.If you believe PE or DVT is a reasonable diagnosis to pursue,testing should be done even if the Wells rule suggests low risk or unlikely and the d-dimer is negative.(show me a blood test that can't be wrong)."