NEJM study finds “low diagnostic yield” for coronary angiography

Only a bit more than one-third of patients without known coronary artery disease (CAD)  are found to have obstructive CAD upon elective catheterization, according to data from the ACC National Cardiovascular Data Registry (NCDR).

Manesh Patel and colleagues report in the New England Journal of Medicine on nearly 400,000 patients who underwent elective catheterization. Here are the main results:

  • 37.6% (149,739) were found to have obstructive CAD (53% of these had multivessel disease).
  • 39.2% had no coronary artery disease (defined as <20% stenosis in all vessels).
  • Prior to catheterization, 83.9% of patients had some form of noninvasive testing (though this included a resting ECG, as the NCDR database was not designed to assess this point).
  • Diagnostic yield was higher in patients with higher Framingham risk scores.
  • In the entire NCDR population, which includes patients with known disease and all symptoms, the rate of obstructive disease was 60.3%

A Higher Threshold?

Discussing the implications of the low diagnostic yield, the investigators wrote:

Our results suggest that greater focus should be placed on the 30.0% of patients who were noted to have no symptoms, including no angina. Presumably the decision to proceed with invasive catheterization in the case of these patients was driven by clinical assessment of risk, testing for ischemia, or both. Given that the primary benefit of invasive treatment for obstructive coronary artery disease is relief of symptoms, we think that the threshold for invasive angiography may need to be higher in asymptomatic patients, for whom the potential benefits remain uncertain.

The best possible “bang for the rad”?

In an accompanying comment, David Brenner points out that 30% of radiation exposure associated with medical diagnostic imaging now comes from cardiac imaging. He asks “whether current coronary imaging techniques are being used optimally. Specifically, are they providing us with the maximum information relative to the population exposure that they involve — the best possible ‘bang for the rad’? Brenner observes that each year in the US approximately 9 million myocardial perfusion studies are performed, “and this test represents one of the single largest man-made contributors to radiation exposure in the U.S. population.”

The application of gatekeeper tests like these needs to be optimized, writes Brenner, “in order to decrease the disturbingly large proportion of invasive coronary angiographic procedures that yield negative results. Ironically, there is evidence that, in many situations, a better gatekeeper test may be yet another radiographic imaging technique — namely, multidetector-row computed-tomographic (CT) angiography.”

“The more we believe in the test, the less well it will appear to perform…”

Taking his usual broad perspective, George Diamond sent the following comment to CardioBrief:

The practical implication of these observations is very difficult to discern. On the one hand, the prevalence of disease in my 1979 paper was 72% vs the 38% here. This could mean that physicians have become much more liberal in their referral of patients for coronary angiography. This is not necessarily a bad thing. From an information management perspective, any diagnostic test performs optimally when uncertainty is at a maximum. This occurs at a 50% probability of disease. The problem, however, is that many patients so detected might have clinically inconsequential disease and wind up being overtreated (vis a vis COURAGE).

On the other hand, these same observations could imply that physicians have come to rely more on non-invasive testing in deciding who to refer. This is not necessarily a good thing. Here’s a simple example: suppose we have a test with a sensitivity of 90% and a specificity of 90% (this is better than most all our tests, but is consistent with the beliefs of the advocates of the test; the example works also with more realistic numbers such as 80% sensitivity and 80% specificity). Suppose further that physicians have come to rely so much on the accuracy of this test that they refer all positive responders to angiography and refer all negative responders away from angiography. If the test is applied to a population of 100 individuals with a 10% prevalence of disease, there will be 9 true positive responders (10 x 0.9) and 9 false positive responders (90 x 0.1). If all 18 of these positive responders are referred for angiography (because we believe the test result is 100% diagnostic), the observed prevalence of disease will only be 50% (9/18). Hence, the more we believe in the test, the less well it will appear to perform. Sic transit Gloria mundi.

Leave a Reply

%d bloggers like this: