Correct math shows no improvement on clinical judgment

I have a problem with the math in Table 1 of the article “Chest pain investigation in patients at low or intermediate risk. What is the best first-line test to rule out coronary artery disease?” which appeared in the January issue of Canadian Family Physician.¹ First, pretest probability is not actually specified, but let us do the calculations using 10% as low risk and 50% as intermediate risk. I will also take the midpoint of the reported ranges of sensitivity and specificity for the purposes of illustration.

The following 2 × 2 tables are generated, the first using a pretest likelihood of 10% and the second using 50%; both use an N of 1000 (Tables 1 and 2).

In neither case are the predictive values reported in Table 1 in the original article (positive predictive value [PPV] of 44% to 64% and negative predictive value [NPV] of 95% to 100%) accurate and in neither case is this test alone a good enough clinical tool.

In the first case of low pretest likelihood, we have little confidence in either the PPV or NPV, and the test improves very little on our clinical judgment. The PPV is only 16%. We go from 10% certain the patient has coronary artery disease (CAD) to 16% certain. The NPV is 97%, which is only an absolute 7% better than the pretest likelihood based on clinical judgment! We go from 90% certain to 97% certain the patient does not have CAD. This illustrates the fallacy of testing when pretest probabilities are low.

In the second case, where the clinical judgment is equivalent to a coin toss, the PPV is 63% and the NPV is 77%. We go from 50% sure the patient has CAD to 63% sure, and 50% sure the patient does not have CAD to 77% sure.

I would assert that a second test is needed in both the positive and negative groups. The positives need to be tested with a test of high specificity, and the negatives need to be tested with a test of greater sensitivity, which is why the patient in the clinical vignette proceeded ultimately to angiography.

I suspect but I cannot prove that our clinical judgments are more refined than we believe. As generalists, we look at the whole picture, the nature of the concern along with family history, lifestyle risk factors—diet, smoking, exercise—blood pressure, lipid levels, and medications. In general practice, the low-probability cases are weeded out on clinical grounds alone, leaving the intermediate- and high-probability cases for referral. Emergency physicians have a different dilemma and I will leave it to them to comment further. Either way, the search for absolute certainty is a fool’s errand and we need to know how to manage uncertainty in conversation with our patients.

• Copyright© the College of Family Physicians of Canada