Olin and colleagues [1] present convincing evidence that duplex ultrasonography is a sensitive and specific diagnostic test for substantial (> 60%) renal artery stenosis. The authors erred, however, in their definitions of positive and negative predictive values and, in doing so, may have provided an overly optimistic view of test performance in most patient populations.

The authors correctly defined the sensitivity of ultrasound (U) as the proportion of positive arteriograms (A) that were positive by ultrasonography [P (U⁺/A⁺)] and defined specificity as the proportion of negative arteriograms that were negative by ultrasonography [P (U^–/A^–)], with arteriography considered the gold standard. The authors defined the positive predictive value as the proportion of positive ultrasounds that were positive by arteriography [P (A⁺/U⁺)] and the negative predictive value as the proportion of negative ultrasounds that were negative by arteriography [P (A^–/U^–)]. In their calculation of positive and negative predictive values, however, the authors did not account for disease prevalence. The following equations are based on Bayes formula [2]: Equation 1 The prevalence of renovascular disease (in this case, P [A⁺]) varies widely across patient populations. If the sensitivity and specificity reported by the authors is used, the positive predictive value of duplex ultrasonography would be 33%, 72%, and 84% at population prevalences of 1%, 5%, and 10%, respectively. The negative predictive values would be greater than 99% at the corresponding prevalence estimates.

(1)

Duplex ultrasonography may prove to be a useful addition in the diagnosis of renovascular disease associated with hypertension and ischemic nephropathy. The calculations above highlight the influence of disease prevalence on predictive value and the importance of considering prevalence (that is, previous probability) in interpreting imperfect diagnostic tests.