In a recent study, Van Werkhoven et al described the utility of 64-slice computed tomographic coronary angiography (CTA) in patients with intermediate pretest likelihood for coronary artery disease. They examined the ability of CTA to exclude coronary stenosis of >50% at the level of the individual coronary vascular segment, vessel, and patient, in 61 patients. The investigators concluded that CTA can be used as a “highly effective gatekeeper for invasive coronary angiography.”

Although their study was designed to examine disease-oriented outcomes, not clinical outcomes, we believe that Van Werkhoven et al’s reporting of their results and subsequent interpretation are fundamentally flawed. We agree that this test would be most useful in patients with intermediate probability of disease and that Diamond and Forrester’s is an excellent (although dated) method of using readily available clinical variables to objectively quantify the likelihood of disease in a patient. Clinicians interested in using the results of this study would most likely be interested in the performance of this test on a per patient basis.

Statistically, although per vessel and per segment analyses would be interesting, the investigators’ technique fails to allow the within-subject (per vessel) or within-vessel, within-patient (per segment) correlations in the data. The easiest way to account for this would be to conduct distinct analyses within each vessel or segment. Each patient would contribute 1 sample to each analysis. Alternatively, for a more comprehensive approach, one might use generalized estimating equations or generalized linear mixed models to accommodate the nested structure.

In their Table 2, Van Werkhoven et al reported that the sensitivity of the test was 100% (16 of 16), as was the negative predictive value (40 of 40). They also reported that the 95% confidence intervals for sensitivity and negative predictive value were 100% to 100%. A simple statistical estimation of zero numerator results, previously described as the “rule of 3,” can be used to demonstrate that the results in this patient population are much more modest. Using this technique, one would expect the lower bound of the 95% confidence interval for sensitivity to be (3/16) × 100 = 81.3%. Using this same technique for the lower bound of the 95% confidence interval of negative predictive value (3/40 × 100) would result in a 95% confidence interval of 92.5% to 100%. These results alone would call into question the use of this test as a “gatekeeper” for coronary angiography if it is possible that approximately 1 in 6 patients could have significant stenosis missed.

Additionally, Van Werkhoven et al report the negative likelihood ratio as 0. However, the negative likelihood ratio is 0, with a 95% confidence interval of 0 to 0.5. It is generally recognized that likelihood ratios of >10 or <0.1 may significantly affect clinical decision making, particularly in patients with intermediate pretest probability. If one is to use these results in conjunction with Bayesian decision making, one would conclude that the results are modestly promising at best. The investigators describe using this test in a patient population with intermediate probability of disease ranging from 13.4% to 87.2%. In fact, using this test in someone with a pretest probability of disease similar to the prevalence of disease in the patients in this study (26%) would result in a posterior probability of disease with a 95% confidence interval of 0% to 15%. Most clinicians would hardly be willing to accept these probabilities, especially in higher probability patients.

Although we agree with Van Werkhoven et al that these results are promising, we respectfully suggest that further analysis and interpretation of the data demonstrate the need for larger studies before one can make the conclusion that CTA may be a “gatekeeper” for coronary angiography in this patient population.