Response

We are grateful to Dr. Heston for his valuable comments and suggestions on our study (1). He poses the question of variation of predictive values with disease prevalence and proposes an alternative way to report standardized predictive values and reduce prevalence bias. Although the method proposed is indeed intriguing, we think there remain a few important problems to be discussed. First, it is not uncommon that even large registries do not have any information regarding the undoubtedly large population of patients who were evaluated but did not undergo the diagnostic test (2). As stated in our article, we aimed to put the emphasis on not overlooking a serious condition such as coronary artery disease (CAD) by means of noninvasive testing (cardiac magnetic resonance imaging [MRI]) in a patient population with stable angina pectoris and predominantly intermediate to high pretest risk according to the method proposed by Morise and Jalisi (3). Although Dr. Heston addresses the point that “The surrogate risk calculator that looked at overall mortality cannot substitute for hard numbers of patients with obstructive disease,” the approach of score-based pretest assessment is widely used and correlates significantly with the incidence of disease (2). Second, the major point raised by the author of this letter is that “We need to know the overall prevalence of disease in the entire population undergoing the test before making a judgment as to its diagnostic utility.” We agree, but is pretest knowledge of disease prevalence achievable in a population presenting with suspected CAD? Currently not, as long as you deal with “real world” patients. Reaffirmation of this sobering cognizance comes from the recent large-scale study by Patel et al (2), who determined patterns of noninvasive testing and the diagnostic yield of catheterization among patients with suspected CAD in a contemporary U.S. national sample. As a main result of their CathPCI Registry, including a total of 398,978 patients subjected to elective coronary angiography in the routine clinical setting, only slightly more than one-third of patients in fact had obstructive CAD, even though 68.6% had a positive stress test. This underscores the unsolved dilemma faced by clinical cardiology: Contemporary “real world” assessment of disease prevalence for clinically and stress test-based suspected CAD is burdened with a high risk of error, which inevitably translates into the assessment of predictive values for all tests in the field. Third, while we welcome the detailed presentation of the calculation method for predictive values including positive test results, there may be a limitation in the example given due to a rather arbitrary assumption for prevalence of disease in those with a positive test (assumed as “more than 5 times greater at 20%”). In fact, CAD has been reported to be present in 85% (574/679) of patients with abnormal stress adenosine cardiac MRI in a large meta-analysis (4), a proportion which is substantially higher than the assumption used for the example calculation. In applying this number to the presented approach (true positives now: 158 × 0.85 = 134) with likewise keeping fixed the more realistic values obtained for sensitivity (134/140, 96%) and specificity (152/176, 86%), the virtual test result for the proposed putative prevalence of 75% would yield a negative predictive value of 87%, far better than the presented odds of “only about 1-to-1” to have no obstructive disease given a negative test. Yet we acknowledge that Dr. Heston deserves the credit of having highlighted the important implication of altering the pretest probability for posttest probability results in our data (Baye's Theorem). Finally, while we believe that systematic evaluation of true incidences should be reserved for prospective registries such as currently EuroCMR (5), we have complied with Dr. Heston's request and reviewed our whole population examined in the study period. With all due caution for such an a posteriori database, we consider that in retrospect 462 (62.7%) patients had a negative result from the total of 737 patients who underwent cardiac MRI as a gatekeeper exam prior to cardiac catheterization. As reported, the 158 patients who consented to invasive coronary angiography were studied within our study protocol. We would like to comment on two aspects: First, we admit that there was no invasive correlation for the patients with normal MRI outside the study. However, the ethics of conducting a trial restricting cardiac MRI exams to patients who a priori also give consent to invasive angiography even in the case of normal MRI results would be questionable. As surrogate information, outcome studies on normal cardiac MRI support the hypothesis that clinically important CAD was not missed by adenosine stress cardiac MRI (6, 7). Second, the high rate of negative results might be interpreted as MRI having been performed in a nonrepresentative population with low prevalence of disease, in which the negative predictive value could be high “even for poorly accurate tests.” Yet the magnitude of negative test results in our MRI population was almost identical to the large invasive CathPCI Registry (2), suggesting that we dealt with a clinically representative population. As long as no definitive prestratification test for the individual patient with suspected CAD is available, it may be preferable to potentially “overuse” noninvasive MRI instead of invasive cardiac catheterization, as it is contemporarily practiced in this population.