Author + information
- Peter Martus, PhD,
- Sabine Schueler, MD and
- Marc Dewey, MD, PhD⁎ ()
- ↵⁎Department of Radiology, Charité, Charitéplatz 1, 10117 Berlin, Germany
We read with great interest the paper about the DISCOVER-FLOW (Diagnosis of Ischemia-Causing Stenoses Obtained Via Noninvasive Fractional Flow Reserve) study, which compared fractional flow reserve (FFR) derived from coronary computed tomography angiography (CTA) with invasive FFR measurements (1). CTA is a reliable test to rule out coronary artery disease based on its high sensitivity and negative predictive value compared with conventional angiography as the reference (2). Considering the importance of invasive FFR as part of conventional coronary angiography for subsequent revascularization decisions (3), it would be game-changing if a reliable estimation of FFR could be performed noninvasively (4).
The DISCOVER-FLOW study is an important step in this direction. We would like to discuss 2 issues related to the study design and statistics.
1. Patients with an at least 50% diameter stenosis determined by the clinical site on CTA were studied using invasive FFR, which served as the reference in the study, but was done as clinically indicated. Thus, the final cohort of vessels that had invasive FFR measurements represents a subgroup that may be biased by the local CTA reading and a higher prevalence. This results in an overestimation of sensitivity and an underestimation of specificity. This can be seen from a different version of the Bayes formula needed to calculate sensitivity from predictive values: sensitivity = PPV · T+/(PPV · T+ + (1 − NPV) · T−), where T+ and T− are the proportion of test positives and negatives in the study sample and PPV and NPV are the positive and negative predictive values.
The proportion of test negatives, T− is underestimated, if T− is calculated in the sample of verified subjects or vessels only. A very conservative estimate is to assume 3 vessels per person (i.e., 309 vessels overall). According to Koo et al. (1), 114 vessels were positive (53 true positive, 61 false positive) on CTA. Thus, we assume 195 negative vessels by CTA from which only 45 were assessed by the reference standard FFR. If we further assume that the observed negative predictive value is the true one, we can obtain corrected diagnostic performance estimates (Table 1). It is likely that the same holds true for computed tomography FFR. However, it is difficult to assess this without the correlation structure of both index tests within truly diseased and truly unaffected vessels. If both tests are conditionally independent (i.e., independent within the true positives and the true negatives), the naive estimates of sensitivity and specificity are unbiased.
2. The limits of agreement between FFR and computed tomography FFR resulting from a regression of absolute differences according to Altman (5) suggest that these are not constant but increase linearly with smaller FFR results (r2 linear model = 0.335, r2 quadratic model = 0.340, r2 cubic model = 0.340) (Fig. 1). In other words, the more positive (lower) the FFR results become, the larger are the limits of agreement (95% confidence intervals).
These additions may be important when appraising the impressive DISCOVER-FLOW study results.
- American College of Cardiology Foundation
- Koo B.K.,
- Erglis A.,
- Doh J.H.,
- et al.
- Pijls N.H.,
- Fearon W.F.,
- Tonino P.A.,
- et al.
- Achenbach S.