Letter to the Editor: “Diagnostic accuracy of coronary computed tomography angiography-derived fractional flow reserve (CT-FFR) in patients before liver transplantation using CT-FFR machine learning algorithm”
by Tao Cui (email@example.com)Diagnostic accuracy of coronary computed tomography angiography-derived fractional flow reserve (CT-FFR) in patients before liver transplantation using CT-FFR machine learning algorithm
We read the article by Schuessler et al with interest . In their study, they used a CT-FFR machine learning algorithm for differentiation of hemodynamically significant and non-significant coronary stenosis in patients evaluated for LT. In this study, CT-FFR measurements yielded a sensitivity of 71%, a specificity of 90%, a PPV of 67%, and a NPV of 91%, and they concluded that the algorithm seems a very promising noninvasive approach .
Two limitations would be figured out. First, only sensitivity (SE) and specificity (SP) were elucidated as measures of diagnostic accuracy, but those indexes were not sufficient for diagnostic accuracy evaluation. Some other systematical indexes like diagnostic odds ratio, negative-to-positive likelihood ratio, and area under receiver operating characteristic curves (AUC) might reflect the methods’ accuracy from different terms, and a combined diagnostic accuracy is always defined as (true positives + true negatives)/(true positives + false negatives + false positives + true negatives), considering both SP and SE. Moreover, receiver operating characteristic curve (ROC curve) and area under curves (AUC) were necessities when comparing different methods’ accuracy, as the ROC curves demonstrate different SE and SP under various cut-points. In addition, as some accuracy measures can be admissible, while the diagnostic added values ultimately might be clinically negligible . Therefore, only by yielding AUCs of tests, the diagnostic added value of the new technology regardless of cut-points would be understood.
In conclusion, revisions of aforementioned flaws would solidify the conclusions of Schuessler and his colleagues.