Correspondence
In Reply
; ;
The results from our colleagues confirm our experience and at the same time provide further evidence that the predictive values (PPV, NPV) of imaging methods cannot be scaled to scenarios of other prevalences. Radiologists always adapt their perception or decision-making behavior (weighing sensitivity versus specificity) to the clinical situation (in this case, lower prevalence). This shift in perception behavior will at least partially compensate for the effect of different prevalences on the prediction values. Based on the information about the likelihood ratio of LD-CT in Jena, and given the lower prevalence of COVID-19 in Jena versus Aachen/Düren, it can be calculated that radiologists increased their perception behavior towards a higher specificity (94.7% versus 91.4%) and lower sensitivity (84.4% versus 94.7%). This is the only way to explain why the PPV observed in Jena (58.2%) was lower than that in Aachen/Düren (87.7%) yet higher than the value resulting from a purely arithmetical scaling to the Jena prevalence (48.9%).
Radiologists shift their perception behavior not only depending on disease prevalence but also when a false positive or false negative result would have very different consequences for a patient. For example, the same CT findings that were acquired with the same question (“COVID-19?”) are interpreted differently for a patient prior to a planned stem cell transplant as for one who is otherwise healthy but has a cough.
Thus, it is undisputed that predictive values can be estimated using a nomogram or calculated using digital methods. The predictive values that a radiological examination method actually achieves in clinical practice can and will differ considerably from the calculated values—unlike laboratory diagnostics. Using standardized findings or diagnostic classification schemes such as COV-RADS do not change this fact.
DOI: 10.3238/arztebl.m2021.0037
On behalf of the authors
Prof. Dr. med. Christiane Kuhl
Dr. med. Maximilian Schulze-Hagen
Dipl.-Inform. Heribert Bieling
Department of Diagnostic and Interventional Radiology
RWTH Aachen University Hospital, Germany
ckuhl@ukaachen.de
Conflict of interest statement
The authors of the contributions declare that no conflict of interest exists.
| 1. | Schulze-Hagen M, Hübel C, Meier-Schroers M, et al.: Low-dose chest CT for the diagnosis of COVID-19—a systematic, prospective comparison with PCR. Dtsch Arztebl Int 2020; 117: 389–95 VOLLTEXT |
