Georg M. Schuetz, Presenter: Nothing to Disclose

Peter Schlattmann PhD, Abstract Co-Author: Nothing to Disclose

Marc Dewey MD, Abstract Co-Author: Research Grant, General Electric Company Research Grant, Bracco Group Research Grant, Guerbet SA Research Grant, Toshiba Corporation Speakers Bureau, Toshiba Corporation Speakers Bureau, Bayer AG Speakers Bureau, Guerbet SA Consultant, Guerbet SA Author, Springer Science+Business Media Deutschland GmbH Institutional research agreement, Siemens AG Institutional research agreement, Koninklijke Philips Electronics NV Institutional research agreement, Toshiba Corporation

There is no consensus on how to handle nonevaluable results in diagnostic accuracy studies and common approaches (excluding or declaring them as either positive or negative) overestimate diagnostic accuracy values. Analyzing the field of coronary CT angiography, we propose the use of a 3-by-2 table for transparent reporting and as basis for appropriate inclusion of nonevaluable results (intention-to-diagnose approach).

Medline, Embase, and ISI Web of Science databases were systematically searched. Eligible studies had to be prospectively performed and to be published in English or German. They had to compare coronary CT with conventional coronary angiography in all patients and provide sufficient data on a patient-level. The full-texts of relevant studies were explored for sufficient information to calculate an alternative 3-by-2 table. Summary diagnostic performance values were calculated from standard data of a 2-by-2 table (after excluding nonevaluable results) or the 3-by-2 table, including nonevaluable results as either “false negative” or “false positive” according to the reference standard results.

120 full-texts of eligible studies were analyzed. The studies greatly varied in handling nonevaluable outcomes. Overall, it was possible for 26 studies including 2,298 patients to calculate both a 2-by-2 and a 3-by-2 table. According to a bivariate random-effects calculation the "classical" 2-by-2 table and the 3-by-2 table were compared to each other. A statistically significant (p<0.05) difference was found for pooled sensitivity (98.2 (95% confidence interval 96.7 to 99.1) vs. 92.7 (88.5 to 95.3)), area under the curve (0.99 (0.98 to 1.00) vs. 0.93 (0.91 to 0.95)), positive likelihood ratio (9.1 (6.2 to13.3) vs. 4.4 (3.3 to 6.0)), and negative likelihood ratio (0.02 (0.01 to 0.04) vs. 0.09 (0.06 to 0.15)).

Diagnostic performance parameters significantly decrease when nonevaluable results are included by using a 3-by-2 table for analysis.

An intention-to-diagnose approach using a 3-by-2 table can be expected to give a more realistic picture of the clinical potential of diagnostic tests in general and is thus recommended.

Schuetz, G, Schlattmann, P, Dewey, M, A 3-by-2 Table with an Intention-to-Diagnose Approach Avoids Overestimation of Diagnostic Accuracy: A Meta-analytical Evaluation of Noninvasive Coronary CT Angiography Studies.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12026550.html