Independent Confirmation

Abstract

Mair and colleagues most recently presented the results from the RITeS study,1 which retrospectively investigated the performance of the e-ASPECTS software (e-Stroke, Brainomix, Oxford, UK) using clinical and computed tomography data from nine different prospective trials. e-ASPECTS' performance was tested in two different populations, a “target” and a “representative” population, with the latter comprising not only patients having acute ischemic stroke or not, but also mimics, including hemorrhage, tumors, or vascular malformations. In this population, e-ASPECTS was also tested regarding its capability to detect hyperdense vessel signs, as well as intracranial hemorrhage. Being the largest e-ASPECTS study to date, RITeS is also stated to be independent in contrast to several previous studies reporting potential conflicts of interest of their authors.2, 3 Although the primary endpoint was met; that is, non-inferiority of e-ASPECTS against experts, the authors focused on the results from the “representative” population, where e-ASPECTS performed worse than experts. In conclusion, the authors mandate for use of the software “as intended (by the vendor)” only; that is, as an assistive device to experienced physicians. Besides “proving the obvious”, the concept of the “representative” population appears questionable, as it seems to imply an intent by whoever to replace the physician by the computer. Another point on which the emphasis is set by the authors is the failure rate of the software, which is reported to be 10% (429/4,100). The largest e-ASPECTS study so far on 2,426 computed tomography scans from the ENCHANTED trial found 5% (84/1564) of scans that could not be processed by e-ASPECTS (using an Intel Core i5-6,300 U processor) after excluding cases according to defined criteria.4 Unfortunately, this is cited incorrectly in the present RITeS study, with a failure rate of 31%. Furthermore, the failure rate in the present study might be reported too high: of 449 failures, 297 were either due to “upload failure” or “cancellation of processing”, both of which are attributable to hardware equipment about which no information is given. Therefore, the actual failure rate might as well have been less than 3%, which would better match previous reports and our clinical experience. Nevertheless, the RITeS study must be acknowledged to be the largest study on e-ASPECTS' performance to date confirming the results of previous, smaller, non-independent studies; that is, that e-ASPECTS performs non-inferior to human experts in detecting early ischemic changes on computed tomography scans. Thereby, it can be a valuable support even to the experienced expert. C. Herweh and S. Nagel received consultancy fees and travel support by Brainomix.

Ähnliche Arbeiten