The role of baseline lesion selection on reader performance in oncologic clinical trials.

Abstract

e14082 Background: Blinded independent central radiology review is integral to modern clinical trials in oncology. Based on previously published data based on overall adjudication rates, it is generally accepted that selection of a larger and more complete number of target and non-target lesions at baseline results in greater reader concordance and therefore improved data fidelity. Methods: A retrospective database review was performed across 27 trials spanning 15 clinical indications and 4 response criteria involving a total of 20,942 independent reads across numerous timepoints in 10,471 subjects by 119 readers. Adjudication rates and adjudication selection rates were calculated across and within all trials. Results: Readers selected a mean of 2.3 target lesions and 1.2 non-target lesions in each subject. The mean adjudication rate across all trials was 33%. However, adjudication selection rate should be a more accurate representation of reader performance. Within any one trial, 44% of readers who selected the greatest number of target and non-target lesions at baseline actually had adjudication selection rates below average. Conclusions: Readers who select more target and non-target lesions than their peers at baseline do not demonstrate significantly higher adjudication selection rates. This suggests that the number of lesions chosen at baseline may not be as strong a contributer to reader performance and data quality as previously surmised.