The adjudication rates between readers in Blinded Independent Central Review (BICR) of advanced esophageal cancer trials with or without immune checkpoint initiators as first-line therapy.

Abstract

4039 Background: It is challenging to use RECIST 1.1 for the response assessment in esophageal cancer, because it excludes certain examines such as the use of barium meal and endoscopy. In some esophageal trials, the study protocols also recommend not selecting the primary tumor as a target lesion, because esophageal lesion could be infiltrative in a cavity organ and difficult to measure reliable especially after treatment. Moreover, it can even be more difficult for independent central readers, as they are often blinded to patient clinical symptoms and outcomes. The aim of this study is to analyze a pool of advanced esophageal cancer trials which used RECIST 1.1, and to document the proportion of reader discrepancies, reader performance through monitoring procedures, and to provide suggestions for the reduction of read inconsistency. This study provides benchmarks of reader adjudication rate in novel esophageal cancer therapeutic trials with or without immune checkpoint inhibitors, which will help to trigger corrective actions, such as reader initial training and follow up re-training. Methods: We analyzed 4 esophageal cancer BICR trials that included 1,875 patients (8,501 time-points) involving 14 radiologists. We analyzed the adjudication rate of each trial as well as testing inter-trial differences. The analysis of adjudication allowed to compute trial and reader adjudication rate, readers endorsement rate, root causes of adjudications. Results: Trials had an average adjudication rate of 45.28% [42.60%-47.99%], while readers endorsement rates ranged [23.1%-81.6%]. The differences of target lesion selection (34.4%) and lesion measurement (40.3%) are the two main reasons that led to discordance per adjudicators’ assessment. The difference of new lesion determination and identification of non-target lesion progression contributed 17.8% and 7.5% of the discordance reasons, respectively. Conclusions: We provided benchmark performances for monitoring reader performance in trials with double reads. The discordances of baseline lesion selection and lesion measurement in follow up visits are the main reasons triggering adjudications in esophageal cancer central reading. Appropriate reader training and monitoring are solutions which can not only mitigate a large portion of the commonly encountered reading errors and help to reach more consensus on lesion selection and measurement between readers.