For image-based evaluations, RECIST 1.1 remains the most used criteria for assessing the therapeutic response in clinical trials. The variabilities of evaluations are generally mitigated by double reading the images, with a third reader adjudicating the discrepancies. Blinded independent central review (BICR) with double read and adjudication is a complex management that needs to be closely monitored. The rate of inter-reader discrepancies is one of those metrics of choice for detecting quality issues in trials. The aim of our study was to provide reference values metrics for the monitoring of RECIST 1.1 BICR with double read plus adjudication in clinical trials.
From the list of clinical trials recorded in the Median Technologies database, we selected a subset of trials according to the following inclusion criteria: 1) Phase II and III 2) Response criteria: RECIST, 3) Trial status: completed, 4) Trial setting: central double read with adjudication, 5) Trial endpoint: Overall Response and PD and 6) Readers monitoring was enabled. For the selected trials, we analyzed, per trial and per readers, the rate of inter-reader discrepancy and the rate of readers’ endorsement by the adjudicator. We compared the discrepancy rate between the indications using Marascuillo test.
Out of the 103 recorded trials, 5 conformed the inclusion criteria. Their indications were: Lung (1), Skin (1) biliary track (1), Gastric (1) and multiple (1) cancers. A total of 1561 patients (mean=312/per trial) and 5986 time points (mean=1197/per trial) were analyzed by 25 readers; 8 adjudicators were involved. Per reader, the discrepancy rate ranged from 27.4% to 68.5% (mean=50.1%) with an endorsement rate ranging from 11.5% to 91.1%. Per trial, the average discrepancy rate was 50.8% [33.0-63.8]. We found a significant difference in the rate of discrepancy per indications: Biliary (63.8%) vs Multiple cancers (33.0%) (p<0.001).
In BICR clinical trials with double reads and adjudication, readers’ monitoring is highly recommended. Monitoring metrics reported a wide range of discrepancy rate and of individual readers performances. Discrepancy rate and readers performances would be indication dependent.
Beaumont Hubert.
Has not received any funding.
H. Beaumont, A. Iannessi, J. Cillario, Y. Liu: Full/Part-time employment: Median Technologies.